nips nips2005 nips2005-173 knowledge-graph by maker-knowledge-mining

173 nips-2005-Sensory Adaptation within a Bayesian Framework for Perception

Source: pdf

Author: Alan Stocker, Eero P. Simoncelli

Abstract: We extend a previously developed Bayesian framework for perception to account for sensory adaptation. We ﬁrst note that the perceptual effects of adaptation seems inconsistent with an adjustment of the internally represented prior distribution. Instead, we postulate that adaptation increases the signal-to-noise ratio of the measurements by adapting the operational range of the measurement stage to the input range. We show that this changes the likelihood function in such a way that the Bayesian estimator model can account for reported perceptual behavior. In particular, we compare the model’s predictions to human motion discrimination data and demonstrate that the model accounts for the commonly observed perceptual adaptation effects of repulsion and enhanced discriminability. 1 Motivation A growing number of studies support the notion that humans are nearly optimal when performing perceptual estimation tasks that require the combination of sensory observations with a priori knowledge. The Bayesian formulation of these problems deﬁnes the optimal strategy, and provides a principled yet simple computational framework for perception that can account for a large number of known perceptual effects and illusions, as demonstrated in sensorimotor learning [1], cue combination [2], or visual motion perception [3], just to name a few of the many examples. Adaptation is a fundamental phenomenon in sensory perception that seems to occur at all processing levels and modalities. A variety of computational principles have been suggested as explanations for adaptation. Many of these are based on the concept of maximizing the sensory information an observer can obtain about a stimulus despite limited sensory resources [4, 5, 6]. More mechanistically, adaptation can be interpreted as the attempt of the sensory system to adjusts its (limited) dynamic range such that it is maximally informative with respect to the statistics of the stimulus. A typical example is observed in the retina, which manages to encode light intensities that vary over nine orders of magnitude using ganglion cells whose dynamic range covers only two orders of magnitude. This is achieved by adapting to the local mean as well as higher order statistics of the visual input over short time-scales [7]. ∗ corresponding author. If a Bayesian framework is to provide a valid computational explanation of perceptual processes, then it needs to account for the behavior of a perceptual system, regardless of its adaptation state. In general, adaptation in a sensory estimation task seems to have two fundamental effects on subsequent perception: • Repulsion: The estimate of parameters of subsequent stimuli are repelled by those of the adaptor stimulus, i.e. the perceived values for the stimulus variable that is subject to the estimation task are more distant from the adaptor value after adaptation. This repulsive effect has been reported for perception of visual speed (e.g. [8, 9]), direction-of-motion [10], and orientation [11]. • Increased sensitivity: Adaptation increases the observer’s discrimination ability around the adaptor (e.g. for visual speed [12, 13]), however it also seems to decrease it further away from the adaptor as shown in the case of direction-of-motion discrimination [14]. In this paper, we show that these two perceptual effects can be explained within a Bayesian estimation framework of perception. Note that our description is at an abstract functional level - we do not attempt to provide a computational model for the underlying mechanisms responsible for adaptation, and this clearly separates this paper from other work which might seem at ﬁrst glance similar [e.g., 15]. 2 Adaptive Bayesian estimator framework Suppose that an observer wants to estimate a property of a stimulus denoted by the variable θ, based on a measurement m. In general, the measurement can be vector-valued, and is corrupted by both internal and external noise. Hence, combining the noisy information gained by the measurement m with a priori knowledge about θ is advantageous. According to Bayes’ rule 1 p(θ|m) = p(m|θ)p(θ) . (1) α That is, the probability of stimulus value θ given m (posterior) is the product of the likelihood p(m|θ) of the particular measurement and the prior p(θ). The normalization constant α serves to ensure that the posterior is a proper probability distribution. Under the assumpˆ tion of a squared-error loss function, the optimal estimate θ(m) is the mean of the posterior, thus ∞ ˆ θ(m) = θ p(θ|m) dθ . (2) 0 ˆ Note that θ(m) describes an estimate for a single measurement m. As discussed in [16], the measurement will vary stochastically over the course of many exposures to the same stimulus, and thus the estimator will also vary. We return to this issue in Section 3.2. Figure 1a illustrates a Bayesian estimator, in which the shape of the (arbitrary) prior distribution leads on average to a shift of the estimate toward a lower value of θ than the true stimulus value θstim . The likelihood and the prior are the fundamental constituents of the Bayesian estimator model. Our goal is to describe how adaptation alters these constituents so as to account for the perceptual effects of repulsion and increased sensitivity. Adaptation does not change the prior ... An intuitively sensible hypothesis is that adaptation changes the prior distribution. Since the prior is meant to reﬂect the knowledge the observer has about the distribution of occurrences of the variable θ in the world, repeated viewing of stimuli with the same parameter a b probability probability attraction ! posterior likelihood prior modified prior Ã θ θ ˆ θ' θ θadapt Figure 1: Hypothetical model in which adaptation alters the prior distribution. a) Unadapted Bayesian estimation conﬁguration in which the prior leads to a shift of the estimate ˆ θ, relative to the stimulus parameter θstim . Both the likelihood function and the prior distriˆ bution contribute to the exact value of the estimate θ (mean of the posterior). b) Adaptation acts by increasing the prior distribution around the value, θadapt , of the adapting stimulus ˆ parameter. Consequently, an subsequent estimate θ of the same stimulus parameter value θstim is attracted toward the adaptor. This is the opposite of observed perceptual effects, and we thus conclude that adjustments of the prior in a Bayesian model do not account for adaptation. value θadapt should presumably increase the prior probability in the vicinity of θadapt . Figure 1b schematically illustrates the effect of such a change in the prior distribution. The estimated (perceived) value of the parameter under the adapted condition is attracted to the adapting parameter value. In order to account for observed perceptual repulsion effects, the prior would have to decrease at the location of the adapting parameter, a behavior that seems fundamentally inconsistent with the notion of a prior distribution. ... but increases the reliability of the measurements Since a change in the prior distribution is not consistent with repulsion, we are led to the conclusion that adaptation must change the likelihood function. But why, and how should this occur? In order to answer this question, we reconsider the functional purpose of adaptation. We assume that adaptation acts to allocate more resources to the representation of the parameter values in the vicinity of the adaptor [4], resulting in a local increase in the signal-to-noise ratio (SNR). This can be accomplished, for example, by dynamically adjusting the operational range to the statistics of the input. This kind of increased operational gain around the adaptor has been effectively demonstrated in the process of retinal adaptation [17]. In the context of our Bayesian estimator framework, and restricting to the simple case of a scalar-valued measurement, adaptation results in a narrower conditional probability density p(m|θ) in the immediate vicinity of the adaptor, thus an increase in the reliability of the measurement m. This is offset by a broadening of the conditional probability density p(m|θ) in the region beyond the adaptor vicinity (we assume that total resources are conserved, and thus an increase around the adaptor must necessarily lead to a decrease elsewhere). Figure 2 illustrates the effect of this local increase in signal-to-noise ratio on the likeli- unadapted adapted θadapt p(m2| θ )' 1/SNR θ θ θ1 θ2 θ1 p(m2|θ) θ2 θ m2 p(m1| θ )' m1 m m p(m1|θ) θ θ θ θadapt p(m| θ2)' p(m|θ2) likelihoods p(m|θ1) p(m| θ1)' p(m|θadapt )' conditionals Figure 2: Measurement noise, conditionals and likelihoods. The two-dimensional conditional density, p(m|θ), is shown as a grayscale image for both the unadapted and adapted cases. We assume here that adaptation increases the reliability (SNR) of the measurement around the parameter value of the adaptor. This is balanced by a decrease in SNR of the measurement further away from the adaptor. Because the likelihood is a function of θ (horizontal slices, shown plotted at right), this results in an asymmetric change in the likelihood that is in agreement with a repulsive effect on the estimate. a b ^ ∆θ ^ ∆θ [deg] + 0 60 30 0 -30 - θ θ adapt -60 -180 -90 90 θadapt 180 θ [deg] Figure 3: Repulsion: Model predictions vs. human psychophysics. a) Difference in perceived direction in the pre- and post-adaptation condition, as predicted by the model. Postadaptive percepts of motion direction are repelled away from the direction of the adaptor. b) Typical human subject data show a qualitatively similar repulsive effect. Data (and ﬁt) are replotted from [10]. hood function. The two gray-scale images represent the conditional probability densities, p(m|θ), in the unadapted and the adapted state. They are formed by assuming additive noise on the measurement m of constant variance (unadapted) or with a variance that decreases symmetrically in the vicinity of the adaptor parameter value θadapt , and grows slightly in the region beyond. In the unadapted state, the likelihood is convolutional and the shape and variance are equivalent to the distribution of measurement noise. However, in the adapted state, because the likelihood is a function of θ (horizontal slice through the conditional surface) it is no longer convolutional around the adaptor. As a result, the mean is pushed away from the adaptor, as illustrated in the two graphs on the right. Assuming that the prior distribution is fairly smooth, this repulsion effect is transferred to the posterior distribution, and thus to the estimate. 3 Simulation Results We have qualitatively demonstrated that an increase in the measurement reliability around the adaptor is consistent with the repulsive effects commonly seen as a result of perceptual adaptation. In this section, we simulate an adapted Bayesian observer by assuming a simple model for the changes in signal-to-noise ratio due to adaptation. We address both repulsion and changes in discrimination threshold. In particular, we compare our model predictions with previously published data from psychophysical experiments examining human perception of motion direction. 3.1 Repulsion In the unadapted state, we assume the measurement noise to be additive and normally distributed, and constant over the whole measurement space. Thus, assuming that m and θ live in the same space, the likelihood is a Gaussian of constant width. In the adapted state, we assume a simple functional description for the variance of the measurement noise around the adapter. Speciﬁcally, we use a constant plus a difference of two Gaussians, a b relative discrimination threshold relative discrimination threshold 1.8 1 θ θadapt 1.6 1.4 1.2 1 0.8 -40 -20 θ adapt 20 40 θ [deg] Figure 4: Discrimination thresholds: Model predictions vs. human psychophysics. a) The model predicts that thresholds for direction discrimination are reduced at the adaptor. It also predicts two side-lobes of increased threshold at further distance from the adaptor. b) Data of human psychophysics are in qualitative agreement with the model. Data are replotted from [14] (see also [11]). each having equal area, with one twice as broad as the other (see Fig. 2). Finally, for simplicity, we assume a ﬂat prior, but any reasonable smooth prior would lead to results that are qualitatively similar. Then, according to (2) we compute the predicted estimate of motion direction in both the unadapted and the adapted case. Figure 3a shows the predicted difference between the pre- and post-adaptive average estimate of direction, as a function of the stimulus direction, θstim . The adaptor is indicated with an arrow. The repulsive effect is clearly visible. For comparison, Figure 3b shows human subject data replotted from [10]. The perceived motion direction of a grating was estimated, under both adapted and unadapted conditions, using a two-alternative-forced-choice experimental paradigm. The plot shows the change in perceived direction as a function of test stimulus direction relative to that of the adaptor. Comparison of the two panels of Figure 3 indicate that despite the highly simpliﬁed construction of the model, the prediction is quite good, and even includes the small but consistent repulsive effects observed 180 degrees from the adaptor. 3.2 Changes in discrimination threshold Adaptation also changes the ability of human observers to discriminate between the direction of two different moving stimuli. In order to model discrimination thresholds, we need to consider a Bayesian framework that can account not only for the mean of the estimate but also its variability. We have recently developed such a framework, and used it to quantitatively constrain the likelihood and the prior from psychophysical data [16]. This framework accounts for the effect of the measurement noise on the variability of the ˆ ˆ estimate θ. Speciﬁcally, it provides a characterization of the distribution p(θ|θstim ) of the estimate for a given stimulus direction in terms of its expected value and its variance as a function of the measurement noise. As in [16] we write ˆ ∂ θ(m) 2 ˆ var θ|θstim = var m ( ) |m=θstim . (3) ∂m Assuming that discrimination threshold is proportional to the standard deviation, ˆ var θ|θstim , we can now predict how discrimination thresholds should change after adaptation. Figure 4a shows the predicted change in discrimination thresholds relative to the unadapted condition for the same model parameters as in the repulsion example (Figure 3a). Thresholds are slightly reduced at the adaptor, but increase symmetrically for directions further away from the adaptor. For comparison, Figure 4b shows the relative change in discrimination thresholds for a typical human subject [14]. Again, the behavior of the human observer is qualitatively well predicted. 4 Discussion We have shown that adaptation can be incorporated into a Bayesian estimation framework for human sensory perception. Adaptation seems unlikely to manifest itself as a change in the internal representation of prior distributions, as this would lead to perceptual bias effects that are opposite to those observed in human subjects. Instead, we argue that adaptation leads to an increase in reliability of the measurement in the vicinity of the adapting stimulus parameter. We show that this change in the measurement reliability results in changes of the likelihood function, and that an estimator that utilizes this likelihood function will exhibit the commonly-observed adaptation effects of repulsion and changes in discrimination threshold. We further conﬁrm our model by making quantitative predictions and comparing them with known psychophysical data in the case of human perception of motion direction. Many open questions remain. The results demonstrated here indicate that a resource allocation explanation is consistent with the functional effects of adaptation, but it seems unlikely that theory alone can lead to a unique quantitative prediction of the detailed form of these effects. Speciﬁcally, the constraints imposed by biological implementation are likely to play a role in determining the changes in measurement noise as a function of adaptor parameter value, and it will be important to characterize and interpret neural response changes in the context of our framework. Also, although we have argued that changes in the prior seem inconsistent with adaptation effects, it may be that such changes do occur but are offset by the likelihood effect, or occur only on much longer timescales. Last, if one considers sensory perception as the result of a cascade of successive processing stages (with both feedforward and feedback connections), it becomes necessary to expand the Bayesian description to describe this cascade [e.g., 18, 19]. For example, it may be possible to interpret this cascade as a sequence of Bayesian estimators, in which the measurement of each stage consists of the estimate computed at the previous stage. Adaptation could potentially occur in each of these processing stages, and it is of fundamental interest to understand how such a cascade can perform useful stable computations despite the fact that each of its elements is constantly readjusting its response properties. References [1] K. K¨ rding and D. Wolpert. Bayesian integration in sensorimotor learning. o 427(15):244–247, January 2004. Nature, [2] D C Knill and W Richards, editors. Perception as Bayesian Inference. Cambridge University Press, 1996. [3] Y. Weiss, E. Simoncelli, and E. Adelson. Motion illusions as optimal percept. Nature Neuroscience, 5(6):598–604, June 2002. [4] H.B. Barlow. Vision: Coding and Efﬁciency, chapter A theory about the functional role and synaptic mechanism of visual after-effects, pages 363–375. Cambridge University Press., 1990. [5] M.J. Wainwright. Visual adaptation as optimal information transmission. Vision Research, 39:3960–3974, 1999. [6] N. Brenner, W. Bialek, and R. de Ruyter van Steveninck. Adaptive rescaling maximizes information transmission. Neuron, 26:695–702, June 2000. [7] S.M. Smirnakis, M.J. Berry, D.K. Warland, W. Bialek, and M. Meister. Adaptation of retinal processing to image contrast and spatial scale. Nature, 386:69–73, March 1997. [8] P. Thompson. Velocity after-effects: the effects of adaptation to moving stimuli on the perception of subsequently seen moving stimuli. Vision Research, 21:337–345, 1980. [9] A.T. Smith. Velocity coding: evidence from perceived velocity shifts. Vision Research, 25(12):1969–1976, 1985. [10] P. Schrater and E. Simoncelli. Local velocity representation: evidence from motion adaptation. Vision Research, 38:3899–3912, 1998. [11] C.W. Clifford. Perceptual adaptation: motion parallels orientation. Trends in Cognitive Sciences, 6(3):136–143, March 2002. [12] C. Clifford and P. Wenderoth. Adaptation to temporal modulaton can enhance differential speed sensitivity. Vision Research, 39:4324–4332, 1999. [13] A. Kristjansson. Increased sensitivity to speed changes during adaptation to ﬁrst-order, but not to second-order motion. Vision Research, 41:1825–1832, 2001. [14] R.E. Phinney, C. Bowd, and R. Patterson. Direction-selective coding of stereoscopic (cyclopean) motion. Vision Research, 37(7):865–869, 1997. [15] N.M. Grzywacz and R.M. Balboa. A Bayesian framework for sensory adaptation. Neural Computation, 14:543–559, 2002. [16] A.A. Stocker and E.P. Simoncelli. Constraining a Bayesian model of human visual speed perception. In Lawrence K. Saul, Yair Weiss, and L´ on Bottou, editors, Advances in Neural Infore mation Processing Systems NIPS 17, pages 1361–1368, Cambridge, MA, 2005. MIT Press. [17] D. Tranchina, J. Gordon, and R.M. Shapley. Retinal light adaptation – evidence for a feedback mechanism. Nature, 310:314–316, July 1984. ´ [18] S. Deneve. Bayesian inference in spiking neurons. In Lawrence K. Saul, Yair Weiss, and L eon Bottou, editors, Adv. Neural Information Processing Systems (NIPS*04), vol 17, Cambridge, MA, 2005. MIT Press. [19] R. Rao. Hierarchical Bayesian inference in networks of spiking neurons. In Lawrence K. Saul, Yair Weiss, and L´ on Bottou, editors, Adv. Neural Information Processing Systems (NIPS*04), e vol 17, Cambridge, MA, 2005. MIT Press.

