DOCSLIB.ORG

THOR: Trace-Based Hardware-Driven Layer-Oriented Natural Gradient

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
THOR: Trace-Based Hardware-Driven Layer-Oriented Natural Gradient PRELIMINARY VERSION: DO NOT CITE The AAAI Digital Library will contain the published version some time after the conference THOR, Trace-based Hardware-driven layer-ORiented Natural Gradient Descent Computation Mengyun Chen1*, Kai-Xin Gao2*, Xiao-Lei Liu2*, Zidong Wang1*, Ningxi Ni1*, Qian Zhang3∗ Lei Chen4y, Chao Ding5, Zheng-Hai Huang2, Min Wang1, Shuangling Wang1, Fan Yu1, Xinyuan Zhao3, Dachuan Xu3 1Huawei Technologies Co. Ltd, 2Tianjin University, 3Beijing University of Technology 4Hong Kong University of Science and Technology, 5Chinese Academy of Sciences 1chenmengyun1, wang1, niningxi, wangmin106, wangshuangling1, [email protected] 2gaokaixin, liuxiaolei, [email protected], 3zhangqian, xyzhao, [email protected] [email protected], [email protected] Abstract an objective function J(θ) with respect to the parameters θ, i.e., θ is updated as: θ θ − αrθJ(θ), where rθJ(θ) It is well-known that second-order optimizer can accelerate the is gradient, α represents the learning rate. Using SGD to training of deep neural networks, however, the huge compu- tation cost of second-order optimization makes it impractical optimize the parameter training faces two challenges, which to apply in real practice. In order to reduce the cost, many are: 1) it is difficult to choose a proper learning rate, and 2) methods have been proposed to approximate a second-order it is hard to escape saddle points. Therefore, many variants matrix. Inspired by KFAC, we propose a novel Trace-based of SGD such as Momentum [Qian 1999], AdaGrad [Zeiler Hardware-driven layer-ORiented Natural Gradient Descent 2012], Adam [Kingma and Ba 2014] and etc. have been Computation method, called THOR, to make the second-order introduced in the past two decades. Though choosing learning optimization applicable in the real application models. Specif- rate becomes easy in these algorithms, they still cannot escape ically, we gradually increase the update interval and use the saddle points when the object function is non-convex, which matrix trace to determine which blocks of Fisher Informa- is often the case in many real application models. tion Matrix (FIM) need to be updated. Moreover, by resorting To address the challenging issues encountered by SGD, it the power of hardware, we have designed a Hardware-driven approximation method for computing FIM to achieve better is easy to think of using second-order optimizer since it can performance. To demonstrate the effectiveness of THOR, we avoid saddle points and most importantly, it can accelerate have conducted extensive experiments. The results show that convergence by using the curvature information. Specifically, −1 training ResNet-50 on ImageNet with THOR only takes 66.7 the parameters θ are updated by θ θ − αG rθJ(θ), minutes to achieve a top-1 accuracy of 75.9 % under an 8 As- where G−1 is the inverse of second-order information matrix cend 910 environment with MindSpore, a new deep learning G. The definitions of G in different second-order optimiza- computing framework. Moreover, with more computational tion algorithms are not the same. Common second-order op- resources, THOR can only takes 2.7 minutes to 75.9 % with timization algorithms include Newton’s method and natural 256 Ascend 910. gradient method, where their second-order information ma- trix G is Hessian matrix (HM) and Fisher information matrix 1 Introduction (FIM), respectively. The biggest challenge to use second- Recently, deep learning has made significantly progress in order optimizer is its computation increases cubically and various computer vision and natural language applications. space cost increases quadratically compared to SGD. There- However, with the increase of complexity of models, tons of fore, it is quite impractical to compute the inverse of second- parameters needed to be trained. For example, according to order information matrix directly. [Devlin et al. 2018] and [He et al. 2016], training BERT (over To reduce the computation cost of the second-order op- 340 million parameters) and ResNet-50 (over 23 million timizer, quite a few approximation approaches have been trainable parameters) will take around 3 days on 16 TPUv3 proposed. For instance, for Newtons method, Quasi-Newton and 29 hours on 8 Tesla P100 GPU, respectively. Therefore, methods [Nocedal and Wright 2006] can be used to approxi- many efforts have been put to propose optimization solutions mate the inverse of HM. One of the advantages of these