nips nips2003 nips2003-95 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Felix A. Wichmann, Arnulf B. Graf
Abstract: We attempt to understand visual classification in humans using both psychophysical and machine learning techniques. Frontal views of human faces were used for a gender classification task. Human subjects classified the faces and their gender judgment, reaction time and confidence rating were recorded. Several hyperplane learning algorithms were used on the same classification task using the Principal Components of the texture and shape representation of the faces. The classification performance of the learning algorithms was estimated using the face database with the true gender of the faces as labels, and also with the gender estimated by the subjects. We then correlated the human responses to the distance of the stimuli to the separating hyperplane of the learning algorithms. Our results suggest that human classification can be modeled by some hyperplane algorithms in the feature space we used. For classification, the brain needs more processing for stimuli close to that hyperplane than for those further away. 1
Reference: text
sentIndex sentText sentNum sentScore
1 de Abstract We attempt to understand visual classification in humans using both psychophysical and machine learning techniques. [sent-8, score-0.227]
2 Frontal views of human faces were used for a gender classification task. [sent-9, score-0.604]
3 Human subjects classified the faces and their gender judgment, reaction time and confidence rating were recorded. [sent-10, score-0.869]
4 Several hyperplane learning algorithms were used on the same classification task using the Principal Components of the texture and shape representation of the faces. [sent-11, score-0.175]
5 The classification performance of the learning algorithms was estimated using the face database with the true gender of the faces as labels, and also with the gender estimated by the subjects. [sent-12, score-0.946]
6 We then correlated the human responses to the distance of the stimuli to the separating hyperplane of the learning algorithms. [sent-13, score-0.547]
7 Our results suggest that human classification can be modeled by some hyperplane algorithms in the feature space we used. [sent-14, score-0.266]
8 For classification, the brain needs more processing for stimuli close to that hyperplane than for those further away. [sent-15, score-0.341]
9 1 Introduction The last decade has seen tremendous technological advances in neuroscience from the microscopic to the macroscopic scale (e. [sent-16, score-0.107]
10 On an algorithmic level, however, methods and understanding of brain processes are still limited. [sent-19, score-0.072]
11 Here we report on a study combining psychophysical and machine learning techniques in order to improve our understanding of human classification of visual stimuli. [sent-20, score-0.274]
12 What algorithms best describe the way the human brain classifies? [sent-21, score-0.219]
13 Might humans use something akin to hyperplanes for classification? [sent-22, score-0.203]
14 If so, is the learning rule as simple as in mean-of-class prototype learners or are more sophisticated algorithms better candidates? [sent-23, score-0.191]
15 In our experiments, subjects and machines classified human faces according to gender. [sent-24, score-0.694]
16 The stimuli were presented and we collected the subjects’ responses, which are the estimated gender, reaction time and confidence rating (sec. [sent-25, score-0.345]
17 For every subject two personal new datasets were created: the original faces either with the true or with the subject’s labels (true or estimated gender response). [sent-27, score-0.565]
18 We then applied a Principal Component Analysis to a texture and shape representation of the faces. [sent-28, score-0.056]
19 Various algorithms such as Support Vec- tor Machines, Relevance Vector Machines, Prototype and K-means Learners (sec. [sent-29, score-0.03]
20 3) were applied on this low-dimensional dataset with either the true or the subjects’ labels. [sent-30, score-0.06]
21 The resulting classification performances were compared, the corresponding decision hyperplanes were computed and the distances of the faces to the hyperplanes were correlated with the subjects’ responses, the data being pooled among all subjects and stimuli or on a stimulus-by-stimulus basis (sec. [sent-31, score-0.94]
22 2 Human Classification Behaviour We used grey-scale frontal views of human faces taken from the MPI face database [1]. [sent-33, score-0.688]
23 Because of technical inhomogeneities of the faces in the database we post-processed each face such that all faces have same mean intensity, same pixel-surface area and are centred [2]. [sent-34, score-0.72]