Reference: text

Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 Simoncelli Howard Hughes Medical Institute and Center for Neural Science New York University Abstract We extend a previously developed Bayesian framework for perception to account for sensory adaptation. [sent-3, score-0.351]

2 We ﬁrst note that the perceptual effects of adaptation seems inconsistent with an adjustment of the internally represented prior distribution. [sent-4, score-0.901]

3 Instead, we postulate that adaptation increases the signal-to-noise ratio of the measurements by adapting the operational range of the measurement stage to the input range. [sent-5, score-0.827]

4 We show that this changes the likelihood function in such a way that the Bayesian estimator model can account for reported perceptual behavior. [sent-6, score-0.439]

5 In particular, we compare the model’s predictions to human motion discrimination data and demonstrate that the model accounts for the commonly observed perceptual adaptation effects of repulsion and enhanced discriminability. [sent-7, score-1.327]

6 1 Motivation A growing number of studies support the notion that humans are nearly optimal when performing perceptual estimation tasks that require the combination of sensory observations with a priori knowledge. [sent-8, score-0.326]

7 Adaptation is a fundamental phenomenon in sensory perception that seems to occur at all processing levels and modalities. [sent-10, score-0.4]

8 Many of these are based on the concept of maximizing the sensory information an observer can obtain about a stimulus despite limited sensory resources [4, 5, 6]. [sent-12, score-0.566]

9 More mechanistically, adaptation can be interpreted as the attempt of the sensory system to adjusts its (limited) dynamic range such that it is maximally informative with respect to the statistics of the stimulus. [sent-13, score-0.526]

10 This is achieved by adapting to the local mean as well as higher order statistics of the visual input over short time-scales [7]. [sent-15, score-0.134]

11 If a Bayesian framework is to provide a valid computational explanation of perceptual processes, then it needs to account for the behavior of a perceptual system, regardless of its adaptation state. [sent-17, score-0.813]

12 In general, adaptation in a sensory estimation task seems to have two fundamental effects on subsequent perception: • Repulsion: The estimate of parameters of subsequent stimuli are repelled by those of the adaptor stimulus, i. [sent-18, score-1.287]

13 the perceived values for the stimulus variable that is subject to the estimation task are more distant from the adaptor value after adaptation. [sent-20, score-0.623]

14 This repulsive effect has been reported for perception of visual speed (e. [sent-21, score-0.405]

15 • Increased sensitivity: Adaptation increases the observer’s discrimination ability around the adaptor (e. [sent-24, score-0.568]

16 for visual speed [12, 13]), however it also seems to decrease it further away from the adaptor as shown in the case of direction-of-motion discrimination [14]. [sent-26, score-0.736]

17 In this paper, we show that these two perceptual effects can be explained within a Bayesian estimation framework of perception. [sent-27, score-0.352]

18 2 Adaptive Bayesian estimator framework Suppose that an observer wants to estimate a property of a stimulus denoted by the variable θ, based on a measurement m. [sent-31, score-0.643]

19 In general, the measurement can be vector-valued, and is corrupted by both internal and external noise. [sent-32, score-0.251]

20 Hence, combining the noisy information gained by the measurement m with a priori knowledge about θ is advantageous. [sent-33, score-0.251]

21 (1) α That is, the probability of stimulus value θ given m (posterior) is the product of the likelihood p(m|θ) of the particular measurement and the prior p(θ). [sent-35, score-0.58]

22 (2) 0 ˆ Note that θ(m) describes an estimate for a single measurement m. [sent-38, score-0.293]

23 As discussed in [16], the measurement will vary stochastically over the course of many exposures to the same stimulus, and thus the estimator will also vary. [sent-39, score-0.318]

24 Figure 1a illustrates a Bayesian estimator, in which the shape of the (arbitrary) prior distribution leads on average to a shift of the estimate toward a lower value of θ than the true stimulus value θstim . [sent-42, score-0.297]

25 The likelihood and the prior are the fundamental constituents of the Bayesian estimator model. [sent-43, score-0.331]

26 Our goal is to describe how adaptation alters these constituents so as to account for the perceptual effects of repulsion and increased sensitivity. [sent-44, score-1.112]

27 An intuitively sensible hypothesis is that adaptation changes the prior distribution. [sent-48, score-0.595]

28 Since the prior is meant to reﬂect the knowledge the observer has about the distribution of occurrences of the variable θ in the world, repeated viewing of stimuli with the same parameter a b probability probability attraction ! [sent-49, score-0.235]

29 posterior likelihood prior modified prior Ã θ θ ˆ θ' θ θadapt Figure 1: Hypothetical model in which adaptation alters the prior distribution. [sent-50, score-0.885]

30 a) Unadapted Bayesian estimation conﬁguration in which the prior leads to a shift of the estimate ˆ θ, relative to the stimulus parameter θstim . [sent-51, score-0.348]

31 Both the likelihood function and the prior distriˆ bution contribute to the exact value of the estimate θ (mean of the posterior). [sent-52, score-0.225]

32 b) Adaptation acts by increasing the prior distribution around the value, θadapt , of the adapting stimulus ˆ parameter. [sent-53, score-0.415]

33 Consequently, an subsequent estimate θ of the same stimulus parameter value θstim is attracted toward the adaptor. [sent-54, score-0.248]

34 This is the opposite of observed perceptual effects, and we thus conclude that adjustments of the prior in a Bayesian model do not account for adaptation. [sent-55, score-0.342]

35 value θadapt should presumably increase the prior probability in the vicinity of θadapt . [sent-56, score-0.255]

36 Figure 1b schematically illustrates the effect of such a change in the prior distribution. [sent-57, score-0.209]

37 The estimated (perceived) value of the parameter under the adapted condition is attracted to the adapting parameter value. [sent-58, score-0.21]

38 In order to account for observed perceptual repulsion effects, the prior would have to decrease at the location of the adapting parameter, a behavior that seems fundamentally inconsistent with the notion of a prior distribution. [sent-59, score-0.911]

39 but increases the reliability of the measurements Since a change in the prior distribution is not consistent with repulsion, we are led to the conclusion that adaptation must change the likelihood function. [sent-63, score-0.796]

40 We assume that adaptation acts to allocate more resources to the representation of the parameter values in the vicinity of the adaptor [4], resulting in a local increase in the signal-to-noise ratio (SNR). [sent-66, score-0.977]

41 This kind of increased operational gain around the adaptor has been effectively demonstrated in the process of retinal adaptation [17]. [sent-68, score-0.965]

42 In the context of our Bayesian estimator framework, and restricting to the simple case of a scalar-valued measurement, adaptation results in a narrower conditional probability density p(m|θ) in the immediate vicinity of the adaptor, thus an increase in the reliability of the measurement m. [sent-69, score-0.988]

43 This is offset by a broadening of the conditional probability density p(m|θ) in the region beyond the adaptor vicinity (we assume that total resources are conserved, and thus an increase around the adaptor must necessarily lead to a decrease elsewhere). [sent-70, score-1.014]

44 The two-dimensional conditional density, p(m|θ), is shown as a grayscale image for both the unadapted and adapted cases. [sent-72, score-0.38]

45 We assume here that adaptation increases the reliability (SNR) of the measurement around the parameter value of the adaptor. [sent-73, score-0.794]

46 This is balanced by a decrease in SNR of the measurement further away from the adaptor. [sent-74, score-0.326]

47 Because the likelihood is a function of θ (horizontal slices, shown plotted at right), this results in an asymmetric change in the likelihood that is in agreement with a repulsive effect on the estimate. [sent-75, score-0.386]

48 a b ^ ∆θ ^ ∆θ [deg] + 0 60 30 0 -30 - θ θ adapt -60 -180 -90 90 θadapt 180 θ [deg] Figure 3: Repulsion: Model predictions vs. [sent-76, score-0.178]

49 a) Difference in perceived direction in the pre- and post-adaptation condition, as predicted by the model. [sent-78, score-0.218]

50 Postadaptive percepts of motion direction are repelled away from the direction of the adaptor. [sent-79, score-0.349]

51 b) Typical human subject data show a qualitatively similar repulsive effect. [sent-80, score-0.272]

52 The two gray-scale images represent the conditional probability densities, p(m|θ), in the unadapted and the adapted state. [sent-83, score-0.38]

53 They are formed by assuming additive noise on the measurement m of constant variance (unadapted) or with a variance that decreases symmetrically in the vicinity of the adaptor parameter value θadapt , and grows slightly in the region beyond. [sent-84, score-0.746]

54 In the unadapted state, the likelihood is convolutional and the shape and variance are equivalent to the distribution of measurement noise. [sent-85, score-0.623]

55 However, in the adapted state, because the likelihood is a function of θ (horizontal slice through the conditional surface) it is no longer convolutional around the adaptor. [sent-86, score-0.27]

56 Assuming that the prior distribution is fairly smooth, this repulsion effect is transferred to the posterior distribution, and thus to the estimate. [sent-88, score-0.439]

57 3 Simulation Results We have qualitatively demonstrated that an increase in the measurement reliability around the adaptor is consistent with the repulsive effects commonly seen as a result of perceptual adaptation. [sent-89, score-1.288]

58 In this section, we simulate an adapted Bayesian observer by assuming a simple model for the changes in signal-to-noise ratio due to adaptation. [sent-90, score-0.303]

59 We address both repulsion and changes in discrimination threshold. [sent-91, score-0.52]

60 In particular, we compare our model predictions with previously published data from psychophysical experiments examining human perception of motion direction. [sent-92, score-0.405]

61 1 Repulsion In the unadapted state, we assume the measurement noise to be additive and normally distributed, and constant over the whole measurement space. [sent-94, score-0.766]

62 In the adapted state, we assume a simple functional description for the variance of the measurement noise around the adapter. [sent-96, score-0.448]

63 Speciﬁcally, we use a constant plus a difference of two Gaussians, a b relative discrimination threshold relative discrimination threshold 1. [sent-97, score-0.464]

64 8 -40 -20 θ adapt 20 40 θ [deg] Figure 4: Discrimination thresholds: Model predictions vs. [sent-102, score-0.178]

65 a) The model predicts that thresholds for direction discrimination are reduced at the adaptor. [sent-104, score-0.34]

66 b) Data of human psychophysics are in qualitative agreement with the model. [sent-106, score-0.089]

67 Finally, for simplicity, we assume a ﬂat prior, but any reasonable smooth prior would lead to results that are qualitatively similar. [sent-110, score-0.154]

68 Then, according to (2) we compute the predicted estimate of motion direction in both the unadapted and the adapted case. [sent-111, score-0.596]

69 Figure 3a shows the predicted difference between the pre- and post-adaptive average estimate of direction, as a function of the stimulus direction, θstim . [sent-112, score-0.219]

70 For comparison, Figure 3b shows human subject data replotted from [10]. [sent-115, score-0.168]

71 The perceived motion direction of a grating was estimated, under both adapted and unadapted conditions, using a two-alternative-forced-choice experimental paradigm. [sent-116, score-0.628]

72 The plot shows the change in perceived direction as a function of test stimulus direction relative to that of the adaptor. [sent-117, score-0.497]

73 Comparison of the two panels of Figure 3 indicate that despite the highly simpliﬁed construction of the model, the prediction is quite good, and even includes the small but consistent repulsive effects observed 180 degrees from the adaptor. [sent-118, score-0.278]

