DOCSLIB.ORG

Unsupervised Representation Learning with Autoencoders By

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Unsupervised Representation Learning with Autoencoders By Unsupervised Representation Learning with Autoencoders by Alireza Makhzani A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Electrical and Computer Engineering University of Toronto c Copyright 2018 by Alireza Makhzani Abstract Unsupervised Representation Learning with Autoencoders Alireza Makhzani Doctor of Philosophy Graduate Department of Electrical and Computer Engineering University of Toronto 2018 Despite the recent progress in machine learning and deep learning, unsupervised learning still remains a largely unsolved problem. It is widely recognized that unsupervised learning algorithms that can learn useful representations are needed for solving problems with limited label information. In this thesis, we study the problem of learning unsupervised representations using autoencoders, and propose regularization techniques that enable autoencoders to learn useful representations of data in unsupervised and semi-supervised settings. First, we exploit sparsity as a generic prior on the representations of autoencoders and propose sparse autoencoders that can learn sparse representations with very fast infer- ence processes, making them well-suited to large problem sizes where conventional sparse coding algorithms cannot be applied. Next, we study autoencoders from a probabilistic perspective and propose generative autoencoders that use a generative adversarial network (GAN) to match the distribution of the latent code of the autoencoder with a pre-defined prior. We show that these generative autoencoders can learn posterior approximations that are more expressive than tractable densities often used in variational inference. We demonstrate the performance of developed methods of this thesis on real world image datasets and show their applications in generative modeling, clustering, semi-supervised classification and dimensionality reduction. ii To my beloved parents and sister: Nasrin, Hassan, and Parastesh. iii Acknowledgements Pursuing a PhD was one of the best decisions of my life. I wish to express my sincerest thanks to my advisor, Brendan Frey. I thank Brendan for his continuous support and encouragement, for believing in the work I wanted to pursue, and for being a great inspiration in my life. Brendan is a teacher and mentor with an unmatched combination of intellect, intuition and wit. I was truly blessed to have him as my advisor. I am very grateful to all my committee members, Rich Zemel, Raquel Urtasun, David Duvenaud and Graham Taylor for offering me their thoughtful comments and feedback. I would like to thank all the past and present members of the PSI lab and the Machine Learning group at U of T, especially Babak Alipanahi, Andrew Delong, Christopher Srinivasa, Jimmy Ba, Hannes Bretschneider, Alice Gao, Hui Xiong, Leo Lee, Michael Leung, and Oren Kraus for sharing ideas and collaborating with me. During my PhD, I did two intenrships at Google, both of which were a wonderful learning experience for me. I would like to thank all the members of Google Brain team and Google DeepMind team, especially Oriol Vinyals, Jon Shlens, Navdeep Jaitly, Ian Goodfellow, Ilya Sutskever, Timothy Lillicrap, Ali Eslami, Sam Bowman and Jon Gauthier. I would like to especially thank Alireza Moghaddamjoo and Hamid Sheikhzadeh Nadjar for inspiring me to pursue academic research and working with me during my undergraduate years at Amirkabir University of Technology in Iran. I had the pleasure of going through my PhD journey with many great friends. In particular, I would like to thank Sadegh Jalali, Aynaz Vatankhah, Masoud Barekatain, Amin Heidari, Weria Havary-Nassab, David Jorjani, Parisa Zareapour, Ehsan Shojaei, Siavash Fazeli and Mohammad Norouzi. I take this opportunity to especially thank Nasrin Tehrani and Hamid Emami. I felt home in Canada thanks to their continuous support during the past years. My deepest gratitude and love, of course, belong to my parents, Nasrin and Hassan, and my sister, Parastesh, for their unconditional love and support throughout my life. To them I owe all that I am and all that I have ever accomplished. iv Contents 1 Introduction1 1.1 Overview . .1 1.2 Unsupervised Representation Learning . .2 1.3 Contributions and Outline . 10 2 k-Sparse Autoencoders 13 2.1 Introduction . 13 2.2 k-Sparse Autoencoders . 13 2.3 Analysis of k-Sparse Autoencoders . 15 2.4 Experiments . 19 2.5 Conclusion . 26 3 Winner-Take-All Autoencoders 27 3.1 Introduction . 27 3.2 Fully-Connected Winner-Take-All Autoencoders . 27 3.3 Convolutional Winner-Take-All Autoencoders . 30 3.4 Experiments . 35 3.5 Discussion . 38 3.6 Conclusion . 40 3.7 Appendix . 41 4 Adversarial Autoencoders 44 4.1 Introduction . 