We present a novel method of compression of deep Convolutional Neural Networks (CNNs). Our method reduces the number of parameters of each convolutional layer by learning a 3D tensor termed Filter Summary (FS). The convolutional filters are extracted from FS as overlapping 3D blocks, and nearby filters in FS share weights in their overlapping regions in a natural way. The resultant neural network based on such weight sharing scheme, termed Filter Summary CNNs or FSNet, has a FS in each convolution layer instead of a set of independent filters in the conventional convolution layer. FSNet has the same architecture as that of the baseline CNN to be compressed, and each convolution layer of FSNet generates the same number of filters from FS as that of the basline CNN in the forward process. Without hurting the inference speed, the parameter space of FSNet is much smaller than that of the baseline CNN. In addition, FSNet is compatible with weight quantization, leading to even higher compression ratio when combined with weight quantization. Experiments demonstrate the effectiveness of FSNet in compression of CNNs for computer vision tasks including image classification and object detection. For classification task, FSNet of 0.22M effective parameters has prediction accuracy of 93.91% on the CIFAR-10 dataset with less than 0.3% accuracy drop, using ResNet-18 of 11.18M parameters as baseline. Furthermore, FSNet version of ResNet-50 with 2.75M effective parameters achieves the top-1 and top-5 accuracy of 63.80% and 85.72% respectively on ILSVRC-12 benchmark. For object detection task, FSNet is used to compress the Single Shot MultiBox Detector (SSD300) of 26.32M parameters. FSNet of 0.45M effective parameters achieves mAP of 67.63% on the VOC2007 test data with weight quantization, and FSNet of 0.68M effective parameters achieves mAP of 70.00% with weight quantization on the same test data.

Variational methods that rely on a recognition network to approximate the posterior of directed graphical models offer better inference and learning than previous methods. Recent advances that exploit the capacity and flexibility in this approach have expanded what kinds of models can be trained. However, as a proposal for the posterior, the capacity of the recognition network is limited, which can constrain the representational power of the generative model and increase the variance of Monte Carlo estimates. To address these issues, we introduce an iterative refinement procedure for improving the approximate posterior of the recognition network and show that training with the refined posterior is competitive with state-of-the-art methods. The advantages of refinement are further evident in an increased effective sample size, which implies a lower variance of gradient estimates.

Similarity-based clustering and semi-supervised learning methods separate the data into clusters or classes according to the pairwise similarity between the data, and the pairwise similarity is crucial for their performance. In this paper, we propose a novel discriminative similarity learning framework which learns discriminative similarity for either data clustering or semi-supervised learning. The proposed framework learns classifier from each hypothetical labeling, and searches for the optimal labeling by minimizing the generalization error of the learned classifiers associated with the hypothetical labeling. Kernel classifier is employed in our framework. By generalization analysis via Rademacher complexity, the generalization error bound for the kernel classifier learned from hypothetical labeling is expressed as the sum of pairwise similarity between the data from different classes, parameterized by the weights of the kernel classifier. Such pairwise similarity serves as the discriminative similarity for the purpose of clustering and semi-supervised learning, and discriminative similarity with similar form can also be induced by the integrated squared error bound for kernel density classification. Based on the discriminative similarity induced by the kernel classifier, we propose new clustering and semi-supervised learning methods.

Variational methods that rely on a recognition network to approximate the posterior of directed graphical models offer better inference and learning than previous methods. Recent advances that exploit the capacity and flexibility in this approach have expanded what kinds of models can be trained. However, as a proposal for the posterior, the capacity of the recognition network is limited, which can constrain the representational power of the generative model and increase the variance of Monte Carlo estimates. To address these issues, we introduce an iterative refinement procedure for improving the approximate posterior of the recognition network and show that training with the refined posterior is competitive with state-of-the-art methods. The advantages of refinement are further evident in an increased effective sample size, which implies a lower variance of gradient estimates.

The counting grid is a grid of microtopics, sparse word/feature distributions. The generative model associated with the grid does not use these microtopics individually. Rather, it groups them in overlapping rectangular windows and uses these grouped microtopics as either mixture or admixture components. This paper builds upon the basic counting grid model and it shows that hierarchical reasoning helps avoid bad local minima, produces better classification accuracy and, most interestingly, allows for extraction of large numbers of coherent microtopics even from small datasets. We evaluate this in terms of consistency, diversity and clarity of the indexed content, as well as in a user study on word intrusion tasks. We demonstrate that these models work well as a technique for embedding raw images and discuss interesting parallels between hierarchical CG models and other deep architectures.

