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

Reinforcement learning, one of the most active research areas in artificial intelligence, is a computational approach to learning whereby an agent tries to maximize the total amount of reward it receives when interacting with a complex, uncertain environment. In Reinforcement Learning, Richard Sutton and Andrew Barto provide a clear and simple account of the key ideas and algorithms of reinforcement learning. Their discussion ranges from the history of the field's intellectual foundations to the most recent developments and applications. The only necessary mathematical background is familiarity with elementary concepts of probability. The book is divided into three parts. Part I defines the reinforcement learning problem in terms of Markov decision processes. Part II provides basic solution methods: dynamic programming, Monte Carlo methods, and temporal-difference learning. Part III presents a unified view of the solution methods and incorporates artificial neural networks, eligibility traces, and planning; the two final chapters present case studies and consider the future of reinforcement learning.

/pdf/reinforcement-learning-an-introduction-rzxgej9p17.pdf

Reinforcement Learning: An Introduction

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 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

In this thesis, I discuss two probabilistic modeling problems arising in metazoan genomic analysis: identifying motifs and cis-regulatory modules (CRMs) from transcriptional regulatory sequences, and inferring haplotypes from genotypes of single nucleotide polymorphisms. Motif and CRM identification is important for understanding the gene regulatory network underlying metazoan development and functioning. I discuss a modular Bayesian model that captures rich structural characteristics of the transcriptional regulatory sequences and supports a variety of motif detection tasks. Haplotype inference is essential for the understanding of genetic variation within and among populations, with important applications to the genetic analysis of disease propensities. I discuss a Bayesian model based on a prior distribution constructed from a Dirichlet process—a nonparametric prior which provides control over the size of the unknown pool of population haplotypes, and on a likelihood function that allows statistical errors in the haplotype/genotype relationship. Our models use the “probabilistic graphical model” formalism, a formalism that exploits the conjoined capabilities of graph theory and probability theory to build complex models out of simpler pieces. I discuss the mathematical underpinnings for the models, how they formally incorporate biological prior knowledge about the data, and I present a generalized mean field theory and a generic algorithm for approximate inference on such models.

Probabilistic graphical models and algorithms for genomic analysis

Active learning with mixture models

Statistical debugging is a combination of statistical machine learning and software debugging. Given sampled run-time profiles from both successful and failed runs, our task is to select a small set of program predicates that can succinctly capture the failure modes, thereby leading to the locations of the bugs. Given the diverse nature of software bugs and coding structure, this is not a trivial task. 
We start by assuming that there is only one bug in the program. This allows us to concentrate on the problem of non-deterministic bugs. We design a utility function whose components may be adjusted based on the suspected level of determinism of the bug. The algorithm proves to work well on two real world programs. 
The problems becomes much more complicated once we do away with the single-bug assumption. The original single-bug algorithm does not perform well in the presence of multiple bugs. Our initial attempts at clustering fall short of an effective solution. After identifying the main problems in the multi-bug case, we present an iterative predicate scoring algorithm. We demonstrate the algorithm at work on five real world programs, where it successfully clusters runs and identifies important predicates that clearly point to many of the underlying bugs.

/pdf/statistical-software-debugging-572u1zuhy0.pdf

Statistical software debugging

Inspired by the success of Convolutional Neural Networks (CNNs) for supervised prediction in images, we design the Deconvolutional Generative Model (DGM), a new probabilistic generative model whose inference calculations correspond to those in a given CNN architecture. The DGM uses a CNN to design the prior distribution in the probabilistic model. Furthermore, the DGM generates images from coarse to finer scales. It introduces a small set of latent variables at each scale, and enforces dependencies among all the latent variables via a conjugate prior distribution. This conjugate prior yields a new regularizer based on paths rendered in the generative model for training CNNs-the Rendering Path Normalization (RPN). We demonstrate that this regularizer improves generalization, both in theory and in practice. In addition, likelihood estimation in the DGM yields training losses for CNNs, and inspired by this, we design a new loss termed as the Max-Min cross entropy which outperforms the traditional cross-entropy loss for object classification. The Max-Min cross entropy suggests a new deep network architecture, namely the Max-Min network, which can learn from less labeled data while maintaining good prediction performance. Our experiments demonstrate that the DGM with the RPN and the Max-Min architecture exceeds or matches the-state-of-art on benchmarks including SVHN, CIFAR10, and CIFAR100 for semi-supervised and supervised learning tasks.

A Bayesian Perspective of Convolutional Neural Networks through a Deconvolutional Generative Model

Modern approaches to data analysis make extensive use of data-driven model selection. The resulting dependencies between the selected model and data used for inference invalidate statistical guarantees derived from classical theories. The framework of post-selection inference (PoSI) has formalized this problem and proposed corrections which ensure valid inferences. Yet, obtaining general principles that enable computationally-efficient, powerful PoSI methodology with formal guarantees remains a challenge. With this goal in mind, we revisit the PoSI problem through the lens of algorithmic stability. Under an appropriate formulation of stability---one that captures closure under post-processing and compositionality properties---we show that stability parameters of a selection method alone suffice to provide non-trivial corrections to classical z-test and t-test intervals. Then, for several popular model selection methods, including the LASSO, we show how stability can be achieved through simple, computationally efficient randomization schemes. Our algorithms offer provable unconditional simultaneous coverage and are computationally efficient; in particular, they do not rely on MCMC sampling. Importantly, our proposal explicitly relates the magnitude of randomization to the resulting confidence interval width, allowing the analyst to tune interval width to the loss in utility due to randomizing selection.

Michael I. Jordan

Papers

Probabilistic graphical models and algorithms for genomic analysis

Active learning with mixture models

Statistical software debugging

A Bayesian Perspective of Convolutional Neural Networks through a Deconvolutional Generative Model

Post-Selection Inference via Algorithmic Stability