meth- to reduce the training time. ods over the classical Newton method is that the HM does Among all the proposed optimization techniques, the most not need to be inverted explicitly. In particular, the Limited- popular and promising one is Stochastic Gradient Descent memory BFGS (L-BGFS) algorithm [Zhu et al. 1997] has (SGD) [Robbins and Monro 1951], which is a first-order been implemented and used to speed up the training process optimization algorithm. Specifically, SGD tries to minimize in Deep Neural Networks (DNN) (e.g., [Le et al. 2011]). Other structured stochastic Quasi-Newton methods are al- ∗Equal contribution. so developed and studied recently in [Keskar and Berahas yCorresponding author 2016, Berahas, Jahani, and Taka´cˇ 2019]. Another class of Newton type second-order methods is the Hessian Free opti- 2 Background and Notations mization method [Martens 2010, Kiros 2013, Pan, Innanen, The purpose of deep neural network training is to find a set and Liao 2017], in which the matrix-free conjugate-gradient of model parameters θ 2 Rn to minimize the loss function (CG) algorithms are used to approximate the true Hessian J(θ). Given the cross-entropy loss function: matrix. However, these CG algorithms usually require lots J(θ) = [− log p(yjx; θ)]; (1) of iterations to reach the desired accuracy, in particular for E ill-condition cases. where x, y are the training input and label, p(yjx; θ) repre- Unlike the Newton type methods, Kronecker-factored Ap- sents the density function of a predictive distribution Pyjx. proximate Curvature (KFAC) [Martens and Grosse 2015, 2.1 The Natural Gradient Grosse and Martens 2016, Martens, Ba, and Johnson 2018] is a second-order method based on natural gradient method. Our algorithm is based on the natural gradient proposed by More precisely, in KFAC, one computes the inverse of the [Amari 1998]. Natural gradient gives the steepest direction FIM by computationally tractable approximations such as of the target function when the parameter space has a Rie- block-diagonal approximation and tridiagonal-block diag- mannian metric structure. In other words, it gives the largest onal approximation. [George et al. 2018] have introduced change of the loss function per unit change of the model. The an Eigenvalue-corrected Kronecker Factorization (EKFAC) distance between the distribution Pθ and Pθ+δθ can be mea- which can approximate FIM much better than KFAC does. sured with the K-L divergence. More recent discussion of the natural gradient can be found in [Martens 2014, Ollivier et al. [Osawa et al. 2019, 2020] have demonstrated that KFAC is −1 2017]. Natural gradient is typically defined as F rθJ(θ), efficient in large-scale distributed computing for deep neural n×n networks. Overall, among all these methods, the approxima- where F 2 R is FIM. With the predictive distribution tion scheme for the inverse of FIM is crucial for improving defined as Pyjx, FIM is formulated as T the efficiency of the second-order optimizer, since the current F = E[rθ log p(yjx; θ)rθ log p(yjx; θ) ]: (2) exact strategies still require significant computing power in It is impractical to compute the inverse of FIM directly in a practice. deep neural network since it has over millions of parameters. To address the issues of inefficient computing FIM, in this paper, we propose an efficient approximation algorithm 2.2 KFAC based on natural gradients, named Trace-based Hardware- KFAC is an efficient method for approximation natural gra- driven layer-ORiented Natural Gradient Descent Computa- dient, which approximates FIM by block-diagonal or block- tion (THOR), to compute FIM. Firstly, we observe from tridiagonal matrices. Based on nice motivation and rigorous experiments that the FIM for each layer usually changes mathematical derivation, it has exquisitely settled the prob- rapidly in the first few iterations and then tends to be stable. lem of complex computation for inverting the second order Therefore, it is reasonable to increase the update interval information matrix. [Osawa et al. 2019] have proved that of the inverse of FIM in a proper manner without the loss block-diagonal KFAC has good results in large-scale DNN of convergence rate. Secondly, we make further decision to and block-diagonal KFAC computes more efficiently than decide which blocks of FIM need to be updated. Thirdly, we block tridiagonal. Thus, we focus on block-diagonal KFAC introduce a new approximation scheme by using a Hardware- to approximate FIM in this work. driven matrix splitting scheme to approximate the FIM, which KFAC is a two-step approximation method. In the first step, can be regarded as finding an