In recent scene recognition research images or large image regions are often represented as disorganized "bags" of features which can then be analyzed using models originally developed to capture co-variation of word counts in text. However, image feature counts are likely to be constrained in different ways than word counts in text. For example, as a camera pans upwards from a building entrance over its first few floors and then further up into the sky Fig. 1, some feature counts in the image drop while others rise -- only to drop again giving way to features found more often at higher elevations. The space of all possible feature count combinations is constrained both by the properties of the larger scene and the size and the location of the window into it. To capture such variation, in this paper we propose the use of the counting grid model. This generative model is based on a grid of feature counts, considerably larger than any of the modeled images, and considerably smaller than the real estate needed to tile the images next to each other tightly. Each modeled image is assumed to have a representative window in the grid in which the feature counts mimic the feature distribution in the image. We provide a learning procedure that jointly maps all images in the training set to the counting grid and estimates the appropriate local counts in it. Experimentally, we demonstrate that the resulting representation captures the space of feature count combinations more accurately than the traditional models, not only when the input images come from a panning camera, but even when modeling images of different scenes from the same category.

In text analysis documents are represented as disorganized bags of words, models of count features are typically based on mixing a small number of topics \cite{lda,sam}. Recently, it has been observed that for many text corpora documents evolve into one another in a smooth way, with some features dropping and new ones being introduced. The counting grid \cite{cgUai} models this spatial metaphor literally: it is multidimensional grid of word distributions learned in such a way that a document's own distribution of features can be modeled as the sum of the histograms found in a window into the grid. The major drawback of this method is that it is essentially a mixture and all the content much be generated by a single contiguous area on the grid. This may be problematic especially for lower dimensional grids. In this paper, we overcome to this issue with the \emph{Componential Counting Grid} which brings the componential nature of topic models to the basic counting grid. We also introduce a generative kernel based on the document's grid usage and a visualization strategy useful for understanding large text corpora. We evaluate our approach on document classification and multimodal retrieval obtaining state of the art results on standard benchmarks.

The Lasso is a cornerstone of modern multivariate data analysis, yet its performance suffers in the common situation in which covariates are correlated. This limitation has led to a growing number of \emph{Preconditioned Lasso} algorithms that pre-multiply $X$ and $y$ by matrices $P_X$, $P_y$ prior to running the standard Lasso. A direct comparison of these and similar Lasso-style algorithms to the original Lasso is difficult because the performance of all of these methods depends critically on an auxiliary penalty parameter $\lambda$. In this paper we propose an agnostic, theoretical framework for comparing Preconditioned Lasso algorithms to the Lasso without having to choose $\lambda$. We apply our framework to three Preconditioned Lasso instances and highlight when they will outperform the Lasso. Additionally, our theory offers insights into the fragilities of these algorithms to which we provide partial solutions.

Mixtures of Gaussians, factor analyzers (probabilistic PCA) and hidden Markov models are staples of static and dynamic data modeling and image and video modeling in particular. We show how topographic transformations in the input, such as translation and shearing in images, can be accounted for in these models by including a discrete transformation variable. The resulting models perform clustering, dimensionality reduction and time-series analysis in a way that is invariant to transformations in the input. Using the EM algorithm, these transformation-invariant models can be fit to static data and time series. We give results on filtering microscopy images, face and facial pose clustering, handwritten digit modeling and recognition, video clustering, object tracking, and removal of distractions from video sequences.

Models of bags of words typically assume topic mixing so that the words in a single bag come from a limited number of topics. We show here that many sets of bag of words exhibit a very different pattern of variation than the patterns that are efficiently captured by topic mixing. In many cases, from one bag of words to the next, the words disappear and new ones appear as if the theme slowly and smoothly shifted across documents (providing that the documents are somehow ordered). Examples of latent structure that describe such ordering are easily imagined. For example, the advancement of the date of the news stories is reflected in a smooth change over the theme of the day as certain evolving news stories fall out of favor and new events create new stories. Overlaps among the stories of consecutive days can be modeled by using windows over linearly arranged tight distributions over words. We show here that such strategy can be extended to multiple dimensions and cases where the ordering of data is not readily obvious. We demonstrate that this way of modeling covariation in word occurrences outperforms standard topic models in classification and prediction tasks in applications in biology, text modeling and computer vision.