We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

Deep learning is a form of machine learning that enables computers to learn from experience and understand the world in terms of a hierarchy of concepts. Because the computer gathers knowledge from experience, there is no need for a human computer operator to formally specify all the knowledge that the computer needs. The hierarchy of concepts allows the computer to learn complicated concepts by building them out of simpler ones; a graph of these hierarchies would be many layers deep. This book introduces a broad range of topics in deep learning. The text offers mathematical and conceptual background, covering relevant concepts in linear algebra, probability theory and information theory, numerical computation, and machine learning. It describes deep learning techniques used by practitioners in industry, including deep feedforward networks, regularization, optimization algorithms, convolutional networks, sequence modeling, and practical methodology; and it surveys such applications as natural language processing, speech recognition, computer vision, online recommendation systems, bioinformatics, and videogames. Finally, the book offers research perspectives, covering such theoretical topics as linear factor models, autoencoders, representation learning, structured probabilistic models, Monte Carlo methods, the partition function, approximate inference, and deep generative models. Deep Learning can be used by undergraduate or graduate students planning careers in either industry or research, and by software engineers who want to begin using deep learning in their products or platforms. A website offers supplementary material for both readers and instructors.

Deep Learning

We describe latent Dirichlet allocation (LDA), a generative probabilistic model for collections of discrete data such as text corpora. LDA is a three-level hierarchical Bayesian model, in which each item of a collection is modeled as a finite mixture over an underlying set of topics. Each topic is, in turn, modeled as an infinite mixture over an underlying set of topic probabilities. In the context of text modeling, the topic probabilities provide an explicit representation of a document. We present efficient approximate inference techniques based on variational methods and an EM algorithm for empirical Bayes parameter estimation. We report results in document modeling, text classification, and collaborative filtering, comparing to a mixture of unigrams model and the probabilistic LSI model.

/pdf/latent-dirichlet-allocation-4gmscb06en.pdf

Latent dirichlet allocation

We present a new technique called “t-SNE” that visualizes high-dimensional data by giving each datapoint a location in a two or three-dimensional map. The technique is a variation of Stochastic Neighbor Embedding (Hinton and Roweis, 2002) that is much easier to optimize, and produces significantly better visualizations by reducing the tendency to crowd points together in the center of the map. t-SNE is better than existing techniques at creating a single map that reveals structure at many different scales. This is particularly important for high-dimensional data that lie on several different, but related, low-dimensional manifolds, such as images of objects from multiple classes seen from multiple viewpoints. For visualizing the structure of very large datasets, we show how t-SNE can use random walks on neighborhood graphs to allow the implicit structure of all of the data to influence the way in which a subset of the data is displayed. We illustrate the performance of t-SNE on a wide variety of datasets and compare it with many other non-parametric visualization techniques, including Sammon mapping, Isomap, and Locally Linear Embedding. The visualizations produced by t-SNE are significantly better than those produced by the other techniques on almost all of the datasets.

/pdf/visualizing-data-using-t-sne-2w1n304iq0.pdf

Visualizing Data using t-SNE

We propose a generative model for text and other collections of discrete data that generalizes or improves on several previous models including naive Bayes/unigram, mixture of unigrams [6], and Hof-mann's aspect model, also known as probabilistic latent semantic indexing (pLSI) [3]. In the context of text modeling, our model posits that each document is generated as a mixture of topics, where the continuous-valued mixture proportions are distributed as a latent Dirichlet random variable. Inference and learning are carried out efficiently via variational algorithms. We present empirical results on applications of this model to problems in text modeling, collaborative filtering, and text classification.

/pdf/latent-dirichlet-allocation-rkwbqihm3p.pdf

Latent Dirichlet Allocation

When factorized approximations are used for variational inference (VI), they tend to underestimate the uncertainty -- as measured in various ways -- of the distributions they are meant to approximate. We consider two popular ways to measure the uncertainty deficit of VI: (i) the degree to which it underestimates the componentwise variance, and (ii) the degree to which it underestimates the entropy. To better understand these effects, and the relationship between them, we examine an informative setting where they can be explicitly (and elegantly) analyzed: the approximation of a Gaussian,~$p$, with a dense covariance matrix, by a Gaussian,~$q$, with a diagonal covariance matrix. We prove that $q$ always underestimates both the componentwise variance and the entropy of $p$, \textit{though not necessarily to the same degree}. Moreover we demonstrate that the entropy of $q$ is determined by the trade-off of two competing forces: it is decreased by the shrinkage of its componentwise variances (our first measure of uncertainty) but it is increased by the factorized approximation which delinks the nodes in the graphical model of $p$. We study various manifestations of this trade-off, notably one where, as the dimension of the problem grows, the per-component entropy gap between $p$ and $q$ becomes vanishingly small even though $q$ underestimates every componentwise variance by a constant multiplicative factor. We also use the shrinkage-delinkage trade-off to bound the entropy gap in terms of the problem dimension and the condition number of the correlation matrix of $p$. Finally we present empirical results on both Gaussian and non-Gaussian targets, the former to validate our analysis and the latter to explore its limitations.