74 2 Changes in discrimination threshold Adaptation also changes the ability of human observers to discriminate between the direction of two different moving stimuli. [sent-120, score-0.497]

75 In order to model discrimination thresholds, we need to consider a Bayesian framework that can account not only for the mean of the estimate but also its variability. [sent-121, score-0.298]

76 We have recently developed such a framework, and used it to quantitatively constrain the likelihood and the prior from psychophysical data [16]. [sent-122, score-0.23]

77 This framework accounts for the effect of the measurement noise on the variability of the ˆ ˆ estimate θ. [sent-123, score-0.375]

78 Speciﬁcally, it provides a characterization of the distribution p(θ|θstim ) of the estimate for a given stimulus direction in terms of its expected value and its variance as a function of the measurement noise. [sent-124, score-0.521]

79 As in [16] we write ˆ ∂ θ(m) 2 ˆ var θ|θstim = var m ( ) |m=θstim . [sent-125, score-0.084]

80 (3) ∂m Assuming that discrimination threshold is proportional to the standard deviation, ˆ var θ|θstim , we can now predict how discrimination thresholds should change after adaptation. [sent-126, score-0.566]

81 Figure 4a shows the predicted change in discrimination thresholds relative to the unadapted condition for the same model parameters as in the repulsion example (Figure 3a). [sent-127, score-0.888]

82 Thresholds are slightly reduced at the adaptor, but increase symmetrically for directions further away from the adaptor. [sent-128, score-0.146]

83 For comparison, Figure 4b shows the relative change in discrimination thresholds for a typical human subject [14]. [sent-129, score-0.429]

84 Again, the behavior of the human observer is qualitatively well predicted. [sent-130, score-0.231]

85 4 Discussion We have shown that adaptation can be incorporated into a Bayesian estimation framework for human sensory perception. [sent-131, score-0.682]

86 Adaptation seems unlikely to manifest itself as a change in the internal representation of prior distributions, as this would lead to perceptual bias effects that are opposite to those observed in human subjects. [sent-132, score-0.649]

87 Instead, we argue that adaptation leads to an increase in reliability of the measurement in the vicinity of the adapting stimulus parameter. [sent-133, score-1.13]

88 We show that this change in the measurement reliability results in changes of the likelihood function, and that an estimator that utilizes this likelihood function will exhibit the commonly-observed adaptation effects of repulsion and changes in discrimination threshold. [sent-134, score-1.747]

89 We further conﬁrm our model by making quantitative predictions and comparing them with known psychophysical data in the case of human perception of motion direction. [sent-135, score-0.405]

90 The results demonstrated here indicate that a resource allocation explanation is consistent with the functional effects of adaptation, but it seems unlikely that theory alone can lead to a unique quantitative prediction of the detailed form of these effects. [sent-137, score-0.265]

91 Speciﬁcally, the constraints imposed by biological implementation are likely to play a role in determining the changes in measurement noise as a function of adaptor parameter value, and it will be important to characterize and interpret neural response changes in the context of our framework. [sent-138, score-0.776]

92 Also, although we have argued that changes in the prior seem inconsistent with adaptation effects, it may be that such changes do occur but are offset by the likelihood effect, or occur only on much longer timescales. [sent-139, score-0.923]

93 Last, if one considers sensory perception as the result of a cascade of successive processing stages (with both feedforward and feedback connections), it becomes necessary to expand the Bayesian description to describe this cascade [e. [sent-140, score-0.424]

94 For example, it may be possible to interpret this cascade as a sequence of Bayesian estimators, in which the measurement of each stage consists of the estimate computed at the previous stage. [sent-143, score-0.357]

95 Adaptation could potentially occur in each of these processing stages, and it is of fundamental interest to understand how such a cascade can perform useful stable computations despite the fact that each of its elements is constantly readjusting its response properties. [sent-144, score-0.16]

96 Vision: Coding and Efﬁciency, chapter A theory about the functional role and synaptic mechanism of visual after-effects, pages 363–375. [sent-162, score-0.082]

97 Velocity after-effects: the effects of adaptation to moving stimuli on the perception of subsequently seen moving stimuli. [sent-189, score-0.739]

98 Increased sensitivity to speed changes during adaptation to ﬁrst-order, but not to second-order motion. [sent-213, score-0.525]

99 Constraining a Bayesian model of human visual speed perception. [sent-234, score-0.172]

100 Retinal light adaptation – evidence for a feedback mechanism. [sent-243, score-0.396]

similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('adaptation', 0.396), ('adaptor', 0.345), ('unadapted', 0.264), ('repulsion', 0.253), ('measurement', 0.251), ('stim', 0.184), ('discrimination', 0.177), ('perceptual', 0.169), ('stimulus', 0.146), ('perception', 0.142), ('adapt', 0.139), ('repulsive', 0.138), ('sensory', 0.13), ('bayesian', 0.124), ('effects', 0.116), ('prior', 0.109), ('perceived', 0.105), ('reliability', 0.101), ('observer', 0.097), ('vicinity', 0.097), ('adapting', 0.09), ('changes', 0.09), ('human', 0.089), ('adapted', 0.089), ('motion', 0.088), ('direction', 0.082), ('thresholds', 0.081), ('replotted', 0.079), ('likelihood', 0.074), ('estimator', 0.067), ('cascade', 0.064), ('operational', 0.063), ('deg', 0.059), ('yair', 0.059), ('change', 0.058), ('vision', 0.056), ('seems', 0.056), ('inconsistent', 0.055), ('velocity', 0.055), ('alters', 0.053), ('repelled', 0.053), ('symmetrically', 0.053), ('increase', 0.049), ('retinal', 0.047), ('psychophysical', 0.047), ('snr', 0.047), ('around', 0.046), ('weiss', 0.046), ('constituents', 0.046), ('qualitatively', 0.045), ('lawrence', 0.045), ('visual', 0.044), ('away', 0.044), ('bottou', 0.044), ('var', 0.042), ('effect', 0.042), ('estimate', 0.042), ('illusions', 0.042), ('stocker', 0.042), ('conditionals', 0.042), ('bialek', 0.042), ('framework', 0.04), ('increased', 0.04), ('account', 0.039), ('predictions', 0.039), ('resources', 0.039), ('speed', 0.039), ('functional', 0.038), ('occur', 0.037), ('saul', 0.037), ('posterior', 0.035), ('simoncelli', 0.035), ('vol', 0.035), ('offset', 0.035), ('fundamental', 0.035), ('convolutional', 0.034), ('sensorimotor', 0.032), ('decrease', 0.031), ('threshold', 0.031), ('march', 0.031), ('attracted', 0.031), ('predicted', 0.031), ('coding', 0.03), ('subsequent', 0.029), ('stimuli', 0.029), ('orders', 0.028), ('moving', 0.028), ('demonstrated', 0.028), ('ratio', 0.027), ('conditional', 0.027), ('unlikely', 0.027), ('estimation', 0.027), ('june', 0.026), ('opposite', 0.025), ('cambridge', 0.025), ('description', 0.024), ('despite', 0.024), ('relative', 0.024), ('spiking', 0.024), ('acts', 0.024)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 1.0000002 173 nips-2005-Sensory Adaptation within a Bayesian Framework for Perception