44 4.2 Adversarial Autoencoders . 45 4.3 Likelihood Analysis of Adversarial Autoencoders . 52 4.4 Supervised Adversarial Autoencoders . 53 4.5 Semi-Supervised Adversarial Autoencoders . 55 4.6 Clustering with Adversarial Autoencoders . 57 v 4.7 Dimensionality Reduction with Adversarial Autoencoders . 58 4.8 Conclusion . 61 4.9 Appendix . 63 5 PixelGAN Autoencoders 66 5.1 Introduction . 66 5.2 PixelGAN Autoencoders . 67 5.3 Experiments . 75 5.4 Learning Cross-Domain Relations with PixelGAN Autoencoders . 78 5.5 Conclusion . 80 5.6 Appendix . 81 6 Conclusions 87 6.1 Summary of Contributions . 87 6.2 Future Directions . 88 Bibliography 90 vi Chapter 1 Introduction 1.1 Overview The goal of the field of Artificial Intelligence is to enable computers to understand the rich world around us, and to interact with it in an intelligent manner. More recently, “deep learning” has emerged as one of the most promising approaches to achieve this goal. Deep learning is a way of computational learning of high-level concepts in data using deep, hierarchical neural networks. Emerging deep learning methods have revolutionized many real world applications such as computer vision [1], speech processing [2], and computational biology [3, 4] by achieving breakthrough performance on different tasks. In speech recognition for example, deep learning is the only technology that was able to eclipse the well-established 10-year old state of the art benchmarks. Recently, supervised neural networks have been developed and used successfully to produce representations that have enabled leaps forward in classification accuracy for several tasks [1]. These networks are often trained using convolutional architectures with the new regularization techniques that have been developed in the deep learning community such as dropout [5] or maxout [6]. Despite the recent progress in supervised representation learning, the question that has remained unanswered is whether it is possible to learn as “powerful” representations from unlabeled data without any supervision. It is still widely recognized that unsupervised learning algorithms that can extract useful features are needed for solving problems with limited or weak label information, such as video recognition or pedestrian detection [7]. An advantage of unsupervised learning algorithms is the ability to use them in semi-supervised scenarios where the amount of labeled data is limited. 1 Chapter 1. Introduction 2 1.2 Unsupervised Representation Learning In parallel with research on discriminative architectures, neural networks trained in an unsupervised fashion have been developed and used to produce useful representations for several tasks such as clustering and classification [8, 9, 10, 11, 12]. These approaches include stacked autoencoders, deep Boltzmann machines [8], deep belief networks [9], variational autoencoders [10] and generative adversarial networks [11]. In this section, we briefly review these approaches by broadly categorizing them into three types of learning algorithms: autoencoders, sparse models and generative models. For a more comprehensive introduction to unsupervised representation learning, we recommend the Deep Learning book [13]. 1.2.1 Autoencoders An autoencoder is a neural network that learns to reconstruct its original input with the aim of learning useful representations. The input vector x is first mapped to a hidden representation z = f(x). The function f is parametrized by the encoder network using one or more layers of non-linearity. The hidden representation z is then mapped to the output x^ = g(z) using the decoder network which parametrizes the decoder function g. Similar to the encoder network, the decoder network can consist of multiple layers of non-linearity. The parameters are optimized to minimize the following cost function: 2 L(x; g(f(x))) = kx^ − xk2 (1.1) The cost function of Equation 1.1 is usually optimized with an additional constraint to limit the representational capacity of the autoencoder in order to prevent the autoencoder from learning the useless identity function. The constraint could be imposed on the architecture of the autoencoder (e.g., limiting the latent code dimensionality) or could be in the form of a regularization term in the final objective (e.g., sparsity constraint). Undercomplete Autoencoders One way of leaning useful representations with autoencoders is to limit the capacity of the autoencoder by constraining its code size. In this case, the autoencoder is forced to extract the salient feature of the data. Indeed, it can be shown that if an autoencoder uses the linear activation along with the mean squared error criterion, the resulting architecture is equivalent to the PCA algorithm (i.e., the hidden units of the autoencoder Chapter 1. Introduction 3 learn the principal components of the data).