24 This processing stage is followed by a slight low-pass filtering of each face in the database in order to eliminate, as much as possible, scanning artifacts. [sent-35, score-0.24]
25 The database is gender-balanced and contains 200 Caucasian faces (see Fig. [sent-36, score-0.345]
26 Twenty-seven human 15 13 i eigenvalue log(λ ) 14 12 11 10 9 8 7 0 20 40 60 80 100 120 140 160 180 200 index of component i Figure 1: Female and male faces from the processed database (left). [sent-38, score-0.628]
27 01 · 103 (the last eigenvalue being 0 is not plotted) and λmax = 2. [sent-41, score-0.051]
28 subjects were asked to classify the faces according to their gender and we recorded three responses: estimated class (i. [sent-43, score-0.845]
29 female/male), reaction time (RT) and, after each estimatedclass-response, a confidence rating (CR) on a scale from 1 (unsure) to 3 (sure). [sent-45, score-0.164]
30 The stimuli were presented sequentially to the subjects on a carefully calibrated display using a modified Hanning window (a raised cosine function with a raising time of ttransient = 500ms and a plateau time of tsteady = 1000ms, for a total presentation time t = 2000ms per face). [sent-46, score-0.522]
31 Subjects were asked to answer as fast as possible to obtain perceptual, rather than cognitive, judgements. [sent-47, score-0.041]
32 Most of the time they responded well before the presentation of the stimulus had ended (mean RT over all stimuli and subjects was approximately 900ms). [sent-48, score-0.492]
33 All subjects had normal or corrected-to-normal vision and were paid for their participation. [sent-49, score-0.315]
34 Analysis of the classification performance of humans is based on signal detection theory [3] and we assume that, on the decision axis, the internal signal and noise distributions are Gaussian with same unit variance but different means. [sent-51, score-0.135]
35 We define correct response probabilities for males (+) and females (−) as P+ = P (ˆ = 1|y = 1) and y P− = P (ˆ = −1|y = −1) where y is the estimated class and y the true class of the stimuy ˆ lus. [sent-52, score-0.212]
36 The discriminability of both stimuli can then be computed as: d = Z(P+ ) + Z(P− ) where Z = Φ−1 , and Φ is the cumulative normal distribution with zero mean and unit variance. [sent-53, score-0.21]
37 This value indicates that the classification task is comparatively easy for the subjects, although without being trivial (no ceiling effect). [sent-57, score-0.06]
38 We observe a strong male bias (a large number of females classified as males but very few males classified as females) and express this bias as: η = Z 2 (P+ ) − Z 2 (P− ) = 3. [sent-58, score-0.28]
39 2 show the correlations of (a) RT and classification error, (b) classification error and CR, and (c) RT and CR. [sent-62, score-0.04]
40 8 0 no error error 1 2 CR 3 1 2 CR 3 Figure 2: Human classification behaviour: mutual dependencies of the subject’s responses. [sent-72, score-0.08]
41 RT’s are longer for incorrect answers than for correct ones (a). [sent-73, score-0.044]
42 Second, a high CR is correlated with a low classification error (b) and thus subjects have veridical knowledge about the difficulty of individual responses—this is certainly not the case in many low-level psychophysical settings. [sent-74, score-0.428]
43 It may thus be concluded that a high error (or equivalently a low CR) implies higher RT’s. [sent-78, score-0.101]
44 This may suggest that patterns difficult to classify need more computation, i. [sent-79, score-0.068]
45 longer processing, by the brain than patterns easy to classify. [sent-81, score-0.116]
46 3 Machine Learning Classifiers In the following, various hyperplane classification algorithms are expressed as weighted dual space learners with different learning rules. [sent-82, score-0.188]
47 Given a dataset {xi , yi }p , we assume i=1 classification is done in the input space, i. [sent-83, score-0.076]
48 The hyperplanes can be written using a weight (or normal) vector w and an offset b in order to yield a classification rule as y(x) = sign( w|x + b) in the first three cases whereas in the last one, the decision rule is a collection of hyperplanes. [sent-87, score-0.168]
49 These classifiers are compared on a two-dimensional toy dataset in Fig. [sent-88, score-0.033]
50 The weight vector is given as: w = i αi yi x i where α is obtained by maximising i αi − 1 ij yi yj αi αj xi |xj subject to i αi yi = 2 0 and 0 ≤ αi ≤ C where C is a regularisation parameter, determined using for instance cross-validation. [sent-91, score-0.213]
51 The offset is computed as: b = yi − w|xi i|0<αi < 5 · 10 −4 which allows us to reject the null hypothesis with a high degree of confidence. [sent-92, score-0.073]
52 02 1 50 100 150 200 |δ| to SH 1 1 50 100 150 200 |δ| to SH 1 1 50 100 150 200 |δ| to SH Figure 5: Scatter plots relating the subjects’ responses (classification error, RT and CR) to the distance |δ| to the SH for each face in the database, the pooling being done across subjects. [sent-117, score-0.24]
53 From these results it can be seen that RVMs correlate best all the subject’s responses with the distances of the stimuli to the SH. [sent-118, score-0.272]
54 The RT seems to be the performance measure where most correlation between man and machine can be asserted although all performance measures are related as shown in sec. [sent-119, score-0.145]
55 The prototype algorithm again behaves in the least human-like manner of the four classifiers. [sent-121, score-0.122]
56 The correlation between the classification behaviour of man and machine indicates for RVMs, and to some extent SVMs, that heads far from the SH are more easily processed by humans. [sent-122, score-0.249]
57 It may be concluded that the brain needs to do more processing (higher RT) to classify stimuli close to the decision hyperplane, while stimuli far from it are classified more accurately (low error) and with higher confidence (high CR). [sent-123, score-0.536]
58 Human classification behaviour can thus be modeled by hyperplane algorithms; a piecewise linear decision function as found in Kmean seems however to be not biologically-plausible. [sent-124, score-0.222]
59 5 Conclusions Our study compared classification of faces by man and machine. [sent-125, score-0.31]
60 Psychophysically we noted that a high classification error and a low CR for humans is accompanied by a longer processing of information by the brain (a longer RT). [sent-126, score-0.328]
61 First, SVMs and RVMs can learn to classify faces using the subjects’ labels but perform much better when using the true labels. [sent-129, score-0.353]
62 Second, correlating the average response of humans (classification error, RT or CR) with the distance to the SH on a face-by-face basis using Spearman’s rank correlation coefficients shows that RVMs recreate human performance most closely in every respect. [sent-130, score-0.303]
63 Third, the mean-of-class prototype, its popularity in neuroscience notwithstanding, is the least human-like classifier in all cases examined. [sent-131, score-0.051]
64 Third, when rejecting the prototype learner as a plausible candidate for human classification we assume the representativeness of our face space: we assume that the mean face of our human subjects’ is close to the sample mean of our database. [sent-133, score-0.688]
65 Clearly, a larger face database would be welcome, but is not trivial as we need texture maps and the corresponding shapes. [sent-134, score-0.328]
66 Machines were trained on the dataset proper, whereas humans were assumed to have extracted the relevant information during their lifetime, and they were tested on faces with some cues removed. [sent-136, score-0.36]
67 However, the representation we used does allow the genders to be separated well, as shown by the SVM classification performance on the true labels. [sent-137, score-0.057]
68 As a first attempt to extend the neuroscience community’s toolbox with machine learning methods we believe to have shown the fruitfulness of this approach. [sent-138, score-0.083]
69 Acknowledgements The authors would like to thank Volker Blanz for providing the face database and the flow¨ field algorithms. [sent-139, score-0.24]
70 In addition we are grateful to Gokhan Bakır, Heinrich B¨ lthoff, Jez Hill, u Carl Rasmussen, Gunnar R¨ tsch, Bernhard Sch¨ lkopf and Vladimir Vapnik for helpful a o comments and suggestions. [sent-140, score-0.026]
71 Nonlinear Component Analysis as a Kero u nel Eigenvalue Problem. [sent-211, score-0.028]
wordName wordTfidf (topN-words)
[('sh', 0.444), ('subjects', 0.283), ('rt', 0.278), ('cr', 0.278), ('classi', 0.232), ('faces', 0.227), ('gender', 0.195), ('stimuli', 0.15), ('human', 0.147), ('kmean', 0.14), ('rvms', 0.14), ('face', 0.122), ('prototype', 0.122), ('hyperplane', 0.119), ('database', 0.118), ('cation', 0.106), ('prot', 0.105), ('hyperplanes', 0.103), ('humans', 0.1), ('responses', 0.092), ('graf', 0.091), ('rating', 0.091), ('man', 0.083), ('rvm', 0.083), ('females', 0.077), ('males', 0.077), ('reaction', 0.073), ('brain', 0.072), ('dence', 0.071), ('learners', 0.069), ('classify', 0.068), ('behaviour', 0.068), ('psychophysical', 0.066), ('concluded', 0.061), ('texture', 0.056), ('blanz', 0.055), ('svm', 0.055), ('subject', 0.054), ('neuroscience', 0.051), ('eigenvalue', 0.051), ('male', 0.049), ('longer', 0.044), ('yi', 0.043), ('cognitive', 0.042), ('con', 0.042), ('asked', 0.041), ('error', 0.04), ('frontal', 0.039), ('correlated', 