The Shrinkage-Delinkage Trade-off: An Analysis of Factorized Gaussian Approximations for Variational Inference

We consider when a sparse nonnegative matrix S can be recovered, via an elementwise non-linearity, from a real-valued matrix L of significantly lower rank. Of particular interest is the setting where the positive elements of S encode the similarities of nearby points on a low dimensional manifold. The recovery can then be posed as a problem in manifold learning— in this case, how to learn a norm-preserving and neighborhood-preserving mapping of high dimensional inputs into a lower dimensional space. We describe an algorithm for this problem based on a generalized low-rank decomposition of sparse matrices. This decomposition has the interesting property that it can be encoded by a neural network with one layer of rectified linear units; since the algorithm discovers this encoding, it can be viewed as a layerwise primitive for deep learning. The algorithm regards the inputs x i and x j as similar whenever the cosine of the angle between them exceeds some threshold τ ∈ (0 , 1). Given this threshold, the algorithm attempts to discover a mapping x i 7→ y i by matching the elements of two sparse matrices; in particular, it seeks a mapping for which S = max(0 , L ), where S ij = max(0 , x i · x j − τ ∥ x i ∥∥ x j ∥ ) and L ij = y i · y j − τ ∥ y i ∥∥ y j ∥ . We apply the algorithm to data sets where vector magnitudes and small cosine distances have interpretable meanings (e.g., the brightness of an image, the similarity to other words). On these data sets, the algorithm discovers much lower dimensional representations that preserve these meanings.

A geometrical connection between sparse and low-rank matrices and its application to manifold learning

We investigate a latent variable model for multinomial classification inspired by recent capsule architectures for visual object recognition (Sabour, Frosst, & Hinton, 2017). Capsule architectures use vectors of hidden unit activities to encode the pose of visual objects in an image, and they use the lengths of these vectors to encode the probabilities that objects are present. Probabilities from different capsules can also be propagated through deep multilayer networks to model the part-whole relationships of more complex objects. Notwithstanding the promise of these networks, there still remains much to understand about capsules as primitive computing elements in their own right. In this letter, we study the problem of capsule regression-a higher-dimensional analog of logistic, probit, and softmax regression in which class probabilities are derived from vectors of competing magnitude. To start, we propose a simple capsule architecture for multinomial classification: the architecture has one capsule per class, and each capsule uses a weight matrix to compute the vector of hidden unit activities for patterns it seeks to recognize. Next, we show how to model these hidden unit activities as latent variables, and we use a squashing nonlinearity to convert their magnitudes as vectors into normalized probabilities for multinomial classification. When different capsules compete to recognize the same pattern, the squashing nonlinearity induces nongaussian terms in the posterior distribution over their latent variables. Nevertheless, we show that exact inference remains tractable and use an expectation-maximization procedure to derive least-squares updates for each capsule's weight matrix. We also present experimental results to demonstrate how these ideas work in practice.

An EM Algorithm for Capsule Regression.

We propose a statistical model for automatic speech recognition that relates variations in speaking rate to nonlinear warpings of time. The model describes a discrete random variable, s(t), that evolves as a function of the arc length traversed along a curve, parameterized by x(t). Since arc length does not depend on the rate at which a curve is traversed, this evolution gives rise to a family of Markov processes whose predictions, Pr[sjx], are invariant to nonlinear warpings of time. We describe the use of such models, known as Markov processes on curves (MPCs), for automatic speech recognition, where x are acoustic feature trajectories and s are phonetic transcriptions. On two tasks|recognizing New Jersey town names and connected alpha-digits|we nd that MPCs yield lower word error rates than comparably trained hid-

Modeling the rate of speech by Markov processes on curves.

A method and apparatus for speech recognition using Markov processes on curves are presented. The method and apparatus operate such that input speech utterances are received and represented as multidimensional curves. The curve is split into acoustic segments representing different components based on initial model estimates. The segments are used to create a new statistical model for the curve. The process may be reiterated to produce a more precise statistical model for recognition. As a result, feature vectors are extracted from input speech and contribute to a recognition score in proportion to their arc length. The arc lengths are weighted to minimize recognition errors due to variations in speaking rate. In addition, more importance is attached to short-lived but non-stationary sounds, such as consonants.

Lawrence K. Saul

Papers

The Shrinkage-Delinkage Trade-off: An Analysis of Factorized Gaussian Approximations for Variational Inference

A geometrical connection between sparse and low-rank matrices and its application to manifold learning

An EM Algorithm for Capsule Regression.

Modeling the rate of speech by Markov processes on curves.

Automatic speech recognition using multi-dimensional curve-linear representations