Load more

Recommended publications
  • Radford M. Neal Curriculum Vitae, 25 April 2021
    Radford M. Neal Curriculum Vitae, 25 April 2021 Biographical Information Personal Data Date of Birth: 12 Sept 1956 Citizenship: Canadian Home address: University address: 21 Shaftesbury Avenue, #106 Department of Statistical Sciences Toronto, Ontario M4T 3B4 University of Toronto M5S 3G3 U of T web page: http://www.cs.utoronto.ca/∼radford/ Blog (not affiliated with U of T): https://radfordneal.wordpress.com E-mail addresses: [email protected], [email protected] Degrees Awarded BSc (Honours) in Computer Science, University of Calgary, 1977 MSc in Computer Science, University of Calgary, 1980 Thesis title: An Editor for Trees Supervisor: David Hill Ph.D in Computer Science, University of Toronto, 1995 Thesis title: Bayesian Learning for Neural Networks Supervisor: Geoffrey Hinton Employment and Other Relevant Experience Present position: Professor Emeritus, Department of Statistical Sciences and Department of Computer Science, University of Toronto, Appointed as Lecturer 1 July 1994. Appointed as Associate Member (Limited Term) of the graduate faculty 13 January 1995. Promoted to Assistant Professor 1 July 1995. Promoted to Associate Professor and full member of the graduate faculty 1 July 1999. Promoted fo Full Professor 1 July 2001. Awarded a Canada Research Chair in Statistics and Machine Learning 1 July 2003 (held to 31 December 2016). Cross-appointed to the Dalla Lana School of Public Health 14 February 2006. Retired as of 1 January 2017. Earlier teaching experience: Sessional instructor in computer science at the University of Calgary for the following courses: 1 Summers 1979-1981 Taught half of the core second-year course on machine archi- tecture and programming.
    [Show full text]
  • An Empirical Study on Evaluation Metrics of Generativeadversarialnetworks
    Under review as a conference paper at ICLR 2018 AN EMPIRICAL STUDY ON EVALUATION METRICS OF GENERATIVE ADVERSARIAL NETWORKS Anonymous authors Paper under double-blind review ABSTRACT Despite the widespread interest in generative adversarial networks (GANs), few works have studied the metrics that quantitatively evaluate GANs’ performance. In this paper, we revisit several representative sample-based evaluation metrics for GANs, and address the important problem of how to evaluate the evaluation metrics. We start with a few necessary conditions for metrics to produce meaningful scores, such as distinguishing real from generated samples, identifying mode dropping and mode collapsing, and detecting overfitting. Then with a series of carefully designed experiments, we are able to comprehensively investigate existing sample-based metrics and identify their strengths and limitations in practical settings. Based on these results, we observe that kernel Maximum Mean Discrepancy (MMD) and the 1-Nearest-Neighbour (1-NN) two-sample test seem to satisfy most of the desirable properties, provided that the distances between samples are computed in a suitable feature space. Our experiments also unveil interesting properties about the behavior of several popular GAN models, such as whether they are memorizing training samples, and how far these state-of-the-art GANs are from perfect. 1 INTRODUCTION Generative adversarial networks (GANs) (Goodfellow et al., 2014) have been studied extensively in recent years. Besides producing surprisingly plausible images of faces (Radford et al., 2015; Larsen et al., 2015) and bedrooms (Radford et al., 2015; Arjovsky et al., 2017; Gulrajani et al., 2017), they have also been innovatively applied in, for example, semi-supervised learning (Odena, 2016; Makhzani et al., 2015), image-to-image translation (Isola et al., 2016; Zhu et al., 2017), and simulated image refinement (Shrivastava et al., 2016).
    [Show full text]
  • Uncertainty in Deep Learning
    References Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jefrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geofrey Irving, Michael Isard, Yangqing Jia, Rafal Jozefowicz, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dan Mané, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, Mike Schuster, Jonathon Shlens, Benoit Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Viégas, Oriol Vinyals, Pete Warden, Martin Wattenberg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. TensorFlow: Large-scale machine learning on heterogeneous systems, 2015. URL http://tensorĆow.org/. Software available from tensorĆow.org. Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, and Dan Mane. Concrete problems in ai safety. arXiv preprint arXiv:1606.06565, 2016. C Angermueller and O Stegle. Multi-task deep neural network to predict CpG methylation proĄles from low-coverage sequencing data. In NIPS MLCB workshop, 2015. O Anjos, C Iglesias, F Peres, J Martínez, Á García, and J Taboada. Neural networks applied to discriminate botanical origin of honeys. Food chemistry, 175:128Ű136, 2015. Christopher Atkeson and Juan Santamaria. A comparison of direct and model-based reinforcement learning. In In International Conference on Robotics and Automation. Citeseer, 1997. Jimmy Ba and Brendan Frey. Adaptive dropout for training deep neural networks. In Advances in Neural Information Processing Systems, pages 3084Ű3092, 2013. P Baldi, P Sadowski, and D Whiteson. Searching for exotic particles in high-energy physics with deep learning. Nature communications, 5, 2014. Pierre Baldi and Peter J Sadowski. Understanding dropout. In Advances in Neural Information Processing Systems, pages 2814Ű2822, 2013.
    [Show full text]
  • Adaptive Dropout for Training Deep Neural Networks
    Adaptive dropout for training deep neural networks Lei Jimmy Ba Brendan Frey Department of Electrical and Computer Engineering University of Toronto jimmy, [email protected] Abstract Recently, it was shown that deep neural networks can perform very well if the activities of hidden units are regularized during learning, e.g, by randomly drop- ping out 50% of their activities. We describe a method called ‘standout’ in which a binary belief network is overlaid on a neural network and is used to regularize of its hidden units by selectively setting activities to zero. This ‘adaptive dropout network’ can be trained jointly with the neural network by approximately com- puting local expectations of binary dropout variables, computing derivatives using back-propagation, and using stochastic gradient descent. Interestingly, experi- ments show that the learnt dropout network parameters recapitulate the neural network parameters, suggesting that a good dropout network regularizes activities according to magnitude. When evaluated on the MNIST and NORB datasets, we found that our method achieves lower classification error rates than other feature learning methods, including standard dropout, denoising auto-encoders, and re- stricted Boltzmann machines. For example, our method achieves 0.80% and 5.8% errors on the MNIST and NORB test sets, which is better than state-of-the-art results obtained using feature learning methods, including those that use convolu- tional architectures. 1 Introduction For decades, deep networks with broad hidden layers and full connectivity could not be trained to produce useful results, because of overfitting, slow convergence and other issues. One approach that has proven to be successful for unsupervised learning of both probabilistic generative models and auto-encoders is to train a deep network layer by layer in a greedy fashion [7].