Author: Alan Stocker, Eero P. Simoncelli

2 0.12673596 3 nips-2005-A Bayesian Framework for Tilt Perception and Confidence

Author: Odelia Schwartz, Peter Dayan, Terrence J. Sejnowski

Abstract: The misjudgement of tilt in images lies at the heart of entertaining visual illusions and rigorous perceptual psychophysics. A wealth of ﬁndings has attracted many mechanistic models, but few clear computational principles. We adopt a Bayesian approach to perceptual tilt estimation, showing how a smoothness prior offers a powerful way of addressing much confusing data. In particular, we faithfully model recent results showing that conﬁdence in estimation can be systematically affected by the same aspects of images that affect bias. Conﬁdence is central to Bayesian modeling approaches, and is applicable in many other perceptual domains. Perceptual anomalies and illusions, such as the misjudgements of motion and tilt evident in so many psychophysical experiments, have intrigued researchers for decades.1–3 A Bayesian view4–8 has been particularly inﬂuential in models of motion processing, treating such anomalies as the normative product of prior information (often statistically codifying Gestalt laws) with likelihood information from the actual scenes presented. Here, we expand the range of statistically normative accounts to tilt estimation, for which there are classes of results (on estimation conﬁdence) that are so far not available for motion. The tilt illusion arises when the perceived tilt of a center target is misjudged (ie bias) in the presence of ﬂankers. Another phenomenon, called Crowding, refers to a loss in the conﬁdence (ie sensitivity) of perceived target tilt in the presence of ﬂankers. Attempts have been made to formalize these phenomena quantitatively. Crowding has been modeled as compulsory feature pooling (ie averaging of orientations), ignoring spatial positions.9, 10 The tilt illusion has been explained by lateral interactions11, 12 in populations of orientationtuned units; and by calibration.13 However, most models of this form cannot explain a number of crucial aspects of the data. First, the geometry of the positional arrangement of the stimuli affects attraction versus repulsion in bias, as emphasized by Kapadia et al14 (ﬁgure 1A), and others.15, 16 Second, Solomon et al. recently measured bias and sensitivity simultaneously.11 The rich and surprising range of sensitivities, far from ﬂat as a function of ﬂanker angles (ﬁgure 1B), are outside the reach of standard models. Moreover, current explanations do not offer a computational account of tilt perception as the outcome of a normative inference process. Here, we demonstrate that a Bayesian framework for orientation estimation, with a prior favoring smoothness, can naturally explain a range of seemingly puzzling tilt data. We explicitly consider both the geometry of the stimuli, and the issue of conﬁdence in the esti- 6 5 4 3 2 1 0 -1 -2 (B) Attraction Repulsion Sensititvity (1/deg) Bias (deg) (A) 0.6 0.5 0.4 0.3 0.2 0.1 -80 -60 -40 -20 0 20 40 60 80 Flanker tilt (deg) Figure 1: Tilt biases and sensitivities in visual perception. (A) Kapadia et al demonstrated the importance of geometry on tilt bias, with bar stimuli in the fovea (and similar results in the periphery). When 5 degrees clockwise ﬂankers are arranged colinearly, the center target appears attracted in the direction of the ﬂankers; when ﬂankers are lateral, the target appears repulsed. Data are an average of 5 subjects.14 (B) Solomon et al measured both biases and sensitivities for gratings in the visual periphery.11 On the top are example stimuli, with ﬂankers tilted 22.5 degrees clockwise. This constitutes the classic tilt illusion, with a repulsive bias percept. In addition, sensitivities vary as a function of ﬂanker angles, in a systematic way (even in cases when there are no biases at all). Sensitivities are given in units of the inverse of standard deviation of the tilt estimate. More detailed data for both experiments are shown in the results section. mation. Bayesian analyses have most frequently been applied to bias. Much less attention has been paid to the equally important phenomenon of sensitivity. This aspect of our model should be applicable to other perceptual domains. In section 1 we formulate the Bayesian model. The prior is determined by the principle of creating a smooth contour between the target and ﬂankers. We describe how to extract the bias and sensitivity. In section 2 we show experimental data of Kapadia et al and Solomon et al, alongside the model simulations, and demonstrate that the model can account for both geometry, and bias and sensitivity measurements in the data. Our results suggest a more uniﬁed, rational, approach to understanding tilt perception. 1 Bayesian model Under our Bayesian model, inference is controlled by the posterior distribution over the tilt of the target element. This comes from the combination of a prior favoring smooth conﬁgurations of the ﬂankers and target, and the likelihood associated with the actual scene. A complete distribution would consider all possible angles and relative spatial positions of the bars, and marginalize the posterior over all but the tilt of the central element. For simplicity, we make two benign approximations: conditionalizing over (ie clamping) the angles of the ﬂankers, and exploring only a small neighborhood of their positions. We now describe the steps of inference. Smoothness prior: Under these approximations, we consider a given actual conﬁguration (see ﬁg 2A) of ﬂankers f1 = (φ1 , x1 ), f2 = (φ2 , x2 ) and center target c = (φc , xc ), arranged from top to bottom. We have to generate a prior over φc and δ1 = x1 − xc and δ2 = x2 − xc based on the principle of smoothness. As a less benign approximation, we do this in two stages: articulating a principle that determines a single optimal conﬁguration; and generating a prior as a mixture of a Gaussian about this optimum and a uniform distribution, with the mixing proportion of the latter being determined by the smoothness of the optimum. Smoothness has been extensively studied in the computer vision literature.17–20 One widely (B) (C) f1 f1 β1 R Probability max smooth Max smooth target (deg) (A) 40 20 0 -20 c δ1 c -40 Φc f2 f2 1 0.8 0.6 0.4 0.2 0 -80 -60 -40 -20 0 20 40 Flanker tilt (deg) 60 80 -80 -60 -40 20 0 20 40 Flanker tilt (deg) 60 80 Figure 2: Geometry and smoothness for ﬂankers, f1 and f2 , and center target, c. (A) Example actual conﬁguration of ﬂankers and target, aligned along the y axis from top to bottom. (B) The elastica procedure can rotate the target angle (to Φc ) and shift the relative ﬂanker and target positions on the x axis (to δ1 and δ2 ) in its search for the maximally smooth solution. Small spatial shifts (up to 1/15 the size of R) of positions are allowed, but positional shift is overemphasized in the ﬁgure for visibility. (C) Top: center tilt that results in maximal smoothness, as a function of ﬂanker tilt. Boxed cartoons show examples for given ﬂanker tilts, of the optimally smooth conﬁguration. Note attraction of target towards ﬂankers for small ﬂanker angles; here ﬂankers and target are positioned in a nearly colinear arrangement. Note also repulsion of target away from ﬂankers for intermediate ﬂanker angles. Bottom: P [c, f1 , f2 ] for center tilt that yields maximal smoothness. The y axis is normalized between 0 and 1. used principle, elastica, known even to Euler, has been applied to contour completion21 and other computer vision applications.17 The basic idea is to ﬁnd the curve with minimum energy (ie, square of curvature). Sharon et al19 showed that the elastica function can be well approximated by a number of simpler forms. We adopt a version that Leung and Malik18 adopted from Sharon et al.19 We assume that the probability for completing a smooth curve, can be factorized into two terms: P [c, f1 , f2 ] = G(c, f1 )G(c, f2 ) (1) with the term G(c, f1 ) (and similarly, G(c, f2 )) written as: R Dβ 2 2 Dβ = β1 + βc − β1 βc (2) − ) where σR σβ and β1 (and similarly, βc ) is the angle between the orientation at f1 , and the line joining f1 and c. The distance between the centers of f1 and c is given by R. The two constants, σβ and σR , control the relative contribution to smoothness of the angle versus the spatial distance. Here, we set σβ = 1, and σR = 1.5. Figure 2B illustrates an example geometry, in which φc , δ1 , and δ2 , have been shifted from the actual scene (of ﬁgure 2A). G(c, f1 ) = exp(− We now estimate the smoothest solution for given conﬁgurations. Figure 2C shows for given ﬂanker tilts, the center tilt that yields maximal smoothness, and the corresponding probability of smoothness. For near vertical ﬂankers, the spatial lability leads to very weak attraction and high probability of smoothness. As the ﬂanker angle deviates farther from vertical, there is a large repulsion, but also lower probability of smoothness. These observations are key to our model: the maximally smooth center tilt will inﬂuence attractive and repulsive interactions of tilt estimation; the probability of smoothness will inﬂuence the relative weighting of the prior versus the likelihood. From the smoothness principle, we construct a two dimensional prior (ﬁgure 3A). One dimension represents tilt, the other dimension, the overall positional shift between target (B) Likelihood (D) Marginalized Posterior (C) Posterior 20 0.03 10 -10 -20 0 Probability 0 10 Angle Angle Angle 10 0 -10 -20 0.01 -10 -20 0.02 0 -0. 2 0 Position 0.2 (E) Psychometric function 20 -0. 2 0 0.2 -0. 2 0 0.2 Position Position -10 -5 0 Angle 5 10 Probability clockwise (A) Prior 20 1 0.8 0.6 0.4 0.2 0 -20 -10 0 10 20 Target angle (deg) Counter-clockwise Clockwise Figure 3: Bayes model for example ﬂankers and target. (A) Prior 2D distribution for ﬂankers set at 22.5 degrees (note repulsive preference for -5.5 degrees). (B) Likelihood 2D distribution for a target tilt of 3 degrees; (C) Posterior 2D distribution. All 2D distributions are drawn on the same grayscale range, and the presence of a larger baseline in the prior causes it to appear more dimmed. (D) Marginalized posterior, resulting in 1D distribution over tilt. Dashed line represents the mean, with slight preference for negative angle. (E) For this target tilt, we calculate probability clockwise, and obtain one point on psychometric curve. and ﬂankers (called ’position’). The prior is a 2D Gaussian distribution, sat upon a constant baseline.22 The Gaussian is centered at the estimated smoothest target angle and relative position, and the baseline is determined by the probability of smoothness. The baseline, and its dependence on the ﬂanker orientation, is a key difference from Weiss et al’s Gaussian prior for smooth, slow motion. It can be seen as a mechanism to allow segmentation (see Posterior description below). The standard deviation of the Gaussian is a free parameter. Likelihood: The likelihood over tilt and position (ﬁgure 3B) is determined by a 2D Gaussian distribution with an added baseline.22 The Gaussian is centered at the actual target tilt; and at a position taken as zero, since this is the actual position, to which the prior is compared. The standard deviation and baseline constant are free parameters. Posterior and marginalization: The posterior comes from multiplying likelihood and prior (ﬁgure 3C) and then marginalizing over position to obtain a 1D distribution over tilt. Figure 3D shows an example in which this distribution is bimodal. Other likelihoods, with closer agreement between target and smooth prior, give unimodal distributions. Note that the bimodality is a direct consequence of having an added baseline to the prior and likelihood (if these were Gaussian without a baseline, the posterior would always be Gaussian). The viewer is effectively assessing whether the target is associated with the same object as the ﬂankers, and this is reﬂected in the baseline, and consequently, in the bimodality, and conﬁdence estimate. We deﬁne α as the mean angle of the 1D posterior distribution (eg, value of dashed line on the x axis), and β as the height of the probability distribution at that mean angle (eg, height of dashed line). The term β is an indication of conﬁdence in the angle estimate, where for larger values we are more certain of the estimate. Decision of probability clockwise: The probability of a clockwise tilt is estimated from the marginalized posterior: 1 P = 1 + exp (3) −α.∗k − log(β+η) where α and β are deﬁned as above, k is a free parameter and η a small constant. Free parameters are set to a single constant value for all ﬂanker and center conﬁgurations. Weiss et al use a similar compressive nonlinearity, but without the term β. We also tried a decision function that integrates the posterior, but the resulting curves were far from the sigmoidal nature of the data. Bias and sensitivity: For one target tilt, we generate a single probability and therefore a single point on the psychometric function relating tilt to the probability of choosing clockwise. We generate the full psychometric curve from all target tilts and ﬁt to it a cumulative 60 40 20 -5 0 5 Target tilt (deg) 10 80 60 40 20 0 -10 (C) Data -5 0 5 Target tilt (deg) 10 80 60 40 20 0 -10 (D) Model 100 100 100 80 0 -10 Model Frequency responding clockwise (B) Data Frequency responding clockwise Frequency responding clockwise Frequency responding clockwise (A) 100 -5 0 5 Target tilt (deg) 10 80 60 40 20 0 -10 -5 0 5 10 Target tilt (deg) Figure 4: Kapadia et al data,14 versus Bayesian model. Solid lines are ﬁts to a cumulative Gaussian distribution. (A) Flankers are tilted 5 degrees clockwise (black curve) or anti-clockwise (gray) of vertical, and positioned spatially in a colinear arrangement. The center bar appears tilted in the direction of the ﬂankers (attraction), as can be seen by the attractive shift of the psychometric curve. The boxed stimuli cartoon illustrates a vertical target amidst the ﬂankers. (B) Model for colinear bars also produces attraction. (C) Data and (D) model for lateral ﬂankers results in repulsion. All data are collected in the fovea for bars. Gaussian distribution N (µ, σ) (ﬁgure 3E). The mean µ of the ﬁt corresponds to the bias, 1 and σ to the sensitivity, or conﬁdence in the bias. The ﬁt to a cumulative Gaussian and extraction of these parameters exactly mimic psychophysical procedures.11 2 Results: data versus model We ﬁrst consider the geometry of the center and ﬂanker conﬁgurations, modeling the full psychometric curve for colinear and parallel ﬂanks (recall that ﬁgure 1A showed summary biases). Figure 4A;B demonstrates attraction in the data and model; that is, the psychometric curve is shifted towards the ﬂanker, because of the nature of smooth completions for colinear ﬂankers. Figure 4C;D shows repulsion in the data and model. In this case, the ﬂankers are arranged laterally instead of colinearly. The smoothest solution in the model arises by shifting the target estimate away from the ﬂankers. This shift is rather minor, because the conﬁguration has a low probability of smoothness (similar to ﬁgure 2C), and thus the prior exerts only a weak effect. The above results show examples of changes in the psychometric curve, but do not address both bias and, particularly, sensitivity, across a whole range of ﬂanker conﬁgurations. Figure 5 depicts biases and sensitivity from Solomon et al, versus the Bayes model. The data are shown for a representative subject, but the qualitative behavior is consistent across all subjects tested. In ﬁgure 5A, bias is shown, for the condition that both ﬂankers are tilted at the same angle. The data exhibit small attraction at near vertical ﬂanker angles (this arrangement is close to colinear); large repulsion at intermediate ﬂanker angles of 22.5 and 45 degrees from vertical; and minimal repulsion at large angles from vertical. This behavior is also exhibited in the Bayes model (Figure 5B). For intermediate ﬂanker angles, the smoothest solution in the model is repulsive, and the effect of the prior is strong enough to induce a signiﬁcant repulsion. For large angles, the prior exerts almost no effect. Interestingly, sensitivity is far from ﬂat in both data and model. In the data (Figure 5C), there is most loss in sensitivity at intermediate ﬂanker angles of 22.5 and 45 degrees (ie, the subject is less certain); and sensitivity is higher for near vertical or near horizontal ﬂankers. The model shows the same qualitative behavior (Figure 5D). In the model, there are two factors driving sensitivity: one is the probability of completing a smooth curvature for a given ﬂanker conﬁguration, as in Figure 2B; this determines the strength of the prior. The other factor is certainty in a particular center estimation; this is determined by β, derived from the posterior distribution, and incorporated into the decision stage of the model Data 5 0 -60 -40 -80 -60 -40 -20 0 20 40 Flanker tilt (deg) -20 0 20 40 Flanker tilt (deg) 60 60 80 -60 -40 0.6 0.5 0.4 0.3 0.2 0.1 -20 0 20 40 Flanker tilt (deg) 60 80 60 80 -80 -60 -40 -20 0 20 40 Flanker tilt (deg) -80 -60 -40 -20 0 20 40 Flanker tilt (deg) 60 80 -20 0 20 40 Flanker tilt (deg) 60 80 (F) Bias (deg) 10 5 0 -5 0.6 0.5 0.4 0.3 0.2 0.1 -80 (D) 10 -10 -10 80 Sensitivity (1/deg) -80 5 0 -5 -80 -80 -60 -60 -40 -40 -20 0 20 40 Flanker tilt (deg) -20 0 20 40 Flanker tilt (deg) 60 -10 80 (H) 60 80 Sensitivity (1/deg) Sensititvity (1/deg) Bias (deg) 0.6 0.5 0.4 0.3 0.2 0.1 (G) Sensititvity (1/deg) 0 -5 (C) (E) 5 -5 -10 Model (B) 10 Bias (deg) Bias (deg) (A) 10 0.6 0.5 0.4 0.3 0.2 0.1 -80 -60 -40 Figure 5: Solomon et al data11 (subject FF), versus Bayesian model. (A) Data and (B) model biases with same-tilted ﬂankers; (C) Data and (D) model sensitivities with same-tilted ﬂankers; (E;G) data and (F;H) model as above, but for opposite-tilted ﬂankers (note that opposite-tilted data was collected for less ﬂanker angles). Each point in the ﬁgure is derived by ﬁtting a cummulative Gaussian distribution N (µ, σ) to corresponding psychometric curve, and setting bias 1 equal to µ and sensitivity to σ . In all experiments, ﬂanker and target gratings are presented in the visual periphery. Both data and model stimuli are averages of two conﬁgurations, on the left hand side (9 O’clock position) and right hand side (3 O’clock position). The conﬁgurations are similar to Figure 1 (B), but slightly shifted according to an iso-eccentric circle, so that all stimuli are similarly visible in the periphery. (equation 3). For ﬂankers that are far from vertical, the prior has minimal effect because one cannot ﬁnd a smooth solution (eg, the likelihood dominates), and thus sensitivity is higher. The low sensitivity at intermediate angles arises because the prior has considerable effect; and there is conﬂict between the prior (tilt, position), and likelihood (tilt, position). This leads to uncertainty in the target angle estimation . For ﬂankers near vertical, the prior exerts a strong effect; but there is less conﬂict between the likelihood and prior estimates (tilt, position) for a vertical target. This leads to more conﬁdence in the posterior estimate, and therefore, higher sensitivity. The only aspect that our model does not reproduce is the (more subtle) sensitivity difference between 0 and +/- 5 degree ﬂankers. Figure 5E-H depict data and model for opposite tilted ﬂankers. The bias is now close to zero in the data (Figure 5E) and model (Figure 5F), as would be expected (since the maximally smooth angle is now always roughly vertical). Perhaps more surprisingly, the sensitivities continue to to be non-ﬂat in the data (Figure 5G) and model (Figure 5H). This behavior arises in the model due to the strength of prior, and positional uncertainty. As before, there is most loss in sensitivity at intermediate angles. Note that to ﬁt Kapadia et al, simulations used a constant parameter of k = 9 in equation 3, whereas for the Solomon et al. simulations, k = 2.5. This indicates that, in our model, there was higher conﬁdence in the foveal experiments than in the peripheral ones. 3 Discussion We applied a Bayesian framework to the widely studied tilt illusion, and demonstrated the model on examples from two different data sets involving foveal and peripheral estimation. Our results support the appealing hypothesis that perceptual misjudgements are not a consequence of poor system design, but rather can be described as optimal inference.4–8 Our model accounts correctly for both attraction and repulsion, determined by the smoothness prior and the geometry of the scene. We emphasized the issue of estimation conﬁdence. The dataset showing how conﬁdence is affected by the same issues that affect bias,11 was exactly appropriate for a Bayesian formulation; other models in the literature typically do not incorporate conﬁdence in a thoroughly probabilistic manner. In fact, our model ﬁts the conﬁdence (and bias) data more proﬁciently than an account based on lateral interactions among a population of orientationtuned cells.11 Other Bayesian work, by Stocker et al,6 utilized the full slope of the psychometric curve in ﬁtting a prior and likelihood to motion data, but did not examine the issue of conﬁdence. Estimation conﬁdence plays a central role in Bayesian formulations as a whole. Understanding how priors affect conﬁdence should have direct bearing on many other Bayesian calculations such as multimodal integration.23 Our model is obviously over-simpliﬁed in a number of ways. First, we described it in terms of tilts and spatial positions; a more complete version should work in the pixel/ﬁltering domain.18, 19 We have also only considered two ﬂanking elements; the model is extendible to a full-ﬁeld surround, whereby smoothness operates along a range of geometric directions, and some directions are more (smoothly) dominant than others. Second, the prior is constructed by summarizing the maximal smoothness information; a more probabilistically correct version should capture the full probability of smoothness in its prior. Third, our model does not incorporate a formal noise representation; however, sensitivities could be inﬂuenced both by stimulus-driven noise and conﬁdence. Fourth, our model does not address attraction in the so-called indirect tilt illusion, thought to be mediated by a different mechanism. Finally, we have yet to account for neurophysiological data within this framework, and incorporate constraints at the neural implementation level. However, versions of our computations are oft suggested for intra-areal and feedback cortical circuits; and smoothness principles form a key part of the association ﬁeld connection scheme in Li’s24 dynamical model of contour integration in V1. Our model is connected to a wealth of literature in computer vision and perception. Notably, occlusion and contour completion might be seen as the extreme example in which there is no likelihood information at all for the center target; a host of papers have shown that under these circumstances, smoothness principles such as elastica and variants explain many aspects of perception. The model is also associated with many studies on contour integration motivated by Gestalt principles;25, 26 and exploration of natural scene statistics and Gestalt,27, 28 including the relation to contour grouping within a Bayesian framework.29, 30 Indeed, our model could be modiﬁed to include a prior from natural scenes. There are various directions for the experimental test and reﬁnement of our model. Most pressing is to determine bias and sensitivity for different center and ﬂanker contrasts. As in the case of motion, our model predicts that when there is more uncertainty in the center element, prior information is more dominant. Another interesting test would be to design a task such that the center element is actually part of a different ﬁgure and unrelated to the ﬂankers; our framework predicts that there would be minimal bias, because of segmentation. Our model should also be applied to other tilt-based illusions such as the Fraser spiral and Z¨ llner. Finally, our model can be applied to other perceptual domains;31 and given o the apparent similarities between the tilt illusion and the tilt after-effect, we plan to extend the model to adaptation, by considering smoothness in time as well as space. Acknowledgements This work was funded by the HHMI (OS, TJS) and the Gatsby Charitable Foundation (PD). We are very grateful to Serge Belongie, Leanne Chukoskie, Philip Meier and Joshua Solomon for helpful discussions. References [1] J J Gibson. Adaptation, after-effect, and contrast in the perception of tilted lines. Journal of Experimental Psychology, 20:553–569, 1937. [2] C Blakemore, R H S Carpentar, and M A Georgeson. Lateral inhibition between orientation detectors in the human visual system. Nature, 228:37–39, 1970. [3] J A Stuart and H M Burian. A study of separation difﬁculty: Its relationship to visual acuity in normal and amblyopic eyes. American Journal of Ophthalmology, 53:471–477, 1962. [4] A Yuille and H H Bulthoff. Perception as bayesian inference. In Knill and Whitman, editors, Bayesian decision theory and psychophysics, pages 123–161. Cambridge University Press, 1996. [5] Y Weiss, E P Simoncelli, and E H Adelson. Motion illusions as optimal percepts. Nature Neuroscience, 5:598–604, 2002. [6] A Stocker and E P Simoncelli. Constraining a bayesian model of human visual speed perception. Adv in Neural Info Processing Systems, 17, 2004. [7] D Kersten, P Mamassian, and A Yuille. Object perception as bayesian inference. Annual Review of Psychology, 55:271–304, 2004. [8] K Kording and D Wolpert. Bayesian integration in sensorimotor learning. Nature, 427:244–247, 2004. [9] L Parkes, J Lund, A Angelucci, J Solomon, and M Morgan. Compulsory averaging of crowded orientation signals in human vision. Nature Neuroscience, 4:739–744, 2001. [10] D G Pelli, M Palomares, and N J Majaj. Crowding is unlike ordinary masking: Distinguishing feature integration from detection. Journal of Vision, 4:1136–1169, 2002. [11] J Solomon, F M Felisberti, and M Morgan. Crowding and the tilt illusion: Toward a uniﬁed account. Journal of Vision, 4:500–508, 2004. [12] J A Bednar and R Miikkulainen. Tilt aftereffects in a self-organizing model of the primary visual cortex. Neural Computation, 12:1721–1740, 2000. [13] C W Clifford, P Wenderoth, and B Spehar. A functional angle on some after-effects in cortical vision. Proc Biol Sci, 1454:1705–1710, 2000. [14] M K Kapadia, G Westheimer, and C D Gilbert. Spatial distribution of contextual interactions in primary visual cortex and in visual perception. J Neurophysiology, 4:2048–262, 2000. [15] C C Chen and C W Tyler. Lateral modulation of contrast discrimination: Flanker orientation effects. Journal of Vision, 2:520–530, 2002. [16] I Mareschal, M P Sceniak, and R M Shapley. Contextual inﬂuences on orientation discrimination: binding local and global cues. Vision Research, 41:1915–1930, 2001. [17] D Mumford. Elastica and computer vision. In Chandrajit Bajaj, editor, Algebraic geometry and its applications. Springer Verlag, 1994. [18] T K Leung and J Malik. Contour continuity in region based image segmentation. In Proc. ECCV, pages 544–559, 1998. [19] E Sharon, A Brandt, and R Basri. Completion energies and scale. IEEE Pat. Anal. Mach. Intell., 22(10), 1997. [20] S W Zucker, C David, A Dobbins, and L Iverson. The organization of curve detection: coarse tangent ﬁelds. Computer Graphics and Image Processing, 9(3):213–234, 1988. [21] S Ullman. Filling in the gaps: the shape of subjective contours and a model for their generation. Biological Cybernetics, 25:1–6, 1976. [22] G E Hinton and A D Brown. Spiking boltzmann machines. Adv in Neural Info Processing Systems, 12, 1998. [23] R A Jacobs. What determines visual cue reliability? Trends in Cognitive Sciences, 6:345–350, 2002. [24] Z Li. A saliency map in primary visual cortex. Trends in Cognitive Science, 6:9–16, 2002. [25] D J Field, A Hayes, and R F Hess. Contour integration by the human visual system: evidence for a local “association ﬁeld”. Vision Research, 33:173–193, 1993. [26] J Beck, A Rosenfeld, and R Ivry. Line segregation. Spatial Vision, 4:75–101, 1989. [27] M Sigman, G A Cecchi, C D Gilbert, and M O Magnasco. On a common circle: Natural scenes and gestalt rules. PNAS, 98(4):1935–1940, 2001. [28] S Mahumad, L R Williams, K K Thornber, and K Xu. Segmentation of multiple salient closed contours from real images. IEEE Pat. Anal. Mach. Intell., 25(4):433–444, 1997. [29] W S Geisler, J S Perry, B J Super, and D P Gallogly. Edge co-occurence in natural images predicts contour grouping performance. Vision Research, 6:711–724, 2001. [30] J H Elder and R M Goldberg. Ecological statistics of gestalt laws for the perceptual organization of contours. Journal of Vision, 4:324–353, 2002. [31] S R Lehky and T J Sejnowski. Neural model of stereoacuity and depth interpolation based on a distributed representation of stereo disparity. Journal of Neuroscience, 10:2281–2299, 1990.

