nips nips2010 nips2010-103 knowledge-graph by maker-knowledge-mining

103 nips-2010-Generating more realistic images using gated MRF's

Source: pdf

Author: Marc'aurelio Ranzato, Volodymyr Mnih, Geoffrey E. Hinton

Abstract: Probabilistic models of natural images are usually evaluated by measuring performance on rather indirect tasks, such as denoising and inpainting. A more direct way to evaluate a generative model is to draw samples from it and to check whether statistical properties of the samples match the statistics of natural images. This method is seldom used with high-resolution images, because current models produce samples that are very different from natural images, as assessed by even simple visual inspection. We investigate the reasons for this failure and we show that by augmenting existing models so that there are two sets of latent variables, one set modelling pixel intensities and the other set modelling image-speciﬁc pixel covariances, we are able to generate high-resolution images that look much more realistic than before. The overall model can be interpreted as a gated MRF where both pair-wise dependencies and mean intensities of pixels are modulated by the states of latent variables. Finally, we conﬁrm that if we disallow weight-sharing between receptive ﬁelds that overlap each other, the gated MRF learns more efﬁcient internal representations, as demonstrated in several recognition tasks. 1 Introduction and Prior Work The study of the statistical properties of natural images has a long history and has inﬂuenced many ﬁelds, from image processing to computational neuroscience [1]. In this work we focus on probabilistic models of natural images. These models are useful for extracting representations [2, 3, 4] that can be used for discriminative tasks and they can also provide adaptive priors [5, 6, 7] that can be used in applications like denoising and inpainting. Our main focus, however, will be on improving the quality of the generative model, rather than exploring its possible applications. Markov Random Fields (MRF’s) provide a very general framework for modelling natural images. In an MRF, an image is assigned a probability which is a normalized product of potential functions, with each function typically being deﬁned over a subset of the observed variables. In this work we consider a very versatile class of MRF’s in which potential functions are deﬁned over both pixels and latent variables, thus allowing the states of the latent variables to modulate or gate the effective interactions between the pixels. This type of MRF, that we dub gated MRF, was proposed as an image model by Geman and Geman [8]. Welling et al. [9] showed how an MRF in this family1 could be learned for small image patches and their work was extended to high-resolution images by Roth and Black [6] who also demonstrated its success in some practical applications [7]. Besides their practical use, these models were speciﬁcally designed to match the statistical properties of natural images, and therefore, it seems natural to evaluate them in those terms. Indeed, several authors [10, 7] have proposed that these models should be evaluated by generating images and 1 Product of Student’s t models (without pooling) may not appear to have latent variables but each potential can be viewed as an inﬁnite mixture of zero-mean Gaussians where the inverse variance of the Gaussian is the latent variable. 1 checking whether the samples match the statistical properties observed in natural images. It is, therefore, very troublesome that none of the existing models can generate good samples, especially for high-resolution images (see for instance ﬁg. 2 in [7] which is one of the best models of highresolution images reported in the literature so far). In fact, as our experiments demonstrate the generated samples from these models are more similar to random images than to natural images! When MRF’s with gated interactions are applied to small image patches, they actually seem to work moderately well, as demonstrated by several authors [11, 12, 13]. The generated patches have some coherent and elongated structure and, like natural image patches, they are predominantly very smooth with sudden outbreaks of strong structure. This is unsurprising because these models have a built-in assumption that images are very smooth with occasional strong violations of smoothness [8, 14, 15]. However, the extension of these patch-based models to high-resolution images by replicating ﬁlters across the image has proven to be difﬁcult. The receptive ﬁelds that are learned no longer resemble Gabor wavelets but look random [6, 16] and the generated images lack any of the long range structure that is so typical of natural images [7]. The success of these methods in applications such as denoising is a poor measure of the quality of the generative model that has been learned: Setting the parameters to random values works almost as well for eliminating independent Gaussian noise [17], because this can be done quite well by just using a penalty for high-frequency variation. In this work, we show that the generative quality of these models can be drastically improved by jointly modelling both pixel mean intensities and pixel covariances. This can be achieved by using two sets of latent variables, one that gates pair-wise interactions between pixels and another one that sets the mean intensities of pixels, as we already proposed in some earlier work [4]. Here, we show that this modelling choice is crucial to make the gated MRF work well on high-resolution images. Finally, we show that the most widely used method of sharing weights in MRF’s for high-resolution images is overly constrained. Earlier work considered homogeneous MRF’s in which each potential is replicated at all image locations. This has the subtle effect of making learning very difﬁcult because of strong correlations at nearby sites. Following Gregor and LeCun [18] and also Tang and Eliasmith [19], we keep the number of parameters under control by using local potentials, but unlike Roth and Black [6] we only share weights between potentials that do not overlap. 2 Augmenting Gated MRF’s with Mean Hidden Units A Product of Student’s t (PoT) model [15] is a gated MRF deﬁned on small image patches that can be viewed as modelling image-speciﬁc, pair-wise relationships between pixel values by using the states of its latent variables. It is very good at representing the fact that two-pixel have very similar intensities and no good at all at modelling what these intensities are. Failure to model the mean also leads to impoverished modelling of the covariances when the input images have nonzero mean intensity. The covariance RBM (cRBM) [20] is another model that shares the same limitation since it only differs from PoT in the distribution of its latent variables: The posterior over the latent variables is a product of Bernoulli distributions instead of Gamma distributions as in PoT. We explain the fundamental limitation of these models by using a simple toy example: Modelling two-pixel images using a cRBM with only one binary hidden unit, see ﬁg. 1. This cRBM assumes that the conditional distribution over the input is a zero-mean Gaussian with a covariance that is determined by the state of the latent variable. Since the latent variable is binary, the cRBM can be viewed as a mixture of two zero-mean full covariance Gaussians. The latent variable uses the pairwise relationship between pixels to decide which of the two covariance matrices should be used to model each image. When the input data is pre-proessed by making each image have zero mean intensity (the empirical histogram is shown in the ﬁrst row and ﬁrst column), most images lie near the origin because most of the times nearby pixels are strongly correlated. Less frequently we encounter edge images that exhibit strong anti-correlation between the pixels, as shown by the long tails along the anti-diagonal line. A cRBM could model this data by using two Gaussians (ﬁrst row and second column): one that is spherical and tight at the origin for smooth images and another one that has a covariance elongated along the anti-diagonal for structured images. If, however, the whole set of images is normalized by subtracting from every pixel the mean value of all pixels over all images (second row and ﬁrst column), the cRBM fails at modelling structured images (second row and second column). It can ﬁt a Gaussian to the smooth images by discovering 2 Figure 1: In the ﬁrst row, each image is zero mean. In the second row, the whole set of data points is centered but each image can have non-zero mean. The ﬁrst column shows 8x8 images picked at random from natural images. The images in the second column are generated by a model that does not account for mean intensity. The images in the third column are generated by a model that has both “mean” and “covariance” hidden units. The contours in the ﬁrst column show the negative log of the empirical distribution of (tiny) natural two-pixel images (x-axis being the ﬁrst pixel and the y-axis the second pixel). The plots in the other columns are toy examples showing how each model could represent the empirical distribution using a mixture of Gaussians with components that have one of two possible covariances (corresponding to the state of a binary “covariance” latent variable). Models that can change the means of the Gaussians (mPoT and mcRBM) can represent better structured images (edge images lie along the anti-diagonal and are ﬁtted by the Gaussians shown in red) while the other models (PoT and cRBM) fail, overall when each image can have non-zero mean. the direction of strong correlation along the main diagonal, but it is very likely to fail to discover the direction of anti-correlation, which is crucial to represent discontinuities, because structured images with different mean intensity appear to be evenly spread over the whole input space. If the model has another set of latent variables that can change the means of the Gaussian distributions in the mixture (as explained more formally below and yielding the mPoT and mcRBM models), then the model can represent both changes of mean intensity and the correlational structure of pixels (see last column). The mean latent variables effectively subtract off the relevant mean from each data-point, letting the covariance latent variable capture the covariance structure of the data. As before, the covariance latent variable needs only to select between two covariance matrices. In fact, experiments on real 8x8 image patches conﬁrm these conjectures. Fig. 1 shows samples drawn from PoT and mPoT. mPoT (and similarly mcRBM [4]) is not only better at modelling zero mean images but it can also represent images that have non zero mean intensity well. We now describe mPoT, referring the reader to [4] for a detailed description of mcRBM. In PoT [9] the energy function is: E PoT (x, hc ) = i 1 [hc (1 + (Ci T x)2 ) + (1 − γ) log hc ] i i 2 (1) where x is a vectorized image patch, hc is a vector of Gamma “covariance” latent variables, C is a ﬁlter bank matrix and γ is a scalar parameter. The joint probability over input pixels and latent variables is proportional to exp(−E PoT (x, hc )). Therefore, the conditional distribution over the input pixels is a zero-mean Gaussian with covariance equal to: Σc = (Cdiag(hc )C T )−1 . (2) In order to make the mean of the conditional distribution non-zero, we deﬁne mPoT as the normalized product of the above zero-mean Gaussian that models the covariance and a spherical covariance Gaussian that models the mean. The overall energy function becomes: E mPoT (x, hc , hm ) = E PoT (x, hc ) + E m (x, hm ) 3 (3) Figure 2: Illustration of different choices of weight-sharing scheme for a RBM. Links converging to one latent variable are ﬁlters. Filters with the same color share the same parameters. Kinds of weight-sharing scheme: A) Global, B) Local, C) TConv and D) Conv. E) TConv applied to an image. Cells correspond to neighborhoods to which ﬁlters are applied. Cells with the same color share the same parameters. F) 256 ﬁlters learned by a Gaussian RBM with TConv weight-sharing scheme on high-resolution natural images. Each ﬁlter has size 16x16 pixels and it is applied every 16 pixels in both the horizontal and vertical directions. Filters in position (i, j) and (1, 1) are applied to neighborhoods that are (i, j) pixels away form each other. Best viewed in color. where hm is another set of latent variables that are assumed to be Bernoulli distributed (but other distributions could be used). The new energy term is: E m (x, hm ) = 1 T x x− 2 hm Wj T x j (4) j yielding the following conditional distribution over the input pixels: p(x|hc , hm ) = N (Σ(W hm ), Σ), Σ = (Σc + I)−1 (5) with Σc deﬁned in eq. 2. As desired, the conditional distribution has non-zero mean2 . Patch-based models like PoT have been extended to high-resolution images by using spatially localized ﬁlters [6]. While we can subtract off the mean intensity from independent image patches to successfully train PoT, we cannot do that on a high-resolution image because overlapping patches might have different mean. Unfortunately, replicating potentials over the image ignoring variations of mean intensity has been the leading strategy to date [6]3 . This is the major reason why generation of high-resolution images is so poor. Sec. 4 shows that generation can be drastically improved by explicitly accounting for variations of mean intensity, as performed by mPoT and mcRBM. 3 Weight-Sharing Schemes By integrating out the latent variables, we can write the density function of any gated MRF as a normalized product of potential functions (for mPoT refer to eq. 6). In this section we investigate different ways of constraining the parameters of the potentials of a generic MRF. Global: The obvious way to extend a patch-based model like PoT to high-resolution images is to deﬁne potentials over the whole image; we call this scheme global. This is not practical because 1) the number of parameters grows about quadratically with the size of the image making training too slow, 2) we do not need to model interactions between very distant pairs of pixels since their dependence is negligible, and 3) we would not be able to use the model on images of different size. Conv: The most popular way to handle big images is to deﬁne potentials on small subsets of variables (e.g., neighborhoods of size 5x5 pixels) and to replicate these potentials across space while 2 The need to model the means was clearly recognized in [21] but they used conjunctive latent features that simultaneously represented a contribution to the “precision matrix” in a speciﬁc direction and the mean along that same direction. 3 The success of PoT-like models in Bayesian denoising is not surprising since the noisy image effectively replaces the reconstruction term from the mean hidden units (see eq. 5), providing a set of noisy mean intensities that are cleaned up by the patterns of correlation enforced by the covariance latent variables. 4 sharing their parameters at each image location [23, 24, 6]. This yields a convolutional weightsharing scheme, also called homogeneous ﬁeld in the statistics literature. This choice is justiﬁed by the stationarity of natural images. This weight-sharing scheme is extremely concise in terms of number of parameters, but also rather inefﬁcient in terms of latent representation. First, if there are N ﬁlters at each location and these ﬁlters are stepped by one pixel then the internal representation is about N times overcomplete. The internal representation has not only high computational cost, but it is also highly redundant. Since the input is mostly smooth and the parameters are the same across space, the latent variables are strongly correlated as well. This inefﬁciency turns out to be particularly harmful for a model like PoT causing the learned ﬁlters to become “random” looking (see ﬁg 3-iii). A simple intuition follows from the equivalence between PoT and square ICA [15]. If the ﬁlter matrix C of eq. 1 is square and invertible, we can marginalize out the latent variables and write: p(y) = i S(yi ), where yi = Ci T x and S is a Student’s t distribution. In other words, there is an underlying assumption that ﬁlter outputs are independent. However, if the ﬁlters of matrix C are shifted and overlapping versions of each other, this clearly cannot be true. Training PoT with the Conv weight-sharing scheme forces the model to ﬁnd ﬁlters that make ﬁlter outputs as independent as possible, which explains the very high-frequency patterns that are usually discovered [6]. Local: The Global and Conv weight-sharing schemes are at the two extremes of a spectrum of possibilities. For instance, we can deﬁne potentials on a small subset of input variables but, unlike Conv, each potential can have its own set of parameters, as shown in ﬁg. 2-B. This is called local, or inhomogeneous ﬁeld. Compared to Conv the number of parameters increases only slightly but the number of latent variables required and their redundancy is greatly reduced. In fact, the model learns different receptive ﬁelds at different locations as a better strategy for representing the input, overall when the number of potentials is limited (see also ﬁg. 2-F). TConv: Local would not allow the model to be trained and tested on images of different resolution, and it might seem wasteful not to exploit the translation invariant property of images. We therefore advocate the use of a weight-sharing scheme that we call tiled-convolutional (TConv) shown in ﬁg. 2-C and E [18]. Each ﬁlter tiles the image without overlaps with copies of itself (i.e. the stride equals the ﬁlter diameter). This reduces spatial redundancy of latent variables and allows the input images to have arbitrary size. At the same time, different ﬁlters do overlap with each other in order to avoid tiling artifacts. Fig. 2-F shows ﬁlters that were (jointly) learned by a Restricted Boltzmann Machine (RBM) [29] with Gaussian input variables using the TConv weight-sharing scheme. 4 Experiments We train gated MRF’s with and without mean hidden units using different weight-sharing schemes. The training procedure is very similar in all cases. We perform approximate maximum likelihood by using Fast Persistence Contrastive Divergence (FPCD) [25] and we draw samples by using Hybrid Monte Carlo (HMC) [26]. Since all latent variables can be exactly marginalized out we can use HMC on the free energy (negative logarithm of the marginal distribution over the input pixels). For mPoT this is: F mPoT (x) = − log(p(x))+const. = k,i 1 1 γ log(1+ (Cik T xk )2 )+ xT x− 2 2 T log(1+exp(Wjk xk )) (6) k,j where the index k runs over spatial locations and xk is the k-th image patch. FPCD keeps samples, called negative particles, that it uses to represent the model distribution. These particles are all updated after each weight update. For each mini-batch of data-points a) we compute the derivative of the free energy w.r.t. the training samples, b) we update the negative particles by running HMC for one HMC step consisting of 20 leapfrog steps. We start at the previous set of negative particles and use as parameters the sum of the regular parameters and a small perturbation vector, c) we compute the derivative of the free energy at the negative particles, and d) we update the regular parameters by using the difference of gradients between step a) and c) while the perturbation vector is updated using the gradient from c) only. The perturbation is also strongly decayed to zero and is subject to a larger learning rate. The aim is to encourage the negative particles to explore the space more quickly by slightly and temporarily raising the energy at their current position. Note that the use of FPCD as opposed to other estimation methods (like Persistent Contrastive Divergence [27]) turns out to be crucial to achieve good mixing of the sampler even after training. We train on mini-batches of 32 samples using gray-scale images of approximate size 160x160 pixels randomly cropped from the Berkeley segmentation dataset [28]. We perform 160,000 weight updates decreasing the learning by a factor of 4 by the end of training. The initial learning rate is set to 0.1 for the covariance 5 Figure 3: 160x160 samples drawn by A) mPoT-TConv, B) mHPoT-TConv, C) mcRBM-TConv and D) PoTTConv. On the side also i) a subset of 8x8 “covariance” ﬁlters learned by mPoT-TConv (the plot below shows how the whole set of ﬁlters tile a small patch; each bar correspond to a Gabor ﬁt of a ﬁlter and colors identify ﬁlters applied at the same 8x8 location, each group is shifted by 2 pixels down the diagonal and a high-resolution image is tiled by replicating this pattern every 8 pixels horizontally and vertically), ii) a subset of 8x8 “mean” ﬁlters learned by the same mPoT-TConv, iii) ﬁlters learned by PoT-Conv and iv) by PoT-TConv. ﬁlters (matrix C of eq. 1), 0.01 for the mean parameters (matrix W of eq. 4), and 0.001 for the other parameters (γ of eq. 1). During training we condition on the borders and initialize the negative particles at zero in order to avoid artifacts at the border of the image. We learn 8x8 ﬁlters and pre-multiply the covariance ﬁlters by a whitening transform retaining 99% of the variance; we also normalize the norm of the covariance ﬁlters to prevent some of them from decaying to zero during training4 . Whenever we use the TConv weight-sharing scheme the model learns covariance ﬁlters that mostly resemble localized and oriented Gabor functions (see ﬁg. 3-i and iv), while the Conv weight-sharing scheme learns structured but poorly localized high-frequency patterns (see ﬁg. 3-iii) [6]. The TConv models re-use the same 8x8 ﬁlters every 8 pixels and apply a diagonal offset of 2 pixels between neighboring ﬁlters with different weights in order to reduce tiling artifacts. There are 4 sets of ﬁlters, each with 64 ﬁlters for a total of 256 covariance ﬁlters (see bottom plot of ﬁg. 3). Similarly, we have 4 sets of mean ﬁlters, each with 32 ﬁlters. These ﬁlters have usually non-zero mean and exhibit on-center off-surround and off-center on-surround patterns, see ﬁg. 3-ii. In order to draw samples from the learned models, we run HMC for a long time (10,000 iterations, each composed of 20 leap-frog steps). Some samples of size 160x160 pixels are reported in ﬁg. 3 A)D). Without modelling the mean intensity, samples lack structure and do not seem much different from those that would be generated by a simple Gaussian model merely ﬁtting the second order statistics (see ﬁg. 3 in [1] and also ﬁg. 2 in [7]). By contrast, structure, sharp boundaries and some simple texture emerge only from models that have mean latent variables, namely mcRBM, mPoT and mHPoT which differs from mPoT by having a second layer pooling matrix on the squared covariance ﬁlter outputs [11]. A more quantitative comparison is reported in table 1. We ﬁrst compute marginal statistics of ﬁlter responses using the generated images, natural images from the test set, and random images. The statistics are the normalized histogram of individual ﬁlter responses to 24 Gabor ﬁlters (8 orientations and 3 scales). We then calculate the KL divergence between the histograms on random images and generated images and the KL divergence between the histograms on natural images and generated images. The table also reports the average difference of energies between random images and natural images. All results demonstrate that models that account for mean intensity generate images 4 The code used in the experiments can be found at the ﬁrst author’s web-page. 6 MODEL F (R) − F (T ) (104 ) KL(R G) KL(T G) KL(R G) − KL(T PoT - Conv 2.9 0.3 0.6 PoT - TConv 2.8 0.4 1.0 -0.6 mPoT - TConv 5.2 1.0 0.2 0.8 mHPoT - TConv 4.9 1.7 0.8 0.9 mcRBM - TConv 3.5 1.5 1.0 G) -0.3 0.5 Table 1: Comparing MRF’s by measuring: difference of energy (negative log ratio of probabilities) between random images (R) and test natural images (T), the KL divergence between statistics of random images (R) and generated images (G), KL divergence between statistics of test natural images (T) and generated images (G), and difference of these two KL divergences. Statistics are computed using 24 Gabor ﬁlters. that are closer to natural images than to random images, whereas models that do not account for the mean (like the widely used PoT-Conv) produce samples that are actually closer to random images. 4.1 Discriminative Experiments on Weight-Sharing Schemes In future work, we intend to use the features discovered by the generative model for recognition. To understand how the different weight sharing schemes affect recognition performance we have done preliminary tests using the discriminative performance of a simpler model on simpler data. We consider one of the simplest and most versatile models, namely the RBM [29]. Since we also aim to test the Global weight-sharing scheme we are constrained to using fairly low resolution datasets such as the MNIST dataset of handwritten digits [30] and the CIFAR 10 dataset of generic object categories [22]. The MNIST dataset has soft binary images of size 28x28 pixels, while the CIFAR 10 dataset has color images of size 32x32 pixels. CIFAR 10 has 10 classes, 5000 training samples per class and 1000 test samples per class. MNIST also has 10 classes with, on average, 6000 training samples per class and 1000 test samples per class. The energy function of the RBM trained on the CIFAR 10 dataset, modelling input pixels with 3 (R,G,B) Gaussian variables [31], is exactly the one shown in eq. 4; while the RBM trained on MNIST uses logistic units for the pixels and the energy function is again the same as before but without any quadratic term. All models are trained in an unsupervised way to approximately maximize the likelihood in the training set using Contrastive Divergence [32]. They are then used to represent each input image with a feature vector (mean of the posterior over the latent variables) which is fed to a multinomial logistic classiﬁer for discrimination. Models are compared in terms of: 1) recognition accuracy, 2) convergence time and 3) dimensionality of the representation. In general, assuming ﬁlters much smaller than the input image and assuming equal number of latent variables, Conv, TConv and Local models process each sample faster than Global by a factor approximately equal to the ratio between the area of the image and the area of the ﬁlters, which can be very large in practice. In the ﬁrst set of experiments reported on the left of ﬁg. 4 we study the internal representation in terms of discrimination and dimensionality using the MNIST dataset. For each choice of dimensionality all models are trained using the same number of operations. This is set to the amount necessary to complete one epoch over the training set using the Global model. This experiment shows that: 1) Local outperforms all other weight-sharing schemes for a wide range of dimensionalities, 2) TConv does not perform as well as Local probably because the translation invariant assumption is clearly violated for these relatively small, centered, images, 3) Conv performs well only when the internal representation is very high dimensional (10 times overcomplete) otherwise it severely underﬁts, 4) Global performs well when the representation is compact but its performance degrades rapidly as this increases because it needs more than the allotted training time. The right hand side of ﬁg. 4 shows how the recognition performance evolves as we increase the number of operations (or training time) using models that produce a twice overcomplete internal representation. With only very few ﬁlters Conv still underﬁts and it does not improve its performance by training for longer, but Global does improve and eventually it reaches the performance of Local. If we look at the crossing of the error rate at 2% we can see that Local is about 4 times faster than Global. To summarize, Local provides more compact representations than Conv, is much faster than Global while achieving 7 6 2.4 error rate % 5 error rate % 2.6 Global Local TConv Conv 4 3 2 1 0 2.2 Global Local 2 Conv 1.8 1000 2000 3000 4000 5000 dimensionality 6000 7000 1.6 0 8000 2 4 6 8 # flops (relative to # flops per epoch of Global model) 10 Figure 4: Experiments on MNIST using RBM’s with different weight-sharing schemes. Left: Error rate as a function of the dimensionality of the latent representation. Right: Error rate as a function of the number of operations (normalized to those needed to perform one epoch in the Global model); all models have a twice overcomplete latent representation. similar performance in discrimination. Also, Local can easily scale to larger images while Global cannot. Similar experiments are performed using the CIFAR 10 dataset [22] of natural images. Using the same protocol introduced in earlier work by Krizhevsky [22], the RBM’s are trained in an unsupervised way on a subset of the 80 million tiny images dataset [33] and then “ﬁne-tuned” on the CIFAR 10 dataset by supervised back-propagation of the error through the linear classiﬁer and feature extractor. All models produce an approximately 10,000 dimensional internal representation to make a fair comparison. Models using local ﬁlters learn 16x16 ﬁlters that are stepped every pixel. Again, we do not experiment with the TConv weight-sharing scheme because the image is not large enough to allow enough replicas. Similarly to ﬁg. 3-iii the Conv weight-sharing scheme was very difﬁcult to train and did not produce Gabor-like features. Indeed, careful injection of sparsity and long training time seem necessary [31] for these RBM’s. By contrast, both Local and Global produce Gabor-like ﬁlters similar to those shown in ﬁg. 2 F). The model trained with Conv weight-sharing scheme yields an accuracy equal to 56.6%, while Local and Global yield much better performance, 63.6% and 64.8% [22], respectively. Although Local and Global have similar performance, training with the Local weight-sharing scheme took under an hour while using the Global weight-sharing scheme required more than a day. 5 Conclusions and Future Work This work is motivated by the poor generative quality of currently popular MRF models of natural images. These models generate images that are actually more similar to white noise than to natural images. Our contribution is to recognize that current models can beneﬁt from 1) the addition of a simple model of the mean intensities and from 2) the use of a less constrained weight-sharing scheme. By augmenting these models with an extra set of latent variables that model mean intensity we can generate samples that look much more realistic: they are characterized by smooth regions, sharp boundaries and some simple high frequency texture. We validate our approach by comparing the statistics of ﬁlter outputs on natural images and generated images. In the future, we plan to integrate these MRF’s into deeper hierarchical models and to use their internal representation to perform object recognition in high-resolution images. The hope is to further improve generation by capturing longer range dependencies and to exploit this to better cope with missing values and ambiguous sensory inputs. References [1] E.P. Simoncelli. Statistical modeling of photographic images. Handbook of Image and Video Processing, pages 431–441, 2005. 8 [2] A. Hyvarinen, J. Karhunen, and E. Oja. Independent Component Analysis. John Wiley & Sons, 2001. [3] G.E. Hinton and R. R Salakhutdinov. Reducing the dimensionality of data with neural networks. Science, 313(5786):504–507, 2006. [4] M. Ranzato and G.E. Hinton. Modeling pixel means and covariances using factorized third-order boltzmann machines. In CVPR, 2010. [5] M.J. Wainwright and E.P. Simoncelli. Scale mixtures of gaussians and the statistics of natural images. In NIPS, 2000. [6] S. Roth and M.J. Black. Fields of experts: A framework for learning image priors. In CVPR, 2005. [7] U. Schmidt, Q. Gao, and S. Roth. A generative perspective on mrfs in low-level vision. In CVPR, 2010. [8] S. Geman and D. Geman. Stochastic relaxation, gibbs distributions, and the bayesian restoration of images. PAMI, 6:721–741, 1984. [9] M. Welling, G.E. Hinton, and S. Osindero. Learning sparse topographic representations with products of student-t distributions. In NIPS, 2003. [10] S.C. Zhu and D. Mumford. Prior learning and gibbs reaction diffusion. PAMI, pages 1236–1250, 1997. [11] S. Osindero, M. Welling, and G. E. Hinton. Topographic product models applied to natural scene statistics. Neural Comp., 18:344–381, 2006. [12] S. Osindero and G. E. Hinton. Modeling image patches with a directed hierarchy of markov random ﬁelds. In NIPS, 2008. [13] Y. Karklin and M.S. Lewicki. Emergence of complex cell properties by learning to generalize in natural scenes. Nature, 457:83–86, 2009. [14] B. A. Olshausen and D. J. Field. Sparse coding with an overcomplete basis set: a strategy employed by v1? Vision Research, 37:3311–3325, 1997. [15] Y. W. Teh, M. Welling, S. Osindero, and G. E. Hinton. Energy-based models for sparse overcomplete representations. JMLR, 4:1235–1260, 2003. [16] Y. Weiss and W.T. Freeman. What makes a good model of natural images? In CVPR, 2007. [17] S. Roth and M. J. Black. Fields of experts. Int. Journal of Computer Vision, 82:205–229, 2009. [18] K. Gregor and Y. LeCun. Emergence of complex-like cells in a temporal product network with local receptive ﬁelds. arXiv:1006.0448, 2010. [19] C. Tang and C. Eliasmith. Deep networks for robust visual recognition. In ICML, 2010. [20] M. Ranzato, A. Krizhevsky, and G.E. Hinton. Factored 3-way restricted boltzmann machines for modeling natural images. In AISTATS, 2010. [21] N. Heess, C.K.I. Williams, and G.E. Hinton. Learning generative texture models with extended ﬁelds-ofexperts. In BMCV, 2009. [22] A. Krizhevsky. Learning multiple layers of features from tiny images, 2009. MSc Thesis, Dept. of Comp. Science, Univ. of Toronto. [23] A. Waibel, T. Hanazawa, G. Hinton, K. Shikano, and K. Lang. Phoneme recognition using time-delay neural networks. IEEE Acoustics Speech and Signal Proc., 37:328–339, 1989. [24] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. [25] T. Tieleman and G.E. Hinton. Using fast weights to improve persistent contrastive divergence. In ICML, 2009. [26] R.M. Neal. Bayesian learning for neural networks. Springer-Verlag, 1996. [27] T. Tieleman. Training restricted boltzmann machines using approximations to the likelihood gradient. In ICML, 2008. [28] http://www.cs.berkeley.edu/projects/vision/grouping/segbench/. [29] M. Welling, M. Rosen-Zvi, and G.E. Hinton. Exponential family harmoniums with an application to information retrieval. In NIPS, 2005. [30] http://yann.lecun.com/exdb/mnist/. [31] H. Lee, R. Grosse, R. Ranganath, and A. Y. Ng. Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations. In Proc. ICML, 2009. [32] G.E. Hinton. Training products of experts by minimizing contrastive divergence. Neural Computation, 14:1771–1800, 2002. [33] A. Torralba, R. Fergus, and W.T. Freeman. 80 million tiny images: a large dataset for non-parametric object and scene recognition. PAMI, 30:1958–1970, 2008. 9