optimal tradeoff point between KFAC decomposes the FIM into block matrices according to the computational efficiency and the information loss of FIM. the layers of the neural network, by assuming that parameters Overall, the contributions of our work can be summarized of different layers are independent. Then the calculation of as follows: the inverse of FIM is simplified as the inverse of these small • Under the assumption that the FIM converges to a sta- blocks. In the second step, these block matrices are further tionary distribution, we gradually increase the update interval approximated by the Kronecker product of two much smaller of the inverse of FIM to save the overall computational time. matrices which we call Kronecker factors. Since the inverses • Instead of using the Frobenius norm based updating of the Kronecker product of two matrices are equal to the rule proposed in [Osawa et al.
Load more

Recommended publications
  • Training Autoencoders by Alternating Minimization
    Under review as a conference paper at ICLR 2018 TRAINING AUTOENCODERS BY ALTERNATING MINI- MIZATION Anonymous authors Paper under double-blind review ABSTRACT We present DANTE, a novel method for training neural networks, in particular autoencoders, using the alternating minimization principle. DANTE provides a distinct perspective in lieu of traditional gradient-based backpropagation techniques commonly used to train deep networks. It utilizes an adaptation of quasi-convex optimization techniques to cast autoencoder training as a bi-quasi-convex optimiza- tion problem. We show that for autoencoder configurations with both differentiable (e.g. sigmoid) and non-differentiable (e.g. ReLU) activation functions, we can perform the alternations very effectively. DANTE effortlessly extends to networks with multiple hidden layers and varying network configurations. In experiments on standard datasets, autoencoders trained using the proposed method were found to be very promising and competitive to traditional backpropagation techniques, both in terms of quality of solution, as well as training speed. 1 INTRODUCTION For much of the recent march of deep learning, gradient-based backpropagation methods, e.g. Stochastic Gradient Descent (SGD) and its variants, have been the mainstay of practitioners. The use of these methods, especially on vast amounts of data, has led to unprecedented progress in several areas of artificial intelligence. On one hand, the intense focus on these techniques has led to an intimate understanding of hardware requirements and code optimizations needed to execute these routines on large datasets in a scalable manner. Today, myriad off-the-shelf and highly optimized packages exist that can churn reasonably large datasets on GPU architectures with relatively mild human involvement and little bootstrap effort.
    [Show full text]
  • Learning to Learn by Gradient Descent by Gradient Descent
    Learning to learn by gradient descent by gradient descent Marcin Andrychowicz1, Misha Denil1, Sergio Gómez Colmenarejo1, Matthew W. Hoffman1, David Pfau1, Tom Schaul1, Brendan Shillingford1,2, Nando de Freitas1,2,3 1Google DeepMind 2University of Oxford 3Canadian Institute for Advanced Research [email protected] {mdenil,sergomez,mwhoffman,pfau,schaul}@google.com [email protected], [email protected] Abstract The move from hand-designed features to learned features in machine learning has been wildly successful. In spite of this, optimization algorithms are still designed by hand. In this paper we show how the design of an optimization algorithm can be cast as a learning problem, allowing the algorithm to learn to exploit structure in the problems of interest in an automatic way. Our learned algorithms, implemented by LSTMs, outperform generic, hand-designed competitors on the tasks for which they are trained, and also generalize well to new tasks with similar structure. We demonstrate this on a number of tasks, including simple convex problems, training neural networks, and styling images with neural art. 1 Introduction Frequently, tasks in machine learning can be expressed as the problem of optimizing an objective function f(✓) defined over some domain ✓ ⇥. The goal in this case is to find the minimizer 2 ✓⇤ = arg min✓ ⇥ f(✓). While any method capable of minimizing this objective function can be applied, the standard2 approach for differentiable functions is some form of gradient descent, resulting in a sequence of updates ✓ = ✓ ↵ f(✓ ) . t+1 t − tr t The performance of vanilla gradient descent, however, is hampered by the fact that it only makes use of gradients and ignores second-order information.