    [Show full text]
  • Time-Series Generative Adversarial Networks
    Time-series Generative Adversarial Networks Jinsung Yoon∗ Daniel Jarrett∗ University of California, Los Angeles, USA University of Cambridge, UK [email protected] [email protected] Mihaela van der Schaar University of Cambridge, UK University of California, Los Angeles, USA Alan Turing Institute, UK [email protected], [email protected] Abstract A good generative model for time-series data should preserve temporal dynamics, in the sense that new sequences respect the original relationships between variables across time. Existing methods that bring generative adversarial networks (GANs) into the sequential setting do not adequately attend to the temporal correlations unique to time-series data. At the same time, supervised models for sequence prediction—which allow finer control over network dynamics—are inherently deterministic. We propose a novel framework for generating realistic time-series data that combines the flexibility of the unsupervised paradigm with the control afforded by supervised training. Through a learned embedding space jointly optimized with both supervised and adversarial objectives, we encourage the network to adhere to the dynamics of the training data during sampling. Empirically, we evaluate the ability of our method to generate realistic samples using a variety of real and synthetic time-series datasets. Qualitatively and quantitatively, we find that the proposed framework consistently and significantly outperforms state-of-the-art benchmarks with respect to measures of similarity and predictive ability. 1 Introduction What is a good generative model for time-series data? The temporal setting poses a unique challenge to generative modeling. A model is not only tasked with capturing the distributions of features within each time point, it should also capture the potentially complex dynamics of those variables across time.
    [Show full text]
  • Top 100 AI Leaders in Drug Discovery and Advanced Healthcare Introduction
    Top 100 AI Leaders in Drug Discovery and Advanced Healthcare www.dka.global Introduction Over the last several years, the pharmaceutical and healthcare organizations have developed a strong interest toward applying artificial intelligence (AI) in various areas, ranging from medical image analysis and elaboration of electronic health records (EHRs) to more basic research like building disease ontologies, preclinical drug discovery, and clinical trials. The demand for the ML/AI technologies, as well as for ML/AI talent, is growing in pharmaceutical and healthcare industries and driving the formation of a new interdisciplinary industry (‘data-driven healthcare’). Consequently, there is a growing number of AI-driven startups and emerging companies offering technology solutions for drug discovery and healthcare. Another important source of advanced expertise in AI for drug discovery and healthcare comes from top technology corporations (Google, Microsoft, Tencent, etc), which are increasingly focusing on applying their technological resources for tackling health-related challenges, or providing technology platforms on rent bases for conducting research analytics by life science professionals. Some of the leading pharmaceutical giants, like GSK, AstraZeneca, Pfizer and Novartis, are already making steps towards aligning their internal research workflows and development strategies to start embracing AI-driven digital transformation at scale. However, the pharmaceutical industry at large is still lagging behind in adopting AI, compared to more traditional consumer industries -- finance, retail etc. The above three main forces are driving the growth in the AI implementation in pharmaceutical and advanced healthcare research, but the overall success depends strongly on the availability of highly skilled interdisciplinary leaders, able to innovate, organize and guide in this direction.
    [Show full text]
  • Gaussian Error Linear Units (Gelus)
    GAUSSIAN ERROR LINEAR UNITS (GELUS) Dan Hendrycks∗ Kevin Gimpel University of California, Berkeley Toyota Technological Institute at Chicago [email protected] [email protected] ABSTRACT We propose the Gaussian Error Linear Unit (GELU), a high-performing neural network activation function. The GELU activation function is xΦ(x), where Φ(x) the standard Gaussian cumulative distribution function. The GELU nonlinearity weights inputs by their value, rather than gates inputs by their sign as in ReLUs (x1x>0). We perform an empirical evaluation of the GELU nonlinearity against the ReLU and ELU activations and find performance improvements across all considered computer vision, natural language processing, and speech tasks. 1 INTRODUCTION Early artificial neurons utilized binary threshold units (Hopfield, 1982; McCulloch & Pitts, 1943). These hard binary decisions are smoothed with sigmoid activations, enabling a neuron to have a “fir- ing rate” interpretation and to train with backpropagation. But as networks became deeper, training with sigmoid activations proved less effective than the non-smooth, less-probabilistic ReLU (Nair & Hinton, 2010) which makes hard gating decisions based upon an input’s sign. Despite having less of a statistical motivation, the ReLU remains a competitive engineering solution which often enables faster and better convergence than sigmoids. Building on the successes of ReLUs, a recent modifi- cation called ELUs (Clevert et al., 2016) allows a ReLU-like nonlinearity to output negative values which sometimes increases training speed. In all, the activation choice has remained a necessary architecture decision for neural networks lest the network be a deep linear classifier. Deep nonlinear classifiers can fit their data so well that network designers are often faced with the choice of including stochastic regularizer like adding noise to hidden layers or applying dropout (Sri- vastava et al., 2014), and this choice remains separate from the activation function.
    [Show full text]
  • Deep Learning Meets Genome Biology
    Deep learning meets genome biology oreilly.com/ideas/deep-learning-meets-genome-biology April 26, 2016 As part of our ongoing series of interviews surveying the frontiers of machine intelligence, I recently interviewed Brendan Frey. Frey is a co-founder of Deep Genomics, a professor at the University of Toronto and a co-founder of its Machine Learning Group, a senior fellow of the Neural Computation program at the Canadian Institute for Advanced Research, and a fellow of the Royal Society of Canada. His work focuses on using machine learning to understand the genome and to realize new possibilities in genomic medicine. Key Takeaways The application of deep learning to genomic medicine is off to a promising start; it could impact diagnostics, intensive care, pharmaceuticals and insurance. The “genotype-phenotype divide”—our inability to connect genetics to disease phenotypes—is preventing genomics from advancing medicine to its potential. Deep learning can bridge the genotype-phenotype divide, by incorporating an exponentially growing amount of data, and accounting for the multiple layers of complex biological processes that relate the genotype to the phenotype. Deep learning has been successful in applications where humans are naturally adept, such as image, text, and speech understanding. The human mind, however, isn’t intrinsically designed to understand the genome. This gap necessitates the application of “super-human intelligence” to the problem. Efforts in this space must account for underlying biological mechanisms; overly simplistic, “black box” approaches will drive only limited value. O'Reilly AI Newsletter David Beyer: Let’s start with your background. Brendan Frey: I completed my Ph.D.
    [Show full text]
  • Deep Learning Approaches to Problems in Speech Recognition, Computational Chemistry, and Natural Language Text Processing
    Deep learning approaches to problems in speech recognition, computational chemistry, and natural language text processing by George Edward Dahl A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Computer Science University of Toronto c Copyright 2015 by George Edward Dahl Abstract Deep learning approaches to problems in speech recognition, computational chemistry, and natural language text processing George Edward Dahl Doctor of Philosophy Graduate Department of Computer Science University of Toronto 2015 The deep learning approach to machine learning emphasizes high-capacity, scalable models that learn distributed representations of their input. This dissertation demonstrates the efficacy and generality of this approach in a series of diverse case studies in speech recognition, computational chemistry, and natural language processing. Throughout these studies, I extend and modify the neural network models as needed to be more effective for each task. In the area of speech recognition, I develop a more accurate acoustic model using a deep neural network. This model, which uses rectified linear units and dropout, improves word error rates on a 50 hour broadcast news task. A similar neural network results in a model for molecular activity prediction substantially more effective than production systems used in the pharmaceutical industry. Even though training assays in drug discovery are not typically very large, it is still possible to train very large models by leveraging data from multiple assays in the same model and by using effective regularization schemes. In the area of natural language processing, I first describe a new restricted Boltzmann machine training algorithm suitable for text data.
    [Show full text]
  • Machine Learning for Computer Graphics: a Manifesto and Tutorial
    Machine Learning for Computer Graphics: A Manifesto and Tutorial Aaron Hertzmann University of Toronto www.dgp.toronto.edu/∼hertzman Abstract motions from an actor in a studio. These “pure data” ap- proaches give the highest quality motions, but at substantial I argue that computer graphics can benefit from a deeper cost in time and the efforts of artists or actors. Moreover, use of machine learning techniques. I give an overview of there is little flexibility: if you discover that you did not get what learning has to offer the graphics community, with an just the right motions in the studio, then you have to go back emphasis on Bayesian techniques. I also attempt to ad- and capture more. The situation is worse for a video game, dress some misconceptions about learning, and to give a where one must capture all motions that might conceivably very brief tutorial on Bayesian reasoning. be needed. Learning techniques promise the best of both worlds: starting from some captured data, we can procedurally syn- 1. Introduction thesize more data in the style of the original. Moreover, we can constrain the synthetic data, for example, according to the requirements of an artist. Of course, we must begin We live in an age of widespread exploration of art and with some data produced by an artist or a capture session. communication using computer graphics and animation. But now, we can do much more with this data than just sim- Filmmakers, scientists, graphic designers, fine artists, and ple playback. For these problems, machine learning offers game designers, are finding new ways to communicate and an attractive set of tools for modeling the patterns of data.
    [Show full text]
  • Deep Speech Recognition
    Deep Speech Recognition New-Generation Models & Methodology for Advancing Speech Technology and Information Processing Li Deng Microsoft Research, Redmond, USA IEEE ChinaSIP Summer School, July 6, 2013 (including joint work with colleagues at MSR, U of Toronto, etc.) Outline PART I: Basics of Deep Learning (DL) --- including impact and recent history of DL (Deep Neural Net, DNN) in speech recognition PART II: Deeper Substance of DL --- including connections to other ML paradigms, examples of incorporating speech knowledge in DL architecture, and recent experiments in speech recognition 2 Deep Learning (DL) Basics 1. Deep Learning (aka Deep Structured Learning, Hierarchical Learning): a class of machine learning techniques, where many layers of information processing stages in hierarchical architectures are exploited for unsupervised feature learning and for pattern analysis/classification. 2. Deep belief nets (DBN): probabilistic generative models composed of multiple layers of stochastic, hidden variables. The top two layers have undirected, symmetric connections between them. The lower layers receive top-down, directed connections from the layer above. (key: stacked RBMs; Hinton: Science, 2006) 3. Boltzmann machine (BM): a network of symmetrically connected, neuron-like units that make stochastic decisions about whether to be on or off. 4. Restricted Boltzmann machine (RBM): a special BM consisting of a layer of visible units and a layer of hidden units with no visible-visible or hidden-hidden connections. (Key: contrastive divergence learning) 5. Deep neural nets (DNN, or “DBN” before Nov 2012): multilayer perceptrons with many hidden layers, whose weights are often initialized (pre-trained) using stacked RBMs or DBN (DBN-DNN) or discriminative pre-training.
    [Show full text]
  • A Study of the Mathematics of Deep Learning
    A STUDY OF THE MATHEMATICS OF DEEP LEARNING by Anirbit Mukherjee A dissertation submitted to Johns Hopkins University in conformity with the requirements for the degree of Doctor of Philosophy. Baltimore, Maryland July, 2020 c 2020 Anirbit Mukherjee ⃝ All Rights Reserved This thesis is dedicated to my mother Dr. Suranjana Sur (“Mame”). Long before I had started learning arithmetic in school, my mother got me a geometry box and taught me how to construct various angles using a compass. Years before I had started formally studying any of the natural sciences, she created artificial clouds inside the home and taught me about condensation and designed experiments with plants to teach me how they do respiration. This thesis began with the scientific knowledge that my mother imparted regularly to me from right since I was a kid. ii Thesis Abstract Thesis Abstract ”Deep Learning”/”Deep Neural Nets” is a technological marvel that is now increasingly deployed at the cutting-edge of artificial intelligence tasks. This ongoing revolution can be said to havebeen ignited by the iconic 2012 paper from the University of Toronto titled “ImageNet Classification with Deep Convolutional Neural Networks” by Alex Krizhevsky, Ilya Sutskever and Geoffrey E. Hinton. This paper showed that deep nets can be used to classify images into meaningful categories with almost human-like accuracies! As of 2020 this approach continues to produce unprecedented per- formance for an ever widening variety of novel purposes ranging from playing chess to self-driving cars to experimental astrophysics and high-energy physics. But this new found astonishing success of deep neural nets in the last few years has been hinged on an enormous amount of heuristics and it has turned out to be extremely challenging to be mathematically rigorously explainable.
    [Show full text]