0.039), ('machines', 0.037), ('processed', 0.036), ('decision', 0.035), ('views', 0.035), ('relevance', 0.034), ('dataset', 0.033), ('trivial', 0.032), ('machine', 0.032), ('normal', 0.032), ('presentation', 0.031), ('labels', 0.031), ('estimated', 0.031), ('tor', 0.03), ('genders', 0.03), ('angeles', 0.03), ('lncs', 0.03), ('lthoff', 0.03), ('mpi', 0.03), ('correlate', 0.03), ('kero', 0.03), ('plateau', 0.03), ('psychophysically', 0.03), ('regularisation', 0.03), ('resonance', 0.03), ('turk', 0.03), ('offset', 0.03), ('svms', 0.03), ('correlation', 0.03), ('visual', 0.029), ('ers', 0.028), ('ceiling', 0.028), ('accompanied', 0.028), ('raised', 0.028), ('ended', 0.028), ('ag', 0.028), ('felix', 0.028), ('discriminability', 0.028), ('eigenfaces', 0.028), ('spemannstra', 0.028), ('bak', 0.028), ('nel', 0.028), ('rejecting', 0.028), ('technological', 0.028), ('tremendous', 0.028), ('true', 0.027), ('smola', 0.027), ('pca', 0.026), ('lkopf', 0.026), ('pooling', 0.026), ('correlating', 0.026), ('centred', 0.026)]
simIndex simValue paperId paperTitle
same-paper 1 1.0000002 95 nips-2003-Insights from Machine Learning Applied to Human Visual Classification
Author: Felix A. Wichmann, Arnulf B. Graf
Abstract: We attempt to understand visual classification in humans using both psychophysical and machine learning techniques. Frontal views of human faces were used for a gender classification task. Human subjects classified the faces and their gender judgment, reaction time and confidence rating were recorded. Several hyperplane learning algorithms were used on the same classification task using the Principal Components of the texture and shape representation of the faces. The classification performance of the learning algorithms was estimated using the face database with the true gender of the faces as labels, and also with the gender estimated by the subjects. We then correlated the human responses to the distance of the stimuli to the separating hyperplane of the learning algorithms. Our results suggest that human classification can be modeled by some hyperplane algorithms in the feature space we used. For classification, the brain needs more processing for stimuli close to that hyperplane than for those further away. 1
2 0.17053135 188 nips-2003-Training fMRI Classifiers to Detect Cognitive States across Multiple Human Subjects
Author: Xuerui Wang, Rebecca Hutchinson, Tom M. Mitchell
Abstract: We consider learning to classify cognitive states of human subjects, based on their brain activity observed via functional Magnetic Resonance Imaging (fMRI). This problem is important because such classifiers constitute “virtual sensors” of hidden cognitive states, which may be useful in cognitive science research and clinical applications. In recent work, Mitchell, et al. [6,7,9] have demonstrated the feasibility of training such classifiers for individual human subjects (e.g., to distinguish whether the subject is reading an ambiguous or unambiguous sentence, or whether they are reading a noun or a verb). Here we extend that line of research, exploring how to train classifiers that can be applied across multiple human subjects, including subjects who were not involved in training the classifier. We describe the design of several machine learning approaches to training multiple-subject classifiers, and report experimental results demonstrating the success of these methods in learning cross-subject classifiers for two different fMRI data sets. 1
Author: G.C. Littlewort, M.S. Bartlett, I.R. Fasel, J. Chenu, T. Kanda, H. Ishiguro, J.R. Movellan
Abstract: Computer animated agents and robots bring a social dimension to human computer interaction and force us to think in new ways about how computers could be used in daily life. Face to face communication is a real-time process operating at a time scale of less than a second. In this paper we present progress on a perceptual primitive to automatically detect frontal faces in the video stream and code them with respect to 7 dimensions in real time: neutral, anger, disgust, fear, joy, sadness, surprise. The face finder employs a cascade of feature detectors trained with boosting techniques [13, 2]. The expression recognizer employs a novel combination of Adaboost and SVM’s. The generalization performance to new subjects for a 7-way forced choice was 93.3% and 97% correct on two publicly available datasets. The outputs of the classifier change smoothly as a function of time, providing a potentially valuable representation to code facial expression dynamics in a fully automatic and unobtrusive manner. The system was deployed and evaluated for measuring spontaneous facial expressions in the field in an application for automatic assessment of human-robot interaction.
4 0.14169259 144 nips-2003-One Microphone Blind Dereverberation Based on Quasi-periodicity of Speech Signals
Author: Tomohiro Nakatani, Masato Miyoshi, Keisuke Kinoshita
Abstract: Speech dereverberation is desirable with a view to achieving, for example, robust speech recognition in the real world. However, it is still a challenging problem, especially when using a single microphone. Although blind equalization techniques have been exploited, they cannot deal with speech signals appropriately because their assumptions are not satisfied by speech signals. We propose a new dereverberation principle based on an inherent property of speech signals, namely quasi-periodicity. The present methods learn the dereverberation filter from a lot of speech data with no prior knowledge of the data, and can achieve high quality speech dereverberation especially when the reverberation time is long. 1
5 0.12423442 161 nips-2003-Probabilistic Inference in Human Sensorimotor Processing
Author: Konrad P. Körding, Daniel M. Wolpert
Abstract: When we learn a new motor skill, we have to contend with both the variability inherent in our sensors and the task. The sensory uncertainty can be reduced by using information about the distribution of previously experienced tasks. Here we impose a distribution on a novel sensorimotor task and manipulate the variability of the sensory feedback. We show that subjects internally represent both the distribution of the task as well as their sensory uncertainty. Moreover, they combine these two sources of information in a way that is qualitatively predicted by optimal Bayesian processing. We further analyze if the subjects can represent multimodal distributions such as mixtures of Gaussians. The results show that the CNS employs probabilistic models during sensorimotor learning even when the priors are multimodal.
6 0.11479385 106 nips-2003-Learning Non-Rigid 3D Shape from 2D Motion
7 0.099001445 14 nips-2003-A Nonlinear Predictive State Representation
8 0.096675463 109 nips-2003-Learning a Rare Event Detection Cascade by Direct Feature Selection
9 0.089818329 90 nips-2003-Increase Information Transfer Rates in BCI by CSP Extension to Multi-class
10 0.089058116 124 nips-2003-Max-Margin Markov Networks
11 0.08549393 53 nips-2003-Discriminating Deformable Shape Classes
12 0.084425285 143 nips-2003-On the Dynamics of Boosting
13 0.081582077 160 nips-2003-Prediction on Spike Data Using Kernel Algorithms
14 0.079687521 64 nips-2003-Estimating Internal Variables and Paramters of a Learning Agent by a Particle Filter
15 0.079183593 9 nips-2003-A Kullback-Leibler Divergence Based Kernel for SVM Classification in Multimedia Applications
16 0.072222702 91 nips-2003-Inferring State Sequences for Non-linear Systems with Embedded Hidden Markov Models
17 0.070448138 113 nips-2003-Learning with Local and Global Consistency
18 0.070094042 28 nips-2003-Application of SVMs for Colour Classification and Collision Detection with AIBO Robots
19 0.067452848 1 nips-2003-1-norm Support Vector Machines
20 0.067351274 120 nips-2003-Locality Preserving Projections
topicId topicWeight
[(0, -0.198), (1, -0.043), (2, 0.076), (3, -0.229), (4, -0.082), (5, -0.071), (6, 0.064), (7, -0.032), (8, -0.028), (9, 0.095), (10, 0.093), (11, 0.0), (12, 0.103), (13, -0.077), (14, 0.065), (15, 0.117), (16, 0.049), (17, 0.062), (18, 0.081), (19, -0.173), (20, 0.168), (21, 0.114), (22, 0.073), (23, -0.218), (24, 0.018), (25, 0.166), (26, 0.059), (27, -0.145), (28, 0.121), (29, 0.019), (30, -0.042), (31, 0.034), (32, -0.039), (33, 0.031), (34, 0.076), (35, -0.066), (36, 0.113), (37, 0.019), (38, -0.085), (39, 0.096), (40, 0.014), (41, 0.134), (42, -0.086), (43, -0.08), (44, -0.022), (45, 0.099), (46, 0.104), (47, -0.01), (48, 0.028), (49, 0.082)]
simIndex simValue paperId paperTitle
same-paper 1 0.94960386 95 nips-2003-Insights from Machine Learning Applied to Human Visual Classification
Author: Felix A. Wichmann, Arnulf B. Graf
Abstract: We attempt to understand visual classification in humans using both psychophysical and machine learning techniques. Frontal views of human faces were used for a gender classification task. Human subjects classified the faces and their gender judgment, reaction time and confidence rating were recorded. Several hyperplane learning algorithms were used on the same classification task using the Principal Components of the texture and shape representation of the faces. The classification performance of the learning algorithms was estimated using the face database with the true gender of the faces as labels, and also with the gender estimated by the subjects. We then correlated the human responses to the distance of the stimuli to the separating hyperplane of the learning algorithms. Our results suggest that human classification can be modeled by some hyperplane algorithms in the feature space we used. For classification, the brain needs more processing for stimuli close to that hyperplane than for those further away. 1
2 0.73777699 188 nips-2003-Training fMRI Classifiers to Detect Cognitive States across Multiple Human Subjects
Author: Xuerui Wang, Rebecca Hutchinson, Tom M. Mitchell
Abstract: We consider learning to classify cognitive states of human subjects, based on their brain activity observed via functional Magnetic Resonance Imaging (fMRI). This problem is important because such classifiers constitute “virtual sensors” of hidden cognitive states, which may be useful in cognitive science research and clinical applications. In recent work, Mitchell, et al. [6,7,9] have demonstrated the feasibility of training such classifiers for individual human subjects (e.g., to distinguish whether the subject is reading an ambiguous or unambiguous sentence, or whether they are reading a noun or a verb). Here we extend that line of research, exploring how to train classifiers that can be applied across multiple human subjects, including subjects who were not involved in training the classifier. We describe the design of several machine learning approaches to training multiple-subject classifiers, and report experimental results demonstrating the success of these methods in learning cross-subject classifiers for two different fMRI data sets. 1
3 0.60046244 90 nips-2003-Increase Information Transfer Rates in BCI by CSP Extension to Multi-class
Author: Guido Dornhege, Benjamin Blankertz, Gabriel Curio, Klaus-Robert Müller
Abstract: Brain-Computer Interfaces (BCI) are an interesting emerging technology that is driven by the motivation to develop an effective communication interface translating human intentions into a control signal for devices like computers or neuroprostheses. If this can be done bypassing the usual human output pathways like peripheral nerves and muscles it can ultimately become a valuable tool for paralyzed patients. Most activity in BCI research is devoted to finding suitable features and algorithms to increase information transfer rates (ITRs). The present paper studies the implications of using more classes, e.g., left vs. right hand vs. foot, for operating a BCI. We contribute by (1) a theoretical study showing under some mild assumptions that it is practically not useful to employ more than three or four classes, (2) two extensions of the common spatial pattern (CSP) algorithm, one interestingly based on simultaneous diagonalization, and (3) controlled EEG experiments that underline our theoretical findings and show excellent improved ITRs. 1
Author: G.C. Littlewort, M.S. Bartlett, I.R. Fasel, J. Chenu, T. Kanda, H. Ishiguro, J.R. Movellan
Abstract: Computer animated agents and robots bring a social dimension to human computer interaction and force us to think in new ways about how computers could be used in daily life. Face to face communication is a real-time process operating at a time scale of less than a second. In this paper we present progress on a perceptual primitive to automatically detect frontal faces in the video stream and code them with respect to 7 dimensions in real time: neutral, anger, disgust, fear, joy, sadness, surprise. The face finder employs a cascade of feature detectors trained with boosting techniques [13, 2]. The expression recognizer employs a novel combination of Adaboost and SVM’s. The generalization performance to new subjects for a 7-way forced choice was 93.3% and 97% correct on two publicly available datasets. The outputs of the classifier change smoothly as a function of time, providing a potentially valuable representation to code facial expression dynamics in a fully automatic and unobtrusive manner. The system was deployed and evaluated for measuring spontaneous facial expressions in the field in an application for automatic assessment of human-robot interaction.
5 0.54187846 14 nips-2003-A Nonlinear Predictive State Representation
Author: Matthew R. Rudary, Satinder P. Singh
Abstract: Predictive state representations (PSRs) use predictions of a set of tests to represent the state of controlled dynamical systems. One reason why this representation is exciting as an alternative to partially observable Markov decision processes (POMDPs) is that PSR models of dynamical systems may be much more compact than POMDP models. Empirical work on PSRs to date has focused on linear PSRs, which have not allowed for compression relative to POMDPs. We introduce a new notion of tests which allows us to define a new type of PSR that is nonlinear in general and allows for exponential compression in some deterministic dynamical systems. These new tests, called e-tests, are related to the tests used by Rivest and Schapire [1] in their work with the diversity representation, but our PSR avoids some of the pitfalls of their representation—in particular, its potential to be exponentially larger than the equivalent POMDP. 1
6 0.49608776 161 nips-2003-Probabilistic Inference in Human Sensorimotor Processing
7 0.47315082 53 nips-2003-Discriminating Deformable Shape Classes
8 0.45745072 28 nips-2003-Application of SVMs for Colour Classification and Collision Detection with AIBO Robots
9 0.45298451 181 nips-2003-Statistical Debugging of Sampled Programs
10 0.45156717 144 nips-2003-One Microphone Blind Dereverberation Based on Quasi-periodicity of Speech Signals
11 0.42504489 3 nips-2003-AUC Optimization vs. Error Rate Minimization
12 0.38770705 109 nips-2003-Learning a Rare Event Detection Cascade by Direct Feature Selection
13 0.38464677 178 nips-2003-Sparse Greedy Minimax Probability Machine Classification
14 0.3676984 124 nips-2003-Max-Margin Markov Networks
15 0.35772377 9 nips-2003-A Kullback-Leibler Divergence Based Kernel for SVM Classification in Multimedia Applications
16 0.32964784 52 nips-2003-Different Cortico-Basal Ganglia Loops Specialize in Reward Prediction at Different Time Scales
17 0.32804352 106 nips-2003-Learning Non-Rigid 3D Shape from 2D Motion
18 0.3251003 118 nips-2003-Link Prediction in Relational Data
19 0.3219485 89 nips-2003-Impact of an Energy Normalization Transform on the Performance of the LF-ASD Brain Computer Interface
20 0.31546515 130 nips-2003-Model Uncertainty in Classical Conditioning
topicId topicWeight
[(0, 0.021), (5, 0.01), (11, 0.024), (35, 0.015), (53, 0.068), (71, 0.034), (76, 0.033), (85, 0.638), (91, 0.067)]
simIndex simValue paperId paperTitle
same-paper 1 0.98377055 95 nips-2003-Insights from Machine Learning Applied to Human Visual Classification
Author: Felix A. Wichmann, Arnulf B. Graf
Abstract: We attempt to understand visual classification in humans using both psychophysical and machine learning techniques. Frontal views of human faces were used for a gender classification task. Human subjects classified the faces and their gender judgment, reaction time and confidence rating were recorded. Several hyperplane learning algorithms were used on the same classification task using the Principal Components of the texture and shape representation of the faces. The classification performance of the learning algorithms was estimated using the face database with the true gender of the faces as labels, and also with the gender estimated by the subjects. We then correlated the human responses to the distance of the stimuli to the separating hyperplane of the learning algorithms. Our results suggest that human classification can be modeled by some hyperplane algorithms in the feature space we used. For classification, the brain needs more processing for stimuli close to that hyperplane than for those further away. 1
2 0.96968997 193 nips-2003-Variational Linear Response
Author: Manfred Opper, Ole Winther
Abstract: A general linear response method for deriving improved estimates of correlations in the variational Bayes framework is presented. Three applications are given and it is discussed how to use linear response as a general principle for improving mean field approximations.
3 0.96390671 136 nips-2003-New Algorithms for Efficient High Dimensional Non-parametric Classification
Author: Ting liu, Andrew W. Moore, Alexander Gray
Abstract: This paper is about non-approximate acceleration of high dimensional nonparametric operations such as k nearest neighbor classifiers and the prediction phase of Support Vector Machine classifiers. We attempt to exploit the fact that even if we want exact answers to nonparametric queries, we usually do not need to explicitly find the datapoints close to the query, but merely need to ask questions about the properties about that set of datapoints. This offers a small amount of computational leeway, and we investigate how much that leeway can be exploited. For clarity, this paper concentrates on pure k-NN classification and the prediction phase of SVMs. We introduce new ball tree algorithms that on real-world datasets give accelerations of 2-fold up to 100-fold compared against highly optimized traditional ball-tree-based k-NN. These results include datasets with up to 106 dimensions and 105 records, and show non-trivial speedups while giving exact answers. 1
4 0.92052144 2 nips-2003-ARA*: Anytime A* with Provable Bounds on Sub-Optimality
Author: Maxim Likhachev, Geoffrey J. Gordon, Sebastian Thrun
Abstract: In real world planning problems, time for deliberation is often limited. Anytime planners are well suited for these problems: they find a feasible solution quickly and then continually work on improving it until time runs out. In this paper we propose an anytime heuristic search, ARA*, which tunes its performance bound based on available search time. It starts by finding a suboptimal solution quickly using a loose bound, then tightens the bound progressively as time allows. Given enough time it finds a provably optimal solution. While improving its bound, ARA* reuses previous search efforts and, as a result, is significantly more efficient than other anytime search methods. In addition to our theoretical analysis, we demonstrate the practical utility of ARA* with experiments on a simulated robot kinematic arm and a dynamic path planning problem for an outdoor rover. 1
5 0.87885606 64 nips-2003-Estimating Internal Variables and Paramters of a Learning Agent by a Particle Filter
Author: Kazuyuki Samejima, Kenji Doya, Yasumasa Ueda, Minoru Kimura
Abstract: When we model a higher order functions, such as learning and memory, we face a difficulty of comparing neural activities with hidden variables that depend on the history of sensory and motor signals and the dynamics of the network. Here, we propose novel method for estimating hidden variables of a learning agent, such as connection weights from sequences of observable variables. Bayesian estimation is a method to estimate the posterior probability of hidden variables from observable data sequence using a dynamic model of hidden and observable variables. In this paper, we apply particle filter for estimating internal parameters and metaparameters of a reinforcement learning model. We verified the effectiveness of the method using both artificial data and real animal behavioral data. 1
6 0.82410991 124 nips-2003-Max-Margin Markov Networks
7 0.77305812 192 nips-2003-Using the Forest to See the Trees: A Graphical Model Relating Features, Objects, and Scenes
8 0.72682005 28 nips-2003-Application of SVMs for Colour Classification and Collision Detection with AIBO Robots
9 0.68621957 3 nips-2003-AUC Optimization vs. Error Rate Minimization
10 0.68113631 134 nips-2003-Near-Minimax Optimal Classification with Dyadic Classification Trees
11 0.64662576 109 nips-2003-Learning a Rare Event Detection Cascade by Direct Feature Selection
12 0.620906 50 nips-2003-Denoising and Untangling Graphs Using Degree Priors
13 0.61273038 20 nips-2003-All learning is Local: Multi-agent Learning in Global Reward Games
14 0.60673332 57 nips-2003-Dynamical Modeling with Kernels for Nonlinear Time Series Prediction
15 0.59258461 52 nips-2003-Different Cortico-Basal Ganglia Loops Specialize in Reward Prediction at Different Time Scales
16 0.58750838 148 nips-2003-Online Passive-Aggressive Algorithms
17 0.58503181 147 nips-2003-Online Learning via Global Feedback for Phrase Recognition
18 0.58469146 188 nips-2003-Training fMRI Classifiers to Detect Cognitive States across Multiple Human Subjects
19 0.57923716 41 nips-2003-Boosting versus Covering
20 0.57178265 101 nips-2003-Large Margin Classifiers: Convex Loss, Low Noise, and Convergence Rates