3 0.12337767 99 nips-2005-Integrate-and-Fire models with adaptation are good enough

Author: Renaud Jolivet, Alexander Rauch, Hans-rudolf Lüscher, Wulfram Gerstner

Abstract: Integrate-and-Fire-type models are usually criticized because of their simplicity. On the other hand, the Integrate-and-Fire model is the basis of most of the theoretical studies on spiking neuron models. Here, we develop a sequential procedure to quantitatively evaluate an equivalent Integrate-and-Fire-type model based on intracellular recordings of cortical pyramidal neurons. We ﬁnd that the resulting effective model is sufﬁcient to predict the spike train of the real pyramidal neuron with high accuracy. In in vivo-like regimes, predicted and recorded traces are almost indistinguishable and a signiﬁcant part of the spikes can be predicted at the correct timing. Slow processes like spike-frequency adaptation are shown to be a key feature in this context since they are necessary for the model to connect between different driving regimes. 1

4 0.12327406 100 nips-2005-Interpolating between types and tokens by estimating power-law generators

Author: Sharon Goldwater, Mark Johnson, Thomas L. Griffiths

Abstract: Standard statistical models of language fail to capture one of the most striking properties of natural languages: the power-law distribution in the frequencies of word tokens. We present a framework for developing statistical models that generically produce power-laws, augmenting standard generative models with an adaptor that produces the appropriate pattern of token frequencies. We show that taking a particular stochastic process – the Pitman-Yor process – as an adaptor justiﬁes the appearance of type frequencies in formal analyses of natural language, and improves the performance of a model for unsupervised learning of morphology.

5 0.11826181 140 nips-2005-Nonparametric inference of prior probabilities from Bayes-optimal behavior

Author: Liam Paninski

Abstract: We discuss a method for obtaining a subject’s a priori beliefs from his/her behavior in a psychophysics context, under the assumption that the behavior is (nearly) optimal from a Bayesian perspective. The method is nonparametric in the sense that we do not assume that the prior belongs to any ﬁxed class of distributions (e.g., Gaussian). Despite this increased generality, the method is relatively simple to implement, being based in the simplest case on a linear programming algorithm, and more generally on a straightforward maximum likelihood or maximum a posteriori formulation, which turns out to be a convex optimization problem (with no non-global local maxima) in many important cases. In addition, we develop methods for analyzing the uncertainty of these estimates. We demonstrate the accuracy of the method in a simple simulated coin-ﬂipping setting; in particular, the method is able to precisely track the evolution of the subject’s posterior distribution as more and more data are observed. We close by brieﬂy discussing an interesting connection to recent models of neural population coding.

6 0.11283839 93 nips-2005-Ideal Observers for Detecting Motion: Correspondence Noise

7 0.10727906 136 nips-2005-Noise and the two-thirds power Law

8 0.086300135 141 nips-2005-Norepinephrine and Neural Interrupts

9 0.079335853 34 nips-2005-Bayesian Surprise Attracts Human Attention

10 0.077602468 156 nips-2005-Prediction and Change Detection

11 0.069651261 101 nips-2005-Is Early Vision Optimized for Extracting Higher-order Dependencies?

12 0.065606259 109 nips-2005-Learning Cue-Invariant Visual Responses

13 0.065544508 28 nips-2005-Analyzing Auditory Neurons by Learning Distance Functions

14 0.063169152 5 nips-2005-A Computational Model of Eye Movements during Object Class Detection

15 0.062412597 45 nips-2005-Conditional Visual Tracking in Kernel Space

16 0.062209927 94 nips-2005-Identifying Distributed Object Representations in Human Extrastriate Visual Cortex

17 0.059705526 194 nips-2005-Top-Down Control of Visual Attention: A Rational Account

18 0.051714696 192 nips-2005-The Information-Form Data Association Filter

19 0.05140793 15 nips-2005-A Theoretical Analysis of Robust Coding over Noisy Overcomplete Channels

20 0.051068772 201 nips-2005-Variational Bayesian Stochastic Complexity of Mixture Models

similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, 0.169), (1, -0.087), (2, 0.026), (3, 0.151), (4, 0.055), (5, -0.001), (6, -0.049), (7, -0.137), (8, -0.081), (9, 0.025), (10, 0.049), (11, -0.015), (12, 0.083), (13, 0.015), (14, -0.037), (15, -0.018), (16, 0.022), (17, 0.192), (18, -0.035), (19, 0.104), (20, -0.088), (21, -0.01), (22, 0.11), (23, -0.105), (24, 0.024), (25, 0.235), (26, -0.093), (27, -0.062), (28, -0.087), (29, 0.012), (30, -0.122), (31, -0.028), (32, -0.079), (33, 0.074), (34, 0.012), (35, 0.1), (36, -0.057), (37, 0.049), (38, -0.091), (39, -0.13), (40, 0.073), (41, 0.046), (42, 0.012), (43, 0.088), (44, 0.032), (45, 0.041), (46, 0.015), (47, 0.03), (48, 0.037), (49, 0.066)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.97095871 173 nips-2005-Sensory Adaptation within a Bayesian Framework for Perception

Author: Alan Stocker, Eero P. Simoncelli

2 0.74451685 3 nips-2005-A Bayesian Framework for Tilt Perception and Confidence

Author: Odelia Schwartz, Peter Dayan, Terrence J. Sejnowski

3 0.73824298 93 nips-2005-Ideal Observers for Detecting Motion: Correspondence Noise

Author: Hongjing Lu, Alan L. Yuille

Abstract: We derive a Bayesian Ideal Observer (BIO) for detecting motion and solving the correspondence problem. We obtain Barlow and Tripathy’s classic model as an approximation. Our psychophysical experiments show that the trends of human performance are similar to the Bayesian Ideal, but overall human performance is far worse. We investigate ways to degrade the Bayesian Ideal but show that even extreme degradations do not approach human performance. Instead we propose that humans perform motion tasks using generic, general purpose, models of motion. We perform more psychophysical experiments which are consistent with humans using a Slow-and-Smooth model and which rule out an alternative model using Slowness. 1

4 0.63760126 156 nips-2005-Prediction and Change Detection

Author: Mark Steyvers, Scott Brown

Abstract: We measure the ability of human observers to predict the next datum in a sequence that is generated by a simple statistical process undergoing change at random points in time. Accurate performance in this task requires the identification of changepoints. We assess individual differences between observers both empirically, and using two kinds of models: a Bayesian approach for change detection and a family of cognitively plausible fast and frugal models. Some individuals detect too many changes and hence perform sub-optimally due to excess variability. Other individuals do not detect enough changes, and perform sub-optimally because they fail to notice short-term temporal trends. 1 Intr oduction Decision-making often requires a rapid response to change. For example, stock analysts need to quickly detect changes in the market in order to adjust investment strategies. Coaches need to track changes in a player’s performance in order to adjust strategy. When tracking changes, there are costs involved when either more or less changes are observed than actually occurred. For example, when using an overly conservative change detection criterion, a stock analyst might miss important short-term trends and interpret them as random fluctuations instead. On the other hand, a change may also be detected too readily. For example, in basketball, a player who makes a series of consecutive baskets is often identified as a “hot hand” player whose underlying ability is perceived to have suddenly increased [1,2]. This might lead to sub-optimal passing strategies, based on random fluctuations. We are interested in explaining individual differences in a sequential prediction task. Observers are shown stimuli generated from a simple statistical process with the task of predicting the next datum in the sequence. The latent parameters of the statistical process change discretely at random points in time. Performance in this task depends on the accurate detection of those changepoints, as well as inference about future outcomes based on the outcomes that followed the most recent inferred changepoint. There is much prior research in statistics on the problem of identifying changepoints [3,4,5]. In this paper, we adopt a Bayesian approach to the changepoint identification problem and develop a simple inference procedure to predict the next datum in a sequence. The Bayesian model serves as an ideal observer model and is useful to characterize the ways in which individuals deviate from optimality. The plan of the paper is as follows. We first introduce the sequential prediction task and discuss a Bayesian analysis of this prediction problem. We then discuss the results from a few individuals in this prediction task and show how the Bayesian approach can capture individual differences with a single “twitchiness” parameter that describes how readily changes are perceived in random sequences. We will show that some individuals are too twitchy: their performance is too variable because they base their predictions on too little of the recent data. Other individuals are not twitchy enough, and they fail to capture fast changes in the data. We also show how behavior can be explained with a set of fast and frugal models [6]. These are cognitively realistic models that operate under plausible computational constraints. 2 A pr ediction task wit h m ult iple c hange points In the prediction task, stimuli are presented sequentially and the task is to predict the next stimulus in the sequence. After t trials, the observer has been presented with stimuli y1, y2, …, yt and the task is to make a prediction about yt+1. After the prediction is made, the actual outcome yt+1 is revealed and the next trial proceeds to the prediction of yt+2. This procedure starts with y1 and is repeated for T trials. The observations yt are D-dimensional vectors with elements sampled from binomial distributions. The parameters of those distributions change discretely at random points in time such that the mean increases or decreases after a change point. This generates a sequence of observation vectors, y1, y2, …, yT, where each yt = {yt,1 … yt,D}. Each of the yt,d is sampled from a binomial distribution Bin(θt,d,K), so 0 ≤ yt,d ≤ K. The parameter vector θt ={θt,1 … θt,D} changes depending on the locations of the changepoints. At each time step, xt is a binary indicator for the occurrence of a changepoint occurring at time t+1. The parameter α determines the probability of a change occurring in the sequence. The generative model is specified by the following algorithm: 1. For d=1..D sample θ1,d from a Uniform(0,1) distribution 2. For t=2..T, (a) Sample xt-1 from a Bernoulli(α) distribution (b) If xt-1=0, then θt=θt-1, else for d=1..D sample θt,d from a Uniform(0,1) distribution (c) for d=1..D, sample yt from a Bin(θt,d,K) distribution Table 1 shows some data generated from the changepoint model with T=20, α=.1,and D=1. In the prediction task, y will be observed, but x and θ are not. Table 1: Example data t x θ y 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 0 0 0 1 0 0 1 0 0 0 0 0 1 0 1 0 0 0 0 0 .68 .68 .68 .68 .48 .48 .48 .74 .74 .74 .74 .74 .74 .19 .19 .87 .87 .87 .87 .87 9 7 8 7 4 4 4 9 8 3 6 7 8 2 1 8 9 9 8 8 3 A Bayesian pr ediction m ode l In both our Bayesian and fast-and-frugal analyses, the prediction task is decomposed into two inference procedures. First, the changepoint locations are identified. This is followed by predictive inference for the next outcome based on the most recent changepoint locations. Several Bayesian approaches have been developed for changepoint problems involving single or multiple changepoints [3,5]. We apply a Markov Chain Monte Carlo (MCMC) analysis to approximate the joint posterior distribution over changepoint assignments x while integrating out θ. Gibbs sampling will be used to sample from this posterior marginal distribution. The samples can then be used to predict the next outcome in the sequence. 3.1 I n f e r e nc e f o r c h a n g e p o i n t a s s i g n m e n t s . To apply Gibbs sampling, we evaluate the conditional probability of assigning a changepoint at time i, given all other changepoint assignments and the current α value. By integrating out θ, the conditional probability is P ( xi | x−i , y, α ) = ∫ P ( xi ,θ , α | x− i , y ) (1) θ where x− i represents all switch point assignments except xi. This can be simplified by considering the location of the most recent changepoint preceding and following time i and the outcomes occurring between these locations. Let niL be the number of time steps from the last changepoint up to and including the current time step i such that xi − nL =1 and xi − nL + j =0 for 0 < niL . Similarly, let niR be the number of time steps that i i follow time step i up to the next changepoint such that xi + n R =1 and xi + nR − j =0 for i R i 0 < n . Let y = L i ∑ i − niL < k ≤ i i yk and y = ∑ k < k ≤i + n R yk . The update equation for the R i i changepoint assignment can then be simplified to P ( xi = m | x−i ) ∝ ( ) ( ( ) D Γ 1 + y L + y R Γ 1 + Kn L + Kn R − y L − y R ⎧ i, j i, j i i i, j i, j ⎪ (1 − α ) ∏ L R Γ 2 + Kni + Kni ⎪ j =1 ⎪ ⎨ L L L R R R ⎪ D Γ 1 + yi, j Γ 1 + Kni − yi, j Γ 1 + yi, j Γ 1 + Kni − yi, j α∏ ⎪ Γ 2 + KniL Γ 2 + KniR ⎪ j =1 ⎩ ( ) ( ( ) ( ) ( ) ) ( ) m=0 ) (2) m =1 We initialize the Gibbs sampler by sampling each xt from a Bernoulli(α) distribution. All changepoint assignments are then updated sequentially by the Gibbs sampling equation above. The sampler is run for M iterations after which one set of changepoint assignments is saved. The Gibbs sampler is then restarted multiple times until S samples have been collected. Although we could have included an update equation for α, in this analysis we treat α as a known constant. This will be useful when characterizing the differences between human observers in terms of differences in α. 3.2 P r e d i c ti v e i n f er e n ce The next latent parameter value θt+1 and outcome yt+1 can be predicted on the basis of observed outcomes that occurred after the last inferred changepoint: θ t+1, j = t ∑ i =t* +1 yt+1, j = round (θt +1, j K ) yi, j / K , (3) where t* is the location of the most recent change point. By considering multiple Gibbs samples, we get a distribution over outcomes yt+1. We base the model predictions on the mean of this distribution. 3.3 I l l u s t r a t i o n o f m o d e l p er f o r m a n c e Figure 1 illustrates the performance of the model on a one dimensional sequence (D=1) generated from the changepoint model with T=160, α=0.05, and K=10. The Gibbs sampler was run for M=30 iterations and S=200 samples were collected. The top panel shows the actual changepoints (triangles) and the distribution of changepoint assignments averaged over samples. The bottom panel shows the observed data y (thin lines) as well as the θ values in the generative model (rescaled between 0 and 10). At locations with large changes between observations, the marginal changepoint probability is quite high. At other locations, the true change in the mean is very small, and the model is less likely to put in a changepoint. The lower right panel shows the distribution over predicted θt+1 values. xt 1 0.5 0 yt 10 1 5 θt+1 0.5 0 20 40 60 80 100 120 140 160 0 Figure 1. Results of model simulation. 4 Prediction experiment We tested performance of 9 human observers in the prediction task. The observers included the authors, a visitor, and one student who were aware of the statistical nature of the task as well as naïve students. The observers were seated in front of an LCD touch screen displaying a two-dimensional grid of 11 x 11 buttons. The changepoint model was used to generate a sequence of T=1500 stimuli for two binomial variables y1 and y2 (D=2, K=10). The change probability α was set to 0.1. The two variables y1 and y2 specified the two-dimensional button location. The same sequence was used for all observers. On each trial, the observer touched a button on the grid displayed on the touch screen. Following each button press, the button corresponding to the next {y1,y2} outcome in the sequence was highlighted. Observers were instructed to press the button that best predicted the next location of the highlighted button. The 1500 trials were divided into three blocks of 500 trials. Breaks were allowed between blocks. The whole experiment lasted between 15 and 30 minutes. Figure 2 shows the first 50 trials from the third block of the experiment. The top and bottom panels show the actual outcomes for the y1 and y2 button grid coordinates as well as the predictions for two observers (SB and MY). The figure shows that at trial 15, the y1 and y2 coordinates show a large shift followed by an immediate shift in observer’s MY predictions (on trial 16). Observer SB waits until trial 17 to make a shift. 10 5 0 outcomes SB predictions MY predictions 10 5 0 0 5 10 15 20 25 Trial 30 35 40 45 50 Figure 2. Trial by trial predictions from two observers. 4.1 T a s k er r o r We assessed prediction performance by comparing the prediction with the actual outcome in the sequence. Task error was measured by normalized city-block distance T 1 (4) task error= ∑ yt ,1 − ytO,1 + yt ,2 − ytO,2 (T − 1) t =2 where yO represents the observer’s prediction. Note that the very first trial is excluded from this calculation. Even though more suitable probabilistic measures for prediction error could have been adopted, we wanted to allow comparison of observer’s performance with both probabilistic and non-probabilistic models. Task error ranged from 2.8 (for participant MY) to 3.3 (for ML). We also assessed the performance of five models – their task errors ranged from 2.78 to 3.20. The Bayesian models (Section 3) had the lowest task errors, just below 2.8. This fits with our definition of the Bayesian models as “ideal observer” models – their task error is lower than any other model’s and any human observer’s task error. The fast and frugal models (Section 5) had task errors ranging from 2.85 to 3.20. 5 Modeling R esults We will refer to the models with the following letter codes: B=Bayesian Model, LB=limited Bayesian model, FF1..3=fast and frugal models 1..3. We assessed model fit by comparing the model’s prediction against the human observers’ predictions, again using a normalized city-block distance model error= T 1 ∑ ytM − ytO,1 + ytM − ytO,2 ,1 ,2 (T − 1) t=2 (5) where yM represents the model’s prediction. The model error for each individual observer is shown in Figure 3. It is important to note that because each model is associated with a set of free parameters, the parameters optimized for task error and model error are different. For Figure 3, the parameters were optimized to minimize Equation (5) for each individual observer, showing the extent to which these models can capture the performance of individual observers, not necessarily providing the best task performance. B LB FF1 FF2 MY MS MM EJ FF3 Model Error 2 1.5 1 0.5 0 PH NP DN SB ML 1 Figure 3. Model error for each individual observer. 5.1 B ay e s i a n p re d i ct i o n m o d e l s At each trial t, the model was provided with the sequence of all previous outcomes. The Gibbs sampling and inference procedures from Eq. (2) and (3) were applied with M=30 iterations and S=200 samples. The change probability α was a free parameter. In the full Bayesian model, the whole sequence of observations up to the current trial is available for prediction, leading to a memory requirement of up to T=1500 trials – a psychologically unreasonable assumption. We therefore also simulated a limited Bayesian model (LB) where the observed sequence was truncated to the last 10 outcomes. The LB model showed almost no decrement in task performance compared to the full Bayesian model. Figure 3 also shows that it fit human data quite well. 5.2 I n d i v i d u a l D i f f er e nc e s The right-hand panel of Figure 4 plots each observer’s task error as a function of the mean city-block distance between their subsequent button presses. This shows a clear U-shaped function. Observers with very variable predictions (e.g., ML and DN) had large average changes between successive button pushes, and also had large task error: These observers were too “twitchy”. Observers with very small average button changes (e.g., SB and NP) were not twitchy enough, and also had large task error. Observers in the middle had the lowest task error (e.g., MS and MY). The left-hand panel of Figure 4 shows the same data, but with the x-axis based on the Bayesian model fits. Instead of using mean button change distance to index twitchiness (as in 1 Error bars indicate bootstrapped 95% confidence intervals. the right-hand panel), the left-hand panel uses the estimated α parameters from the Bayesian model. A similar U-shaped pattern is observed: individuals with too large or too small α estimates have large task errors. 3.3 DN 3.2 Task Error ML SB 3.2 NP 3.1 Task Error 3.3 PH EJ 3 MM MS MY 2.9 2.8 10 -4 10 -3 10 -2 DN NP 3.1 3 PH EJ MM MS 2.9 B ML SB MY 2.8 10 -1 10 0 0.5 1 α 1.5 2 Mean Button Change 2.5 3 Figure 4. Task error vs. “twitchiness”. Left-hand panel indexes twitchiness using estimated α parameters from Bayesian model fits. Right-hand panel uses mean distance between successive predictions. 5.3 F a s t - a n d - F r u g a l ( F F ) p r e d ic t i o n m o d e l s These models perform the prediction task using simple heuristics that are cognitively plausible. The FF models keep a short memory of previous stimulus values and make predictions using the same two-step process as the Bayesian model. First, a decision is made as to whether the latent parameter θ has changed. Second, remembered stimulus values that occurred after the most recently detected changepoint are used to generate the next prediction. A simple heuristic is used to detect changepoints: If the distance between the most recent observation and prediction is greater than some threshold amount, a change is inferred. We defined the distance between a prediction (p) and an observation (y) as the difference between the log-likelihoods of y assuming θ=p and θ=y. Thus, if fB(.|θ, K) is the binomial density with parameters θ and K, the distance between observation y and prediction p is defined as d(y,p)=log(fB(y|y,K))-log(fB(y|p,K)). A changepoint on time step t+1 is inferred whenever d(yt,pt)>C. The parameter C governs the twitchiness of the model predictions. If C is large, only very dramatic changepoints will be detected, and the model will be too conservative. If C is small, the model will be too twitchy, and will detect changepoints on the basis of small random fluctuations. Predictions are based on the most recent M observations, which are kept in memory, unless a changepoint has been detected in which case only those observations occurring after the changepoint are used for prediction. The prediction for time step t+1 is simply the mean of these observations, say p. Human observers were reticent to make predictions very close to the boundaries. This was modeled by allowing the FF model to change its prediction for the next time step, yt+1, towards the mean prediction (0.5). This change reflects a two-way bet. If the probability of a change occurring is α, the best guess will be 0.5 if that change occurs, or the mean p if the change does not occur. Thus, the prediction made is actually yt+1=1/2 α+(1-α)p. Note that we do not allow perfect knowledge of the probability of a changepoint, α. Instead, an estimated value of α is used based on the number of changepoints detected in the data series up to time t. The FF model nests two simpler FF models that are psychologically interesting. If the twitchiness threshold parameter C becomes arbitrarily large, the model never detects a change and instead becomes a continuous running average model. Predictions from this model are simply a boxcar smooth of the data. Alternatively, if we assume no memory the model must based each prediction on only the previous stimulus (i.e., M=1). Above, in Figure 3, we labeled the complete FF model as FF1, the boxcar model as FF2 and the memoryless model FF3. Figure 3 showed that the complete FF model (FF1) fit the data from all observers significantly better than either the boxcar model (FF2) or the memoryless model (FF3). Exceptions were observers PH, DN and ML, for whom all three FF model fit equally well. This result suggests that our observers were (mostly) doing more than just keeping a running average of the data, or using only the most recent observation. The FF1 model fit the data about as well as the Bayesian models for all observers except MY and MS. Note that, in general, the FF1 and Bayesian model fits are very good: the average city block distance between the human data and the model prediction is around 0.75 (out of 10) buttons on both the x- and y-axes. 6 C onclusion We used an online prediction task to study changepoint detection. Human observers had to predict the next observation in stochastic sequences containing random changepoints. We showed that some observers are too “twitchy”: They perform poorly on the prediction task because they see changes where only random fluctuation exists. Other observers are not twitchy enough, and they perform poorly because they fail to see small changes. We developed a Bayesian changepoint detection model that performed the task optimally, and also provided a good fit to human data when sub-optimal parameter settings were used. Finally, we developed a fast-and-frugal model that showed how participants may be able to perform well at the task using minimal information and simple decision heuristics. Acknowledgments We thank Eric-Jan Wagenmakers and Mike Yi for useful discussions related to this work. This work was supported in part by a grant from the US Air Force Office of Scientific Research (AFOSR grant number FA9550-04-1-0317). R e f er e n ce s [1] Gilovich, T., Vallone, R. and Tversky, A. (1985). The hot hand in basketball: on the misperception of random sequences. Cognitive Psychology17, 295-314. [2] Albright, S.C. (1993a). A statistical analysis of hitting streaks in baseball. Journal of the American Statistical Association, 88, 1175-1183. [3] Stephens, D.A. (1994). Bayesian retrospective multiple changepoint identification. Applied Statistics 43(1), 159-178. [4] Carlin, B.P., Gelfand, A.E., & Smith, A.F.M. (1992). Hierarchical Bayesian analysis of changepoint problems. Applied Statistics 41(2), 389-405. [5] Green, P.J. (1995). Reversible jump Markov chain Monte Carlo computation and Bayesian model determination. Biometrika 82(4), 711-732. [6] Gigerenzer, G., & Goldstein, D.G. (1996). Reasoning the fast and frugal way: Models of bounded rationality. Psychological Review, 103, 650-669.

5 0.60792732 34 nips-2005-Bayesian Surprise Attracts Human Attention

Author: Laurent Itti, Pierre F. Baldi

Abstract: The concept of surprise is central to sensory processing, adaptation, learning, and attention. Yet, no widely-accepted mathematical theory currently exists to quantitatively characterize surprise elicited by a stimulus or event, for observers that range from single neurons to complex natural or engineered systems. We describe a formal Bayesian deﬁnition of surprise that is the only consistent formulation under minimal axiomatic assumptions. Surprise quantiﬁes how data affects a natural or artiﬁcial observer, by measuring the difference between posterior and prior beliefs of the observer. Using this framework we measure the extent to which humans direct their gaze towards surprising items while watching television and video games. We ﬁnd that subjects are strongly attracted towards surprising locations, with 72% of all human gaze shifts directed towards locations more surprising than the average, a ﬁgure which rises to 84% when considering only gaze targets simultaneously selected by all subjects. The resulting theory of surprise is applicable across different spatio-temporal scales, modalities, and levels of abstraction. Life is full of surprises, ranging from a great christmas gift or a new magic trick, to wardrobe malfunctions, reckless drivers, terrorist attacks, and tsunami waves. Key to survival is our ability to rapidly attend to, identify, and learn from surprising events, to decide on present and future courses of action [1]. Yet, little theoretical and computational understanding exists of the very essence of surprise, as evidenced by the absence from our everyday vocabulary of a quantitative unit of surprise: Qualities such as the “wow factor” have remained vague and elusive to mathematical analysis. Informal correlates of surprise exist at nearly all stages of neural processing. In sensory neuroscience, it has been suggested that only the unexpected at one stage is transmitted to the next stage [2]. Hence, sensory cortex may have evolved to adapt to, to predict, and to quiet down the expected statistical regularities of the world [3, 4, 5, 6], focusing instead on events that are unpredictable or surprising. Electrophysiological evidence for this early sensory emphasis onto surprising stimuli exists from studies of adaptation in visual [7, 8, 4, 9], olfactory [10, 11], and auditory cortices [12], subcortical structures like the LGN [13], and even retinal ganglion cells [14, 15] and cochlear hair cells [16]: neural response greatly attenuates with repeated or prolonged exposure to an initially novel stimulus. Surprise and novelty are also central to learning and memory formation [1], to the point that surprise is believed to be a necessary trigger for associative learning [17, 18], as supported by mounting evidence for a role of the hippocampus as a novelty detector [19, 20, 21]. Finally, seeking novelty is a well-identiﬁed human character trait, with possible association with the dopamine D4 receptor gene [22, 23, 24]. In the Bayesian framework, we develop the only consistent theory of surprise, in terms of the difference between the posterior and prior distributions of beliefs of an observer over the available class of models or hypotheses about the world. We show that this deﬁnition derived from ﬁrst principles presents key advantages over more ad-hoc formulations, typically relying on detecting outlier stimuli. Armed with this new framework, we provide direct experimental evidence that surprise best characterizes what attracts human gaze in large amounts of natural video stimuli. We here extend a recent pilot study [25], adding more comprehensive theory, large-scale human data collection, and additional analysis. 1 Theory Bayesian Deﬁnition of Surprise. We propose that surprise is a general concept, which can be derived from ﬁrst principles and formalized across spatio-temporal scales, sensory modalities, and, more generally, data types and data sources. Two elements are essential for a principled deﬁnition of surprise. First, surprise can exist only in the presence of uncertainty, which can arise from intrinsic stochasticity, missing information, or limited computing resources. A world that is purely deterministic and predictable in real-time for a given observer contains no surprises. Second, surprise can only be deﬁned in a relative, subjective, manner and is related to the expectations of the observer, be it a single synapse, neuronal circuit, organism, or computer device. The same data may carry different amount of surprise for different observers, or even for the same observer taken at different times. In probability and decision theory it can be shown that the only consistent and optimal way for modeling and reasoning about uncertainty is provided by the Bayesian theory of probability [26, 27, 28]. Furthermore, in the Bayesian framework, probabilities correspond to subjective degrees of beliefs in hypotheses or models which are updated, as data is acquired, using Bayes’ theorem as the fundamental tool for transforming prior belief distributions into posterior belief distributions. Therefore, within the same optimal framework, the only consistent deﬁnition of surprise must involve: (1) probabilistic concepts to cope with uncertainty; and (2) prior and posterior distributions to capture subjective expectations. Consistently with this Bayesian approach, the background information of an observer is captured by his/her/its prior probability distribution {P (M )}M ∈M over the hypotheses or models M in a model space M. Given this prior distribution of beliefs, the fundamental effect of a new data observation D on the observer is to change the prior distribution {P (M )}M ∈M into the posterior distribution {P (M |D)}M ∈M via Bayes theorem, whereby P (D|M ) ∀M ∈ M, P (M |D) = P (M ). (1) P (D) In this framework, the new data observation D carries no surprise if it leaves the observer beliefs unaffected, that is, if the posterior is identical to the prior; conversely, D is surprising if the posterior distribution resulting from observing D signiﬁcantly differs from the prior distribution. Therefore we formally measure surprise elicited by data as some distance measure between the posterior and prior distributions. This is best done using the relative entropy or Kullback-Leibler (KL) divergence [29]. Thus, surprise is deﬁned by the average of the log-odd ratio: P (M |D) S(D, M) = KL(P (M |D), P (M )) = P (M |D) log dM (2) P (M ) M taken with respect to the posterior distribution over the model class M. Note that KL is not symmetric but has well-known theoretical advantages, including invariance with respect to Figure 1: Computing surprise in early sensory neurons. (a) Prior data observations, tuning preferences, and top-down inﬂuences contribute to shaping a set of “prior beliefs” a neuron may have over a class of internal models or hypotheses about the world. For instance, M may be a set of Poisson processes parameterized by the rate λ, with {P (M )}M ∈M = {P (λ)}λ∈I +∗ the prior distribution R of beliefs about which Poisson models well describe the world as sensed by the neuron. New data D updates the prior into the posterior using Bayes’ theorem. Surprise quantiﬁes the difference between the posterior and prior distributions over the model class M. The remaining panels detail how surprise differs from conventional model ﬁtting and outlier-based novelty. (b) In standard iterative Bayesian model ﬁtting, at every iteration N , incoming data DN is used to update the prior {P (M |D1 , D2 , ..., DN −1 )}M ∈M into the posterior {P (M |D1 , D2 , ..., DN )}M ∈M . Freezing this learning at a given iteration, one then picks the currently best model, usually using either a maximum likelihood criterion, or a maximum a posteriori one (yielding MM AP shown). (c) This best model is used for a number of tasks at the current iteration, including outlier-based novelty detection. New data is then considered novel at that instant if it has low likelihood for the best model b a (e.g., DN is more novel than DN ). This focus onto the single best model presents obvious limitations, especially in situations where other models are nearly as good (e.g., M∗ in panel (b) is entirely ignored during standard novelty computation). One palliative solution is to consider mixture models, or simply P (D), but this just amounts to shifting the problem into a different model class. (d) Surprise directly addresses this problem by simultaneously considering all models and by measuring how data changes the observer’s distribution of beliefs from {P (M |D1 , D2 , ..., DN −1 )}M ∈M to {P (M |D1 , D2 , ..., DN )}M ∈M over the entire model class M (orange shaded area). reparameterizations. A unit of surprise — a “wow” — may then be deﬁned for a single model M as the amount of surprise corresponding to a two-fold variation between P (M |D) and P (M ), i.e., as log P (M |D)/P (M ) (with log taken in base 2), with the total number of wows experienced for all models obtained through the integration in eq. 2. Surprise and outlier detection. Outlier detection based on the likelihood P (D|M best ) of D given a single best model Mbest is at best an approximation to surprise and, in some cases, is misleading. Consider, for instance, a case where D has very small probability both for a model or hypothesis M and for a single alternative hypothesis M. Although D is a strong outlier, it carries very little information regarding whether M or M is the better model, and therefore very little surprise. Thus an outlier detection method would strongly focus attentional resources onto D, although D is a false positive, in the sense that it carries no useful information for discriminating between the two alternative hypotheses M and M. Figure 1 further illustrates this disconnect between outlier detection and surprise. 2 Human experiments To test the surprise hypothesis — that surprise attracts human attention and gaze in natural scenes — we recorded eye movements from eight na¨ve observers (three females and ı ﬁve males, ages 23-32, normal or corrected-to-normal vision). Each watched a subset from 50 videoclips totaling over 25 minutes of playtime (46,489 video frames, 640 × 480, 60.27 Hz, mean screen luminance 30 cd/m2 , room 4 cd/m2 , viewing distance 80cm, ﬁeld of view 28◦ × 21◦ ). Clips comprised outdoors daytime and nighttime scenes of crowded environments, video games, and television broadcast including news, sports, and commercials. Right-eye position was tracked with a 240 Hz video-based device (ISCAN RK-464), with methods as previously [30]. Two hundred calibrated eye movement traces (10,192 saccades) were analyzed, corresponding to four distinct observers for each of the 50 clips. Figure 2 shows sample scanpaths for one videoclip. To characterize image regions selected by participants, we process videoclips through computational metrics that output a topographic dynamic master response map, assigning in real-time a response value to every input location. A good master map would highlight, more than expected by chance, locations gazed to by observers. To score each metric we hence sample, at onset of every human saccade, master map activity around the saccade’s future endpoint, and around a uniformly random endpoint (random sampling was repeated 100 times to evaluate variability). We quantify differences between histograms of master Figure 2: (a) Sample eye movement traces from four observers (squares denote saccade endpoints). (b) Our data exhibits high inter-individual overlap, shown here with the locations where one human saccade endpoint was nearby (≈ 5◦ ) one (white squares), two (cyan squares), or all three (black squares) other humans. (c) A metric where the master map was created from the three eye movement traces other than that being tested yields an upper-bound KL score, computed by comparing the histograms of metric values at human (narrow blue bars) and random (wider green bars) saccade targets. Indeed, this metric’s map was very sparse (many random saccades landing on locations with nearzero response), yet humans preferentially saccaded towards the three active hotspots corresponding to the eye positions of three other humans (many human saccades landing on locations with near-unity responses). map samples collected from human and random saccades using again the Kullback-Leibler (KL) distance: metrics which better predict human scanpaths exhibit higher distances from random as, typically, observers non-uniformly gaze towards a minority of regions with highest metric responses while avoiding a majority of regions with low metric responses. This approach presents several advantages over simpler scoring schemes [31, 32], including agnosticity to putative mechanisms for generating saccades and the fact that applying any continuous nonlinearity to master map values would not affect scoring. Experimental results. We test six computational metrics, encompassing and extending the state-of-the-art found in previous studies. The ﬁrst three quantify static image properties (local intensity variance in 16 × 16 image patches [31]; local oriented edge density as measured with Gabor ﬁlters [33]; and local Shannon entropy in 16 × 16 image patches [34]). The remaining three metrics are more sensitive to dynamic events (local motion [33]; outlier-based saliency [33]; and surprise [25]). For all metrics, we ﬁnd that humans are signiﬁcantly attracted by image regions with higher metric responses. However, the static metrics typically respond vigorously at numerous visual locations (Figure 3), hence they are poorly speciﬁc and yield relatively low KL scores between humans and random. The metrics sensitive to motion, outliers, and surprising events, in comparison, yield sparser maps and higher KL scores. The surprise metric of interest here quantiﬁes low-level surprise in image patches over space and time, and at this point does not account for high-level or cognitive beliefs of our human observers. Rather, it assumes a family of simple models for image patches, each processed through 72 early feature detectors sensitive to color, orientation, motion, etc., and computes surprise from shifts in the distribution of beliefs about which models better describe the patches (see [25] and [35] for details). We ﬁnd that the surprise metric signiﬁcantly outperforms all other computational metrics (p < 10−100 or better on t-tests for equality of KL scores), scoring nearly 20% better than the second-best metric (saliency) and 60% better than the best static metric (entropy). Surprising stimuli often substantially differ from simple feature outliers; for example, a continually blinking light on a static background elicits sustained ﬂicker due to its locally outlier temporal dynamics but is only surprising for a moment. Similarly, a shower of randomly-colored pixels continually excites all low-level feature detectors but rapidly becomes unsurprising. Strongest attractors of human attention. Clearly, in our and previous eye-tracking experiments, in some situations potentially interesting targets were more numerous than in others. With many possible targets, different observers may orient towards different locations, making it more difﬁcult for a single metric to accurately predict all observers. Hence we consider (Figure 4) subsets of human saccades where at least two, three, or all four observers simultaneously agreed on a gaze target. Observers could have agreed based on bottom-up factors (e.g., only one location had interesting visual appearance at that time), top-down factors (e.g., only one object was of current cognitive interest), or both (e.g., a single cognitively interesting object was present which also had distinctive appearance). Irrespectively of the cause for agreement, it indicates consolidated belief that a location was attractive. While the KL scores of all metrics improved when progressively focusing onto only those locations, dynamic metrics improved more steeply, indicating that stimuli which more reliably attracted all observers carried more motion, saliency, and surprise. Surprise remained signiﬁcantly the best metric to characterize these agreed-upon attractors of human gaze (p < 10−100 or better on t-tests for equality of KL scores). Overall, surprise explained the greatest fraction of human saccades, indicating that humans are signiﬁcantly attracted towards surprising locations in video displays. Over 72% of all human saccades were targeted to locations predicted to be more surprising than on average. When only considering saccades where two, three, or four observers agreed on a common gaze target, this ﬁgure rose to 76%, 80%, and 84%, respectively. Figure 3: (a) Sample video frames, with corresponding human saccades and predictions from the entropy, surprise, and human-derived metrics. Entropy maps, like intensity variance and orientation maps, exhibited many locations with high responses, hence had low speciﬁcity and were poorly discriminative. In contrast, motion, saliency, and surprise maps were much sparser and more speciﬁc, with surprise signiﬁcantly more often on target. For three example frames (ﬁrst column), saccades from one subject are shown (arrows) with corresponding apertures over which master map activity at the saccade endpoint was sampled (circles). (b) KL scores for these metrics indicate signiﬁcantly different performance levels, and a strict ranking of variance < orientation < entropy < motion < saliency < surprise < human-derived. KL scores were computed by comparing the number of human saccades landing onto each given range of master map values (narrow blue bars) to the number of random saccades hitting the same range (wider green bars). A score of zero would indicate equality between the human and random histograms, i.e., humans did not tend to hit various master map values any differently from expected by chance, or, the master map could not predict human saccades better than random saccades. Among the six computational metrics tested in total, surprise performed best, in that surprising locations were relatively few yet reliably gazed to by humans. Figure 4: KL scores when considering only saccades where at least one (all 10,192 saccades), two (7,948 saccades), three (5,565 saccades), or all four (2,951 saccades) humans agreed on a common gaze location, for the static (a) and dynamic metrics (b). Static metrics improved substantially when progressively focusing onto saccades with stronger inter-observer agreement (average slope 0.56 ± 0.37 percent KL score units per 1,000 pruned saccades). Hence, when humans agreed on a location, they also tended to be more reliably predicted by the metrics. Furthermore, dynamic metrics improved 4.5 times more steeply (slope 2.44 ± 0.37), suggesting a stronger role of dynamic events in attracting human attention. Surprising events were signiﬁcantly the strongest (t-tests for equality of KL scores between surprise and other metrics, p < 10−100 ). 3 Discussion While previous research has shown with either static scenes or dynamic synthetic stimuli that humans preferentially ﬁxate regions of high entropy [34], contrast [31], saliency [32], ﬂicker [36], or motion [37], our data provides direct experimental evidence that humans ﬁxate surprising locations even more reliably. These conclusions were made possible by developing new tools to quantify what attracts human gaze over space and time in dynamic natural scenes. Surprise explained best where humans look when considering all saccades, and even more so when restricting the analysis to only those saccades for which human observers tended to agree. Surprise hence represents an inexpensive, easily computable approximation to human attentional allocation. In the absence of quantitative tools to measure surprise, most experimental and modeling work to date has adopted the approximation that novel events are surprising, and has focused on experimental scenarios which are simple enough to ensure an overlap between informal notions of novelty and surprise: for example, a stimulus is novel during testing if it has not been seen during training [9]. Our deﬁnition opens new avenues for more sophisticated experiments, where surprise elicited by different stimuli can be precisely compared and calibrated, yielding predictions at the single-unit as well as behavioral levels. The deﬁnition of surprise — as the distance between the posterior and prior distributions of beliefs over models — is entirely general and readily applicable to the analysis of auditory, olfactory, gustatory, or somatosensory data. While here we have focused on behavior rather than detailed biophysical implementation, it is worth noting that detecting surprise in neural spike trains does not require semantic understanding of the data carried by the spike trains, and thus could provide guiding signals during self-organization and development of sensory areas. At higher processing levels, top-down cues and task demands are known to combine with stimulus novelty in capturing attention and triggering learning [1, 38], ideas which may now be formalized and quantiﬁed in terms of priors, posteriors, and surprise. Surprise, indeed, inherently depends on uncertainty and on prior beliefs. Hence surprise theory can further be tested and utilized in experiments where the prior is biased, for ex- ample by top-down instructions or prior exposures to stimuli [38]. In addition, simple surprise-based behavioral measures such as the eye-tracking one used here may prove useful for early diagnostic of human conditions including autism and attention-deﬁcit hyperactive disorder, as well as for quantitative comparison between humans and animals which may have lower or different priors, including monkeys, frogs, and ﬂies. Beyond sensory biology, computable surprise could guide the development of data mining and compression systems (giving more bits to surprising regions of interest), to ﬁnd surprising agents in crowds, surprising sentences in books or speeches, surprising sequences in genomes, surprising medical symptoms, surprising odors in airport luggage racks, surprising documents on the world-wide-web, or to design surprising advertisements. Acknowledgments: Supported by HFSP, NSF and NGA (L.I.), NIH and NSF (P.B.). We thank UCI’s Institute for Genomics and Bioinformatics and USC’s Center High Performance Computing and Communications (www.usc.edu/hpcc) for access to their computing clusters. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] Ranganath, C. & Rainer, G. Nat Rev Neurosci 4, 193–202 (2003). Rao, R. P. & Ballard, D. H. Nat Neurosci 2, 79–87 (1999). Olshausen, B. A. & Field, D. J. Nature 381, 607–609 (1996). M¨ ller, J. R., Metha, A. B., Krauskopf, J. & Lennie, P. Science 285, 1405–1408 (1999). u Dragoi, V., Sharma, J., Miller, E. K. & Sur, M. Nat Neurosci 5, 883–891 (2002). David, S. V., Vinje, W. E. & Gallant, J. L. J Neurosci 24, 6991–7006 (2004). Maffei, L., Fiorentini, A. & Bisti, S. Science 182, 1036–1038 (1973). Movshon, J. A. & Lennie, P. Nature 278, 850–852 (1979). Fecteau, J. H. & Munoz, D. P. Nat Rev Neurosci 4, 435–443 (2003). Kurahashi, T. & Menini, A. Nature 385, 725–729 (1997). Bradley, J., Bonigk, W., Yau, K. W. & Frings, S. Nat Neurosci 7, 705–710 (2004). Ulanovsky, N., Las, L. & Nelken, I. Nat Neurosci 6, 391–398 (2003). Solomon, S. G., Peirce, J. W., Dhruv, N. T. & Lennie, P. Neuron 42, 155–162 (2004). Smirnakis, S. M., Berry, M. J. & et al. Nature 386, 69–73 (1997). Brown, S. P. & Masland, R. H. Nat Neurosci 4, 44–51 (2001). Kennedy, H. J., Evans, M. G. & et al. Nat Neurosci 6, 832–836 (2003). Schultz, W. & Dickinson, A. Annu Rev Neurosci 23, 473–500 (2000). Fletcher, P. C., Anderson, J. M., Shanks, D. R. et al. Nat Neurosci 4, 1043–1048 (2001). Knight, R. Nature 383, 256–259 (1996). Stern, C. E., Corkin, S., Gonzalez, R. G. et al. Proc Natl Acad Sci U S A 93, 8660–8665 (1996). Li, S., Cullen, W. K., Anwyl, R. & Rowan, M. J. Nat Neurosci 6, 526–531 (2003). Ebstein, R. P., Novick, O., Umansky, R. et al. Nat Genet 12, 78–80 (1996). Benjamin, J., Li, L. & et al. Nat Genet 12, 81–84 (1996). Lusher, J. M., Chandler, C. & Ball, D. Mol Psychiatry 6, 497–499 (2001). Itti, L. & Baldi, P. In Proc. IEEE CVPR. San Siego, CA (2005 in press). Cox, R. T. Am. J. Phys. 14, 1–13 (1964). Savage, L. J. The foundations of statistics (Dover, New York, 1972). (First Edition in 1954). Jaynes, E. T. Probability Theory. The Logic of Science (Cambridge University Press, 2003). Kullback, S. Information Theory and Statistics (Wiley, New York:New York, 1959). Itti, L. Visual Cognition (2005 in press). Reinagel, P. & Zador, A. M. Network 10, 341–350 (1999). Parkhurst, D., Law, K. & Niebur, E. Vision Res 42, 107–123 (2002). Itti, L. & Koch, C. Nat Rev Neurosci 2, 194–203 (2001). Privitera, C. M. & Stark, L. W. IEEE Trans Patt Anal Mach Intell 22, 970–982 (2000). All source code for all metrics is freely available at http://iLab.usc.edu/toolkit/. Theeuwes, J. Percept Psychophys 57, 637–644 (1995). Abrams, R. A. & Christ, S. E. Psychol Sci 14, 427–432 (2003). Wolfe, J. M. & Horowitz, T. S. Nat Rev Neurosci 5, 495–501 (2004).

6 0.57764447 140 nips-2005-Nonparametric inference of prior probabilities from Bayes-optimal behavior

7 0.5186401 35 nips-2005-Bayesian model learning in human visual perception

8 0.4227114 136 nips-2005-Noise and the two-thirds power Law

9 0.39200816 4 nips-2005-A Bayesian Spatial Scan Statistic

10 0.38239574 100 nips-2005-Interpolating between types and tokens by estimating power-law generators

11 0.35153303 141 nips-2005-Norepinephrine and Neural Interrupts

12 0.32540038 99 nips-2005-Integrate-and-Fire models with adaptation are good enough

13 0.32001856 171 nips-2005-Searching for Character Models

14 0.31643713 94 nips-2005-Identifying Distributed Object Representations in Human Extrastriate Visual Cortex

15 0.31072912 101 nips-2005-Is Early Vision Optimized for Extracting Higher-order Dependencies?

16 0.30699489 194 nips-2005-Top-Down Control of Visual Attention: A Rational Account

17 0.30529204 45 nips-2005-Conditional Visual Tracking in Kernel Space

18 0.30393967 48 nips-2005-Context as Filtering

19 0.30234605 119 nips-2005-Learning to Control an Octopus Arm with Gaussian Process Temporal Difference Methods

20 0.30110171 26 nips-2005-An exploration-exploitation model based on norepinepherine and dopamine activity

similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(3, 0.044), (10, 0.061), (11, 0.019), (27, 0.06), (31, 0.049), (34, 0.074), (39, 0.029), (55, 0.061), (56, 0.271), (60, 0.024), (69, 0.063), (88, 0.084), (91, 0.044)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.84143996 173 nips-2005-Sensory Adaptation within a Bayesian Framework for Perception

Author: Alan Stocker, Eero P. Simoncelli

2 0.80066746 111 nips-2005-Learning Influence among Interacting Markov Chains

Author: Dong Zhang, Daniel Gatica-perez, Samy Bengio, Deb Roy

Abstract: We present a model that learns the inﬂuence of interacting Markov chains within a team. The proposed model is a dynamic Bayesian network (DBN) with a two-level structure: individual-level and group-level. Individual level models actions of each player, and the group-level models actions of the team as a whole. Experiments on synthetic multi-player games and a multi-party meeting corpus show the effectiveness of the proposed model. 1

3 0.52739805 3 nips-2005-A Bayesian Framework for Tilt Perception and Confidence

Author: Odelia Schwartz, Peter Dayan, Terrence J. Sejnowski

4 0.5172804 96 nips-2005-Inference with Minimal Communication: a Decision-Theoretic Variational Approach

Author: O. P. Kreidl, Alan S. Willsky

Abstract: Given a directed graphical model with binary-valued hidden nodes and real-valued noisy observations, consider deciding upon the maximum a-posteriori (MAP) or the maximum posterior-marginal (MPM) assignment under the restriction that each node broadcasts only to its children exactly one single-bit message. We present a variational formulation, viewing the processing rules local to all nodes as degrees-of-freedom, that minimizes the loss in expected (MAP or MPM) performance subject to such online communication constraints. The approach leads to a novel message-passing algorithm to be executed ofﬂine, or before observations are realized, which mitigates the performance loss by iteratively coupling all rules in a manner implicitly driven by global statistics. We also provide (i) illustrative examples, (ii) assumptions that guarantee convergence and efﬁciency and (iii) connections to active research areas. 1

5 0.51434636 72 nips-2005-Fast Online Policy Gradient Learning with SMD Gain Vector Adaptation

Author: Jin Yu, Douglas Aberdeen, Nicol N. Schraudolph

Abstract: Reinforcement learning by direct policy gradient estimation is attractive in theory but in practice leads to notoriously ill-behaved optimization problems. We improve its robustness and speed of convergence with stochastic meta-descent, a gain vector adaptation method that employs fast Hessian-vector products. In our experiments the resulting algorithms outperform previously employed online stochastic, ofﬂine conjugate, and natural policy gradient methods. 1

6 0.50604326 144 nips-2005-Off-policy Learning with Options and Recognizers

7 0.50336474 67 nips-2005-Extracting Dynamical Structure Embedded in Neural Activity

8 0.5032258 200 nips-2005-Variable KD-Tree Algorithms for Spatial Pattern Search and Discovery

9 0.50182784 136 nips-2005-Noise and the two-thirds power Law

10 0.50093573 153 nips-2005-Policy-Gradient Methods for Planning

11 0.50048667 74 nips-2005-Faster Rates in Regression via Active Learning

12 0.50028223 30 nips-2005-Assessing Approximations for Gaussian Process Classification

13 0.49940357 23 nips-2005-An Application of Markov Random Fields to Range Sensing

14 0.49849832 32 nips-2005-Augmented Rescorla-Wagner and Maximum Likelihood Estimation

15 0.49786949 78 nips-2005-From Weighted Classification to Policy Search

16 0.49770218 43 nips-2005-Comparing the Effects of Different Weight Distributions on Finding Sparse Representations

17 0.49740657 109 nips-2005-Learning Cue-Invariant Visual Responses

18 0.49366415 50 nips-2005-Convex Neural Networks

19 0.49299163 28 nips-2005-Analyzing Auditory Neurons by Learning Distance Functions

20 0.49114788 154 nips-2005-Preconditioner Approximations for Probabilistic Graphical Models