Reference: text

Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 Generating more realistic images using gated MRF’s Marc’Aurelio Ranzato Volodymyr Mnih Geoffrey E. [sent-1, score-0.423]

2 edu Abstract Probabilistic models of natural images are usually evaluated by measuring performance on rather indirect tasks, such as denoising and inpainting. [sent-4, score-0.416]

3 A more direct way to evaluate a generative model is to draw samples from it and to check whether statistical properties of the samples match the statistics of natural images. [sent-5, score-0.219]

4 The overall model can be interpreted as a gated MRF where both pair-wise dependencies and mean intensities of pixels are modulated by the states of latent variables. [sent-8, score-0.675]

5 Finally, we conﬁrm that if we disallow weight-sharing between receptive ﬁelds that overlap each other, the gated MRF learns more efﬁcient internal representations, as demonstrated in several recognition tasks. [sent-9, score-0.276]

6 1 Introduction and Prior Work The study of the statistical properties of natural images has a long history and has inﬂuenced many ﬁelds, from image processing to computational neuroscience [1]. [sent-10, score-0.432]

7 Markov Random Fields (MRF’s) provide a very general framework for modelling natural images. [sent-14, score-0.177]

8 In this work we consider a very versatile class of MRF’s in which potential functions are deﬁned over both pixels and latent variables, thus allowing the states of the latent variables to modulate or gate the effective interactions between the pixels. [sent-16, score-0.617]

9 This type of MRF, that we dub gated MRF, was proposed as an image model by Geman and Geman [8]. [sent-17, score-0.255]

10 [9] showed how an MRF in this family1 could be learned for small image patches and their work was extended to high-resolution images by Roth and Black [6] who also demonstrated its success in some practical applications [7]. [sent-19, score-0.478]

11 It is, therefore, very troublesome that none of the existing models can generate good samples, especially for high-resolution images (see for instance ﬁg. [sent-23, score-0.312]

12 2 in [7] which is one of the best models of highresolution images reported in the literature so far). [sent-24, score-0.312]

13 In fact, as our experiments demonstrate the generated samples from these models are more similar to random images than to natural images! [sent-25, score-0.461]

14 When MRF’s with gated interactions are applied to small image patches, they actually seem to work moderately well, as demonstrated by several authors [11, 12, 13]. [sent-26, score-0.308]

15 The generated patches have some coherent and elongated structure and, like natural image patches, they are predominantly very smooth with sudden outbreaks of strong structure. [sent-27, score-0.333]

16 This is unsurprising because these models have a built-in assumption that images are very smooth with occasional strong violations of smoothness [8, 14, 15]. [sent-28, score-0.343]

17 However, the extension of these patch-based models to high-resolution images by replicating ﬁlters across the image has proven to be difﬁcult. [sent-29, score-0.459]

18 The receptive ﬁelds that are learned no longer resemble Gabor wavelets but look random [6, 16] and the generated images lack any of the long range structure that is so typical of natural images [7]. [sent-30, score-0.724]

19 In this work, we show that the generative quality of these models can be drastically improved by jointly modelling both pixel mean intensities and pixel covariances. [sent-32, score-0.522]

20 This can be achieved by using two sets of latent variables, one that gates pair-wise interactions between pixels and another one that sets the mean intensities of pixels, as we already proposed in some earlier work [4]. [sent-33, score-0.547]

21 Here, we show that this modelling choice is crucial to make the gated MRF work well on high-resolution images. [sent-34, score-0.269]

22 Finally, we show that the most widely used method of sharing weights in MRF’s for high-resolution images is overly constrained. [sent-35, score-0.294]

23 2 Augmenting Gated MRF’s with Mean Hidden Units A Product of Student’s t (PoT) model [15] is a gated MRF deﬁned on small image patches that can be viewed as modelling image-speciﬁc, pair-wise relationships between pixel values by using the states of its latent variables. [sent-39, score-0.694]

24 It is very good at representing the fact that two-pixel have very similar intensities and no good at all at modelling what these intensities are. [sent-40, score-0.315]

25 Failure to model the mean also leads to impoverished modelling of the covariances when the input images have nonzero mean intensity. [sent-41, score-0.567]

26 The covariance RBM (cRBM) [20] is another model that shares the same limitation since it only differs from PoT in the distribution of its latent variables: The posterior over the latent variables is a product of Bernoulli distributions instead of Gamma distributions as in PoT. [sent-42, score-0.531]

27 We explain the fundamental limitation of these models by using a simple toy example: Modelling two-pixel images using a cRBM with only one binary hidden unit, see ﬁg. [sent-43, score-0.312]

28 This cRBM assumes that the conditional distribution over the input is a zero-mean Gaussian with a covariance that is determined by the state of the latent variable. [sent-45, score-0.308]

29 Since the latent variable is binary, the cRBM can be viewed as a mixture of two zero-mean full covariance Gaussians. [sent-46, score-0.278]

30 The latent variable uses the pairwise relationship between pixels to decide which of the two covariance matrices should be used to model each image. [sent-47, score-0.468]

31 When the input data is pre-proessed by making each image have zero mean intensity (the empirical histogram is shown in the ﬁrst row and ﬁrst column), most images lie near the origin because most of the times nearby pixels are strongly correlated. [sent-48, score-0.741]

32 Less frequently we encounter edge images that exhibit strong anti-correlation between the pixels, as shown by the long tails along the anti-diagonal line. [sent-49, score-0.269]

33 A cRBM could model this data by using two Gaussians (ﬁrst row and second column): one that is spherical and tight at the origin for smooth images and another one that has a covariance elongated along the anti-diagonal for structured images. [sent-50, score-0.462]

34 If, however, the whole set of images is normalized by subtracting from every pixel the mean value of all pixels over all images (second row and ﬁrst column), the cRBM fails at modelling structured images (second row and second column). [sent-51, score-1.291]

35 It can ﬁt a Gaussian to the smooth images by discovering 2 Figure 1: In the ﬁrst row, each image is zero mean. [sent-52, score-0.401]

36 The ﬁrst column shows 8x8 images picked at random from natural images. [sent-54, score-0.37]

37 The images in the second column are generated by a model that does not account for mean intensity. [sent-55, score-0.393]

38 The images in the third column are generated by a model that has both “mean” and “covariance” hidden units. [sent-56, score-0.338]

39 The contours in the ﬁrst column show the negative log of the empirical distribution of (tiny) natural two-pixel images (x-axis being the ﬁrst pixel and the y-axis the second pixel). [sent-57, score-0.468]

40 The plots in the other columns are toy examples showing how each model could represent the empirical distribution using a mixture of Gaussians with components that have one of two possible covariances (corresponding to the state of a binary “covariance” latent variable). [sent-58, score-0.219]

41 Models that can change the means of the Gaussians (mPoT and mcRBM) can represent better structured images (edge images lie along the anti-diagonal and are ﬁtted by the Gaussians shown in red) while the other models (PoT and cRBM) fail, overall when each image can have non-zero mean. [sent-59, score-0.682]

42 the direction of strong correlation along the main diagonal, but it is very likely to fail to discover the direction of anti-correlation, which is crucial to represent discontinuities, because structured images with different mean intensity appear to be evenly spread over the whole input space. [sent-60, score-0.479]

43 The mean latent variables effectively subtract off the relevant mean from each data-point, letting the covariance latent variable capture the covariance structure of the data. [sent-62, score-0.743]

44 As before, the covariance latent variable needs only to select between two covariance matrices. [sent-63, score-0.38]

45 In fact, experiments on real 8x8 image patches conﬁrm these conjectures. [sent-64, score-0.179]

46 mPoT (and similarly mcRBM [4]) is not only better at modelling zero mean images but it can also represent images that have non zero mean intensity well. [sent-67, score-0.859]

47 In PoT [9] the energy function is: E PoT (x, hc ) = i 1 [hc (1 + (Ci T x)2 ) + (1 − γ) log hc ] i i 2 (1) where x is a vectorized image patch, hc is a vector of Gamma “covariance” latent variables, C is a ﬁlter bank matrix and γ is a scalar parameter. [sent-69, score-0.886]

48 The joint probability over input pixels and latent variables is proportional to exp(−E PoT (x, hc )). [sent-70, score-0.628]

49 Therefore, the conditional distribution over the input pixels is a zero-mean Gaussian with covariance equal to: Σc = (Cdiag(hc )C T )−1 . [sent-71, score-0.322]

50 (2) In order to make the mean of the conditional distribution non-zero, we deﬁne mPoT as the normalized product of the above zero-mean Gaussian that models the covariance and a spherical covariance Gaussian that models the mean. [sent-72, score-0.427]

51 The overall energy function becomes: E mPoT (x, hc , hm ) = E PoT (x, hc ) + E m (x, hm ) 3 (3) Figure 2: Illustration of different choices of weight-sharing scheme for a RBM. [sent-73, score-0.686]

52 Each ﬁlter has size 16x16 pixels and it is applied every 16 pixels in both the horizontal and vertical directions. [sent-81, score-0.38]

53 Filters in position (i, j) and (1, 1) are applied to neighborhoods that are (i, j) pixels away form each other. [sent-82, score-0.229]

54 where hm is another set of latent variables that are assumed to be Bernoulli distributed (but other distributions could be used). [sent-84, score-0.319]

55 The new energy term is: E m (x, hm ) = 1 T x x− 2 hm Wj T x j (4) j yielding the following conditional distribution over the input pixels: p(x|hc , hm ) = N (Σ(W hm ), Σ), Σ = (Σc + I)−1 (5) with Σc deﬁned in eq. [sent-85, score-0.466]

56 Patch-based models like PoT have been extended to high-resolution images by using spatially localized ﬁlters [6]. [sent-88, score-0.34]

57 While we can subtract off the mean intensity from independent image patches to successfully train PoT, we cannot do that on a high-resolution image because overlapping patches might have different mean. [sent-89, score-0.537]

58 Unfortunately, replicating potentials over the image ignoring variations of mean intensity has been the leading strategy to date [6]3 . [sent-90, score-0.373]

59 This is the major reason why generation of high-resolution images is so poor. [sent-91, score-0.269]

60 3 Weight-Sharing Schemes By integrating out the latent variables, we can write the density function of any gated MRF as a normalized product of potential functions (for mPoT refer to eq. [sent-94, score-0.383]

61 Global: The obvious way to extend a patch-based model like PoT to high-resolution images is to deﬁne potentials over the whole image; we call this scheme global. [sent-97, score-0.445]

62 Conv: The most popular way to handle big images is to deﬁne potentials on small subsets of variables (e. [sent-99, score-0.393]

63 3 The success of PoT-like models in Bayesian denoising is not surprising since the noisy image effectively replaces the reconstruction term from the mean hidden units (see eq. [sent-102, score-0.268]

64 5), providing a set of noisy mean intensities that are cleaned up by the patterns of correlation enforced by the covariance latent variables. [sent-103, score-0.433]

65 This weight-sharing scheme is extremely concise in terms of number of parameters, but also rather inefﬁcient in terms of latent representation. [sent-107, score-0.248]

66 First, if there are N ﬁlters at each location and these ﬁlters are stepped by one pixel then the internal representation is about N times overcomplete. [sent-108, score-0.172]

67 Since the input is mostly smooth and the parameters are the same across space, the latent variables are strongly correlated as well. [sent-110, score-0.286]

68 1 is square and invertible, we can marginalize out the latent variables and write: p(y) = i S(yi ), where yi = Ci T x and S is a Student’s t distribution. [sent-114, score-0.225]

69 Compared to Conv the number of parameters increases only slightly but the number of latent variables required and their redundancy is greatly reduced. [sent-122, score-0.225]

70 TConv: Local would not allow the model to be trained and tested on images of different resolution, and it might seem wasteful not to exploit the translation invariant property of images. [sent-125, score-0.327]

71 This reduces spatial redundancy of latent variables and allows the input images to have arbitrary size. [sent-131, score-0.524]

72 4 Experiments We train gated MRF’s with and without mean hidden units using different weight-sharing schemes. [sent-135, score-0.236]

73 Since all latent variables can be exactly marginalized out we can use HMC on the free energy (negative logarithm of the marginal distribution over the input pixels). [sent-138, score-0.315]

74 The aim is to encourage the negative particles to explore the space more quickly by slightly and temporarily raising the energy at their current position. [sent-149, score-0.178]

75 We train on mini-batches of 32 samples using gray-scale images of approximate size 160x160 pixels randomly cropped from the Berkeley segmentation dataset [28]. [sent-151, score-0.543]

76 We learn 8x8 ﬁlters and pre-multiply the covariance ﬁlters by a whitening transform retaining 99% of the variance; we also normalize the norm of the covariance ﬁlters to prevent some of them from decaying to zero during training4 . [sent-163, score-0.204]

77 Whenever we use the TConv weight-sharing scheme the model learns covariance ﬁlters that mostly resemble localized and oriented Gabor functions (see ﬁg. [sent-164, score-0.255]

78 The TConv models re-use the same 8x8 ﬁlters every 8 pixels and apply a diagonal offset of 2 pixels between neighboring ﬁlters with different weights in order to reduce tiling artifacts. [sent-167, score-0.454]

79 Some samples of size 160x160 pixels are reported in ﬁg. [sent-174, score-0.247]

80 Without modelling the mean intensity, samples lack structure and do not seem much different from those that would be generated by a simple Gaussian model merely ﬁtting the second order statistics (see ﬁg. [sent-176, score-0.284]

81 By contrast, structure, sharp boundaries and some simple texture emerge only from models that have mean latent variables, namely mcRBM, mPoT and mHPoT which differs from mPoT by having a second layer pooling matrix on the squared covariance ﬁlter outputs [11]. [sent-179, score-0.402]

82 We ﬁrst compute marginal statistics of ﬁlter responses using the generated images, natural images from the test set, and random images. [sent-181, score-0.361]

83 We then calculate the KL divergence between the histograms on random images and generated images and the KL divergence between the histograms on natural images and generated images. [sent-183, score-1.031]

84 The table also reports the average difference of energies between random images and natural images. [sent-184, score-0.331]

85 All results demonstrate that models that account for mean intensity generate images 4 The code used in the experiments can be found at the ﬁrst author’s web-page. [sent-185, score-0.463]

86 that are closer to natural images than to random images, whereas models that do not account for the mean (like the widely used PoT-Conv) produce samples that are actually closer to random images. [sent-208, score-0.486]

87 The MNIST dataset has soft binary images of size 28x28 pixels, while the CIFAR 10 dataset has color images of size 32x32 pixels. [sent-214, score-0.592]

88 The energy function of the RBM trained on the CIFAR 10 dataset, modelling input pixels with 3 (R,G,B) Gaussian variables [31], is exactly the one shown in eq. [sent-217, score-0.475]

89 4; while the RBM trained on MNIST uses logistic units for the pixels and the energy function is again the same as before but without any quadratic term. [sent-218, score-0.308]

90 They are then used to represent each input image with a feature vector (mean of the posterior over the latent variables) which is fed to a multinomial logistic classiﬁer for discrimination. [sent-220, score-0.307]

91 4 shows how the recognition performance evolves as we increase the number of operations (or training time) using models that produce a twice overcomplete internal representation. [sent-229, score-0.187]

92 Left: Error rate as a function of the dimensionality of the latent representation. [sent-238, score-0.203]

93 Right: Error rate as a function of the number of operations (normalized to those needed to perform one epoch in the Global model); all models have a twice overcomplete latent representation. [sent-239, score-0.321]

94 Also, Local can easily scale to larger images while Global cannot. [sent-241, score-0.269]

95 Again, we do not experiment with the TConv weight-sharing scheme because the image is not large enough to allow enough replicas. [sent-246, score-0.173]

96 These models generate images that are actually more similar to white noise than to natural images. [sent-258, score-0.374]

97 Our contribution is to recognize that current models can beneﬁt from 1) the addition of a simple model of the mean intensities and from 2) the use of a less constrained weight-sharing scheme. [sent-259, score-0.198]

98 By augmenting these models with an extra set of latent variables that model mean intensity we can generate samples that look much more realistic: they are characterized by smooth regions, sharp boundaries and some simple high frequency texture. [sent-260, score-0.544]

99 We validate our approach by comparing the statistics of ﬁlter outputs on natural images and generated images. [sent-261, score-0.387]

100 Modeling image patches with a directed hierarchy of markov random ﬁelds. [sent-336, score-0.179]

similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('pot', 0.326), ('tconv', 0.326), ('images', 0.269), ('lters', 0.266), ('mpot', 0.265), ('conv', 0.247), ('mrf', 0.2), ('pixels', 0.19), ('hc', 0.183), ('latent', 0.176), ('gated', 0.154), ('crbm', 0.115), ('modelling', 0.115), ('cifar', 0.107), ('rbm', 0.106), ('covariance', 0.102), ('image', 0.101), ('intensities', 0.1), ('intensity', 0.096), ('hm', 0.094), ('particles', 0.09), ('patches', 0.078), ('mcrbm', 0.077), ('lter', 0.076), ('kl', 0.076), ('potentials', 0.075), ('scheme', 0.072), ('pixel', 0.07), ('hmc', 0.07), ('natural', 0.062), ('global', 0.062), ('fpcd', 0.061), ('internal', 0.061), ('mnist', 0.06), ('energy', 0.06), ('samples', 0.057), ('gabor', 0.056), ('overcomplete', 0.056), ('mean', 0.055), ('contrastive', 0.053), ('welling', 0.052), ('divergence', 0.051), ('tiny', 0.05), ('variables', 0.049), ('local', 0.048), ('gaussians', 0.047), ('replicating', 0.046), ('epoch', 0.046), ('generative', 0.043), ('models', 0.043), ('covariances', 0.043), ('denoising', 0.042), ('flops', 0.041), ('mhpot', 0.041), ('stepped', 0.041), ('geman', 0.039), ('neighborhoods', 0.039), ('column', 0.039), ('boltzmann', 0.038), ('augmenting', 0.037), ('roth', 0.037), ('schemes', 0.037), ('osindero', 0.036), ('receptive', 0.036), ('gregor', 0.036), ('hinton', 0.035), ('ranzato', 0.033), ('student', 0.032), ('perturbation', 0.032), ('pami', 0.032), ('fields', 0.031), ('elongated', 0.031), ('tiling', 0.031), ('smooth', 0.031), ('trained', 0.031), ('input', 0.03), ('learned', 0.03), ('generated', 0.03), ('whole', 0.029), ('krizhevsky', 0.029), ('spherical', 0.029), ('localized', 0.028), ('product', 0.028), ('negative', 0.028), ('resemble', 0.028), ('subtract', 0.028), ('dataset', 0.027), ('units', 0.027), ('training', 0.027), ('seem', 0.027), ('filters', 0.027), ('dimensionality', 0.027), ('cvpr', 0.026), ('interactions', 0.026), ('outputs', 0.026), ('drastically', 0.026), ('learns', 0.025), ('sharing', 0.025), ('gaussian', 0.025), ('normalized', 0.025)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 1.0 103 nips-2010-Generating more realistic images using gated MRF's

Author: Marc'aurelio Ranzato, Volodymyr Mnih, Geoffrey E. Hinton

2 0.14007264 99 nips-2010-Gated Softmax Classification

Author: Roland Memisevic, Christopher Zach, Marc Pollefeys, Geoffrey E. Hinton

Abstract: We describe a ”log-bilinear” model that computes class probabilities by combining an input vector multiplicatively with a vector of binary latent variables. Even though the latent variables can take on exponentially many possible combinations of values, we can efﬁciently compute the exact probability of each class by marginalizing over the latent variables. This makes it possible to get the exact gradient of the log likelihood. The bilinear score-functions are deﬁned using a three-dimensional weight tensor, and we show that factorizing this tensor allows the model to encode invariances inherent in a task by learning a dictionary of invariant basis functions. Experiments on a set of benchmark problems show that this fully probabilistic model can achieve classiﬁcation performance that is competitive with (kernel) SVMs, backpropagation, and deep belief nets. 1

3 0.13742629 206 nips-2010-Phone Recognition with the Mean-Covariance Restricted Boltzmann Machine

Author: George Dahl, Marc'aurelio Ranzato, Abdel-rahman Mohamed, Geoffrey E. Hinton

Abstract: Straightforward application of Deep Belief Nets (DBNs) to acoustic modeling produces a rich distributed representation of speech data that is useful for recognition and yields impressive results on the speaker-independent TIMIT phone recognition task. However, the ﬁrst-layer Gaussian-Bernoulli Restricted Boltzmann Machine (GRBM) has an important limitation, shared with mixtures of diagonalcovariance Gaussians: GRBMs treat different components of the acoustic input vector as conditionally independent given the hidden state. The mean-covariance restricted Boltzmann machine (mcRBM), ﬁrst introduced for modeling natural images, is a much more representationally efﬁcient and powerful way of modeling the covariance structure of speech data. Every conﬁguration of the precision units of the mcRBM speciﬁes a different precision matrix for the conditional distribution over the acoustic space. In this work, we use the mcRBM to learn features of speech data that serve as input into a standard DBN. The mcRBM features combined with DBNs allow us to achieve a phone error rate of 20.5%, which is superior to all published results on speaker-independent TIMIT to date. 1

4 0.13582104 143 nips-2010-Learning Convolutional Feature Hierarchies for Visual Recognition

Author: Koray Kavukcuoglu, Pierre Sermanet, Y-lan Boureau, Karol Gregor, Michael Mathieu, Yann L. Cun

Abstract: We propose an unsupervised method for learning multi-stage hierarchies of sparse convolutional features. While sparse coding has become an increasingly popular method for learning visual features, it is most often trained at the patch level. Applying the resulting ﬁlters convolutionally results in highly redundant codes because overlapping patches are encoded in isolation. By training convolutionally over large image windows, our method reduces the redudancy between feature vectors at neighboring locations and improves the efﬁciency of the overall representation. In addition to a linear decoder that reconstructs the image from sparse features, our method trains an efﬁcient feed-forward encoder that predicts quasisparse features from the input. While patch-based training rarely produces anything but oriented edge detectors, we show that convolutional training produces highly diverse ﬁlters, including center-surround ﬁlters, corner detectors, cross detectors, and oriented grating detectors. We show that using these ﬁlters in multistage convolutional network architecture improves performance on a number of visual recognition and detection tasks. 1

5 0.13483411 101 nips-2010-Gaussian sampling by local perturbations

Author: George Papandreou, Alan L. Yuille

Abstract: We present a technique for exact simulation of Gaussian Markov random ﬁelds (GMRFs), which can be interpreted as locally injecting noise to each Gaussian factor independently, followed by computing the mean/mode of the perturbed GMRF. Coupled with standard iterative techniques for the solution of symmetric positive deﬁnite systems, this yields a very efﬁcient sampling algorithm with essentially linear complexity in terms of speed and memory requirements, well suited to extremely large scale probabilistic models. Apart from synthesizing data under a Gaussian model, the proposed technique directly leads to an efﬁcient unbiased estimator of marginal variances. Beyond Gaussian models, the proposed algorithm is also very useful for handling highly non-Gaussian continuously-valued MRFs such as those arising in statistical image modeling or in the ﬁrst layer of deep belief networks describing real-valued data, where the non-quadratic potentials coupling different sites can be represented as ﬁnite or inﬁnite mixtures of Gaussians with the help of local or distributed latent mixture assignment variables. The Bayesian treatment of such models most naturally involves a block Gibbs sampler which alternately draws samples of the conditionally independent latent mixture assignments and the conditionally multivariate Gaussian continuous vector and we show that it can directly beneﬁt from the proposed methods. 1

6 0.13099892 224 nips-2010-Regularized estimation of image statistics by Score Matching

7 0.12746544 240 nips-2010-Simultaneous Object Detection and Ranking with Weak Supervision

8 0.1188454 133 nips-2010-Kernel Descriptors for Visual Recognition

9 0.11666221 288 nips-2010-Worst-case bounds on the quality of max-product fixed-points

10 0.11321051 266 nips-2010-The Maximal Causes of Natural Scenes are Edge Filters

11 0.10921305 149 nips-2010-Learning To Count Objects in Images

12 0.1066457 89 nips-2010-Factorized Latent Spaces with Structured Sparsity

13 0.10405187 156 nips-2010-Learning to combine foveal glimpses with a third-order Boltzmann machine

14 0.10308155 242 nips-2010-Slice sampling covariance hyperparameters of latent Gaussian models

15 0.097405195 235 nips-2010-Self-Paced Learning for Latent Variable Models

16 0.096357517 209 nips-2010-Pose-Sensitive Embedding by Nonlinear NCA Regression

17 0.092935063 86 nips-2010-Exploiting weakly-labeled Web images to improve object classification: a domain adaptation approach

18 0.089849524 186 nips-2010-Object Bank: A High-Level Image Representation for Scene Classification & Semantic Feature Sparsification

19 0.087494522 213 nips-2010-Predictive Subspace Learning for Multi-view Data: a Large Margin Approach

20 0.08406055 271 nips-2010-Tiled convolutional neural networks

similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, 0.213), (1, 0.106), (2, -0.177), (3, -0.114), (4, -0.023), (5, -0.03), (6, 0.025), (7, 0.128), (8, -0.06), (9, 0.021), (10, -0.047), (11, -0.043), (12, 0.061), (13, -0.124), (14, -0.124), (15, -0.038), (16, 0.004), (17, -0.038), (18, 0.09), (19, -0.06), (20, 0.006), (21, 0.091), (22, -0.025), (23, 0.05), (24, -0.071), (25, 0.073), (26, -0.016), (27, -0.061), (28, 0.029), (29, 0.004), (30, 0.039), (31, -0.026), (32, 0.093), (33, -0.02), (34, 0.063), (35, -0.071), (36, -0.058), (37, -0.125), (38, 0.08), (39, -0.065), (40, 0.083), (41, 0.004), (42, 0.015), (43, 0.098), (44, -0.143), (45, 0.12), (46, -0.094), (47, 0.095), (48, -0.018), (49, 0.058)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.95664316 103 nips-2010-Generating more realistic images using gated MRF's

Author: Marc'aurelio Ranzato, Volodymyr Mnih, Geoffrey E. Hinton

2 0.71041042 101 nips-2010-Gaussian sampling by local perturbations

Author: George Papandreou, Alan L. Yuille

3 0.69488549 266 nips-2010-The Maximal Causes of Natural Scenes are Edge Filters

Author: Jose Puertas, Joerg Bornschein, Joerg Luecke

Abstract: We study the application of a strongly non-linear generative model to image patches. As in standard approaches such as Sparse Coding or Independent Component Analysis, the model assumes a sparse prior with independent hidden variables. However, in the place where standard approaches use the sum to combine basis functions we use the maximum. To derive tractable approximations for parameter estimation we apply a novel approach based on variational Expectation Maximization. The derived learning algorithm can be applied to large-scale problems with hundreds of observed and hidden variables. Furthermore, we can infer all model parameters including observation noise and the degree of sparseness. In applications to image patches we ﬁnd that Gabor-like basis functions are obtained. Gabor-like functions are thus not a feature exclusive to approaches assuming linear superposition. Quantitatively, the inferred basis functions show a large diversity of shapes with many strongly elongated and many circular symmetric functions. The distribution of basis function shapes reﬂects properties of simple cell receptive ﬁelds that are not reproduced by standard linear approaches. In the study of natural image statistics, the implications of using different superposition assumptions have so far not been investigated systematically because models with strong non-linearities have been found analytically and computationally challenging. The presented algorithm represents the ﬁrst large-scale application of such an approach. 1

4 0.67663151 256 nips-2010-Structural epitome: a way to summarize one’s visual experience

Author: Nebojsa Jojic, Alessandro Perina, Vittorio Murino

Abstract: In order to study the properties of total visual input in humans, a single subject wore a camera for two weeks capturing, on average, an image every 20 seconds. The resulting new dataset contains a mix of indoor and outdoor scenes as well as numerous foreground objects. Our ﬁrst goal is to create a visual summary of the subject’s two weeks of life using unsupervised algorithms that would automatically discover recurrent scenes, familiar faces or common actions. Direct application of existing algorithms, such as panoramic stitching (e.g., Photosynth) or appearance-based clustering models (e.g., the epitome), is impractical due to either the large dataset size or the dramatic variations in the lighting conditions. As a remedy to these problems, we introduce a novel image representation, the ”structural element (stel) epitome,” and an associated efﬁcient learning algorithm. In our model, each image or image patch is characterized by a hidden mapping T which, as in previous epitome models, deﬁnes a mapping between the image coordinates and the coordinates in the large ”all-I-have-seen” epitome matrix. The limited epitome real-estate forces the mappings of different images to overlap which indicates image similarity. However, the image similarity no longer depends on direct pixel-to-pixel intensity/color/feature comparisons as in previous epitome models, but on spatial conﬁguration of scene or object parts, as the model is based on the palette-invariant stel models. As a result, stel epitomes capture structure that is invariant to non-structural changes, such as illumination changes, that tend to uniformly affect pixels belonging to a single scene or object part. 1

5 0.665811 224 nips-2010-Regularized estimation of image statistics by Score Matching

Author: Diederik P. Kingma, Yann L. Cun

Abstract: Score Matching is a recently-proposed criterion for training high-dimensional density models for which maximum likelihood training is intractable. It has been applied to learning natural image statistics but has so-far been limited to simple models due to the difﬁculty of differentiating the loss with respect to the model parameters. We show how this differentiation can be automated with an extended version of the double-backpropagation algorithm. In addition, we introduce a regularization term for the Score Matching loss that enables its use for a broader range of problem by suppressing instabilities that occur with ﬁnite training sample sizes and quantized input values. Results are reported for image denoising and super-resolution.

6 0.66209245 99 nips-2010-Gated Softmax Classification

7 0.65201163 245 nips-2010-Space-Variant Single-Image Blind Deconvolution for Removing Camera Shake

8 0.64662385 213 nips-2010-Predictive Subspace Learning for Multi-view Data: a Large Margin Approach

9 0.63284755 82 nips-2010-Evaluation of Rarity of Fingerprints in Forensics

10 0.63057643 156 nips-2010-Learning to combine foveal glimpses with a third-order Boltzmann machine

11 0.58851415 111 nips-2010-Hallucinations in Charles Bonnet Syndrome Induced by Homeostasis: a Deep Boltzmann Machine Model

12 0.57520789 235 nips-2010-Self-Paced Learning for Latent Variable Models

13 0.56604105 206 nips-2010-Phone Recognition with the Mean-Covariance Restricted Boltzmann Machine

14 0.54921001 149 nips-2010-Learning To Count Objects in Images

15 0.53449643 271 nips-2010-Tiled convolutional neural networks

16 0.522102 242 nips-2010-Slice sampling covariance hyperparameters of latent Gaussian models

17 0.51771981 77 nips-2010-Epitome driven 3-D Diffusion Tensor image segmentation: on extracting specific structures

18 0.51261485 262 nips-2010-Switched Latent Force Models for Movement Segmentation

19 0.51164627 89 nips-2010-Factorized Latent Spaces with Structured Sparsity

20 0.50246823 143 nips-2010-Learning Convolutional Feature Hierarchies for Visual Recognition

similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(13, 0.035), (17, 0.041), (27, 0.098), (30, 0.037), (35, 0.029), (45, 0.236), (50, 0.096), (52, 0.055), (59, 0.018), (60, 0.028), (77, 0.04), (90, 0.029), (96, 0.181)]

similar papers list:

simIndex simValue paperId paperTitle

1 0.94249612 245 nips-2010-Space-Variant Single-Image Blind Deconvolution for Removing Camera Shake

Author: Stefan Harmeling, Hirsch Michael, Bernhard Schölkopf

Abstract: Modelling camera shake as a space-invariant convolution simpliﬁes the problem of removing camera shake, but often insufﬁciently models actual motion blur such as those due to camera rotation and movements outside the sensor plane or when objects in the scene have different distances to the camera. In an effort to address these limitations, (i) we introduce a taxonomy of camera shakes, (ii) we build on a recently introduced framework for space-variant ﬁltering by Hirsch et al. and a fast algorithm for single image blind deconvolution for space-invariant ﬁlters by Cho and Lee to construct a method for blind deconvolution in the case of space-variant blur, and (iii), we present an experimental setup for evaluation that allows us to take images with real camera shake while at the same time recording the spacevariant point spread function corresponding to that blur. Finally, we demonstrate that our method is able to deblur images degraded by spatially-varying blur originating from real camera shake, even without using additionally motion sensor information. 1

same-paper 2 0.8671487 103 nips-2010-Generating more realistic images using gated MRF's

Author: Marc'aurelio Ranzato, Volodymyr Mnih, Geoffrey E. Hinton

3 0.82966048 109 nips-2010-Group Sparse Coding with a Laplacian Scale Mixture Prior

Author: Pierre Garrigues, Bruno A. Olshausen

Abstract: We propose a class of sparse coding models that utilizes a Laplacian Scale Mixture (LSM) prior to model dependencies among coefﬁcients. Each coefﬁcient is modeled as a Laplacian distribution with a variable scale parameter, with a Gamma distribution prior over the scale parameter. We show that, due to the conjugacy of the Gamma prior, it is possible to derive efﬁcient inference procedures for both the coefﬁcients and the scale parameter. When the scale parameters of a group of coefﬁcients are combined into a single variable, it is possible to describe the dependencies that occur due to common amplitude ﬂuctuations among coefﬁcients, which have been shown to constitute a large fraction of the redundancy in natural images [1]. We show that, as a consequence of this group sparse coding, the resulting inference of the coefﬁcients follows a divisive normalization rule, and that this may be efﬁciently implemented in a network architecture similar to that which has been proposed to occur in primary visual cortex. We also demonstrate improvements in image coding and compressive sensing recovery using the LSM model. 1

4 0.82788026 96 nips-2010-Fractionally Predictive Spiking Neurons

Author: Jaldert Rombouts, Sander M. Bohte

Abstract: Recent experimental work has suggested that the neural ﬁring rate can be interpreted as a fractional derivative, at least when signal variation induces neural adaptation. Here, we show that the actual neural spike-train itself can be considered as the fractional derivative, provided that the neural signal is approximated by a sum of power-law kernels. A simple standard thresholding spiking neuron sufﬁces to carry out such an approximation, given a suitable refractory response. Empirically, we ﬁnd that the online approximation of signals with a sum of powerlaw kernels is beneﬁcial for encoding signals with slowly varying components, like long-memory self-similar signals. For such signals, the online power-law kernel approximation typically required less than half the number of spikes for similar SNR as compared to sums of similar but exponentially decaying kernels. As power-law kernels can be accurately approximated using sums or cascades of weighted exponentials, we demonstrate that the corresponding decoding of spiketrains by a receiving neuron allows for natural and transparent temporal signal ﬁltering by tuning the weights of the decoding kernel. 1

5 0.82525474 51 nips-2010-Construction of Dependent Dirichlet Processes based on Poisson Processes

Author: Dahua Lin, Eric Grimson, John W. Fisher

Abstract: We present a novel method for constructing dependent Dirichlet processes. The approach exploits the intrinsic relationship between Dirichlet and Poisson processes in order to create a Markov chain of Dirichlet processes suitable for use as a prior over evolving mixture models. The method allows for the creation, removal, and location variation of component models over time while maintaining the property that the random measures are marginally DP distributed. Additionally, we derive a Gibbs sampling algorithm for model inference and test it on both synthetic and real data. Empirical results demonstrate that the approach is effective in estimating dynamically varying mixture models. 1

6 0.82403493 238 nips-2010-Short-term memory in neuronal networks through dynamical compressed sensing

7 0.81669968 21 nips-2010-Accounting for network effects in neuronal responses using L1 regularized point process models

8 0.81612372 17 nips-2010-A biologically plausible network for the computation of orientation dominance

9 0.81569183 44 nips-2010-Brain covariance selection: better individual functional connectivity models using population prior

10 0.81170946 158 nips-2010-Learning via Gaussian Herding

11 0.81112951 98 nips-2010-Functional form of motion priors in human motion perception

12 0.81107098 55 nips-2010-Cross Species Expression Analysis using a Dirichlet Process Mixture Model with Latent Matchings

13 0.80929029 85 nips-2010-Exact learning curves for Gaussian process regression on large random graphs

14 0.80793476 224 nips-2010-Regularized estimation of image statistics by Score Matching

15 0.80746627 239 nips-2010-Sidestepping Intractable Inference with Structured Ensemble Cascades

16 0.80740041 206 nips-2010-Phone Recognition with the Mean-Covariance Restricted Boltzmann Machine

17 0.80727893 54 nips-2010-Copula Processes

18 0.80689895 33 nips-2010-Approximate inference in continuous time Gaussian-Jump processes

19 0.80685931 155 nips-2010-Learning the context of a category

20 0.80674106 218 nips-2010-Probabilistic latent variable models for distinguishing between cause and effect