    [Show full text]
  • Training Neural Networks Without Gradients: a Scalable ADMM Approach
    Training Neural Networks Without Gradients: A Scalable ADMM Approach Gavin Taylor1 [email protected] Ryan Burmeister1 Zheng Xu2 [email protected] Bharat Singh2 [email protected] Ankit Patel3 [email protected] Tom Goldstein2 [email protected] 1United States Naval Academy, Annapolis, MD USA 2University of Maryland, College Park, MD USA 3Rice University, Houston, TX USA Abstract many parameters. Because big datasets provide results that With the growing importance of large network (often dramatically) outperform the prior state-of-the-art in models and enormous training datasets, GPUs many machine learning tasks, researchers are willing to have become increasingly necessary to train neu- purchase specialized hardware such as GPUs, and commit ral networks. This is largely because conven- large amounts of time to training models and tuning hyper- tional optimization algorithms rely on stochastic parameters. gradient methods that don’t scale well to large Gradient-based training methods have several properties numbers of cores in a cluster setting. Further- that contribute to this need for specialized hardware. First, more, the convergence of all gradient methods, while large amounts of data can be shared amongst many including batch methods, suffers from common cores, existing optimization methods suffer when paral- problems like saturation effects, poor condition- lelized. Second, training neural nets requires optimizing ing, and saddle points. This paper explores an highly non-convex objectives that exhibit saddle points, unconventional training method that uses alter- poor conditioning, and vanishing gradients, all of which nating direction methods and Bregman iteration slow down gradient-based methods such as stochastic gra- to train networks without gradient descent steps.
    [Show full text]
  • Training Deep Networks with Stochastic Gradient Normalized by Layerwise Adaptive Second Moments
    Training Deep Networks with Stochastic Gradient Normalized by Layerwise Adaptive Second Moments Boris Ginsburg 1 Patrice Castonguay 1 Oleksii Hrinchuk 1 Oleksii Kuchaiev 1 Ryan Leary 1 Vitaly Lavrukhin 1 Jason Li 1 Huyen Nguyen 1 Yang Zhang 1 Jonathan M. Cohen 1 Abstract We started with Adam, and then (1) replaced the element- wise second moment with the layer-wise moment, (2) com- We propose NovoGrad, an adaptive stochastic puted the first moment using gradients normalized by layer- gradient descent method with layer-wise gradient wise second moment, (3) decoupled weight decay (WD) normalization and decoupled weight decay. In from normalized gradients (similar to AdamW). our experiments on neural networks for image classification, speech recognition, machine trans- The resulting algorithm, NovoGrad, combines SGD’s and lation, and language modeling, it performs on par Adam’s strengths. We applied NovoGrad to a variety of or better than well-tuned SGD with momentum, large scale problems — image classification, neural machine Adam, and AdamW. Additionally, NovoGrad (1) translation, language modeling, and speech recognition — is robust to the choice of learning rate and weight and found that in all cases, it performs as well or better than initialization, (2) works well in a large batch set- Adam/AdamW and SGD with momentum. ting, and (3) has half the memory footprint of Adam. 2. Related Work NovoGrad belongs to the family of Stochastic Normalized 1. Introduction Gradient Descent (SNGD) optimizers (Nesterov, 1984; Hazan et al., 2015). SNGD uses only the direction of the The most popular algorithms for training Neural Networks stochastic gradient gt to update the weights wt: (NNs) are Stochastic Gradient Descent (SGD) with mo- gt mentum (Polyak, 1964; Sutskever et al., 2013) and Adam wt+1 = wt − λt · (Kingma & Ba, 2015).
    [Show full text]
  • CSE 152: Computer Vision Manmohan Chandraker
    CSE 152: Computer Vision Manmohan Chandraker Lecture 15: Optimization in CNNs Recap Engineered against learned features Label Convolutional filters are trained in a Dense supervised manner by back-propagating classification error Dense Dense Convolution + pool Label Convolution + pool Classifier Convolution + pool Pooling Convolution + pool Feature extraction Convolution + pool Image Image Jia-Bin Huang and Derek Hoiem, UIUC Two-layer perceptron network Slide credit: Pieter Abeel and Dan Klein Neural networks Non-linearity Activation functions Multi-layer neural network From fully connected to convolutional networks next layer image Convolutional layer Slide: Lazebnik Spatial filtering is convolution Convolutional Neural Networks [Slides credit: Efstratios Gavves] 2D spatial filters Filters over the whole image Weight sharing Insight: Images have similar features at various spatial locations! Key operations in a CNN Feature maps Spatial pooling Non-linearity Convolution (Learned) . Input Image Input Feature Map Source: R. Fergus, Y. LeCun Slide: Lazebnik Convolution as a feature extractor Key operations in a CNN Feature maps Rectified Linear Unit (ReLU) Spatial pooling Non-linearity Convolution (Learned) Input Image Source: R. Fergus, Y. LeCun Slide: Lazebnik Key operations in a CNN Feature maps Spatial pooling Max Non-linearity Convolution (Learned) Input Image Source: R. Fergus, Y. LeCun Slide: Lazebnik Pooling operations • Aggregate multiple values into a single value • Invariance to small transformations • Keep only most important information for next layer • Reduces the size of the next layer • Fewer parameters, faster computations • Observe larger receptive field in next layer • Hierarchically extract more abstract features Key operations in a CNN Feature maps Spatial pooling Non-linearity Convolution (Learned) . Input Image Input Feature Map Source: R.
    [Show full text]
  • Automated Source Code Generation and Auto-Completion Using Deep Learning: Comparing and Discussing Current Language Model-Related Approaches
    Article Automated Source Code Generation and Auto-Completion Using Deep Learning: Comparing and Discussing Current Language Model-Related Approaches Juan Cruz-Benito 1,* , Sanjay Vishwakarma 2,†, Francisco Martin-Fernandez 1 and Ismael Faro 1 1 IBM Quantum, IBM T.J. Watson Research Center, Yorktown Heights, NY 10598, USA; [email protected] (F.M.-F.); [email protected] (I.F.) 2 Electrical and Computer Engineering, Carnegie Mellon University, Mountain View, CA 94035, USA; [email protected] * Correspondence: [email protected] † Intern at IBM Quantum at the time of writing this paper. Abstract: In recent years, the use of deep learning in language models has gained much attention. Some research projects claim that they can generate text that can be interpreted as human writ- ing, enabling new possibilities in many application areas. Among the different areas related to language processing, one of the most notable in applying this type of modeling is programming languages. For years, the machine learning community has been researching this software engi- neering area, pursuing goals like applying different approaches to auto-complete, generate, fix, or evaluate code programmed by humans. Considering the increasing popularity of the deep learning- enabled language models approach, we found a lack of empirical papers that compare different deep learning architectures to create and use language models based on programming code. This paper compares different neural network architectures like Average Stochastic Gradient Descent (ASGD) Weight-Dropped LSTMs (AWD-LSTMs), AWD-Quasi-Recurrent Neural Networks (QRNNs), and Citation: Cruz-Benito, J.; Transformer while using transfer learning and different forms of tokenization to see how they behave Vishwakarma, S.; Martin- Fernandez, F.; Faro, I.
    [Show full text]
  • GEE: a Gradient-Based Explainable Variational Autoencoder for Network Anomaly Detection
    GEE: A Gradient-based Explainable Variational Autoencoder for Network Anomaly Detection Quoc Phong Nguyen Kar Wai Lim Dinil Mon Divakaran National University of Singapore National University of Singapore Trustwave [email protected] [email protected] [email protected] Kian Hsiang Low Mun Choon Chan National University of Singapore National University of Singapore [email protected] [email protected] Abstract—This paper looks into the problem of detecting number of factors, such as end-user behavior, customer busi- network anomalies by analyzing NetFlow records. While many nesses (e.g., banking, retail), applications, location, time of the previous works have used statistical models and machine learning day, and are expected to evolve with time. Such diversity and techniques in a supervised way, such solutions have the limitations that they require large amount of labeled data for training and dynamism limits the utility of rule-based detection systems. are unlikely to detect zero-day attacks. Existing anomaly detection Next, as capturing, storing and processing raw traffic from solutions also do not provide an easy way to explain or identify such high capacity networks is not practical, Internet routers attacks in the anomalous traffic. To address these limitations, today have the capability to extract and export meta data such we develop and present GEE, a framework for detecting and explaining anomalies in network traffic. GEE comprises of two as NetFlow records [3]. With NetFlow, the amount of infor- components: (i) Variational Autoencoder (VAE) — an unsuper- mation captured is brought down by orders of magnitude (in vised deep-learning technique for detecting anomalies, and (ii) a comparison to raw packet capture), not only because a NetFlow gradient-based fingerprinting technique for explaining anomalies.
    [Show full text]
  • Optimization and Gradient Descent INFO-4604, Applied Machine Learning University of Colorado Boulder
    Optimization and Gradient Descent INFO-4604, Applied Machine Learning University of Colorado Boulder September 11, 2018 Prof. Michael Paul Prediction Functions Remember: a prediction function is the function that predicts what the output should be, given the input. Prediction Functions Linear regression: f(x) = wTx + b Linear classification (perceptron): f(x) = 1, wTx + b ≥ 0 -1, wTx + b < 0 Need to learn what w should be! Learning Parameters Goal is to learn to minimize error • Ideally: true error • Instead: training error The loss function gives the training error when using parameters w, denoted L(w). • Also called cost function • More general: objective function (in general objective could be to minimize or maximize; with loss/cost functions, we want to minimize) Learning Parameters Goal is to minimize loss function. How do we minimize a function? Let’s review some math. Rate of Change The slope of a line is also called the rate of change of the line. y = ½x + 1 “rise” “run” Rate of Change For nonlinear functions, the “rise over run” formula gives you the average rate of change between two points Average slope from x=-1 to x=0 is: 2 f(x) = x -1 Rate of Change There is also a concept of rate of change at individual points (rather than two points) Slope at x=-1 is: f(x) = x2 -2 Rate of Change The slope at a point is called the derivative at that point Intuition: f(x) = x2 Measure the slope between two points that are really close together Rate of Change The slope at a point is called the derivative at that point Intuition: Measure the
    [Show full text]
  • Gradient Descent (PDF)
    18.657: Mathematics of Machine Learning Lecturer: Philippe Rigollet Lecture 11 Scribe: Kevin Li Oct. 14, 2015 2. CONVEX OPTIMIZATION FOR MACHINE LEARNING In this lecture, we will cover the basics of convex optimization as it applies to machine learning. There is much more to this topic than will be covered in this class so you may be interested in the following books. Convex Optimization by Boyd and Vandenberghe Lecture notes on Convex Optimization by Nesterov Convex Optimization: Algorithms and Complexity by Bubeck Online Convex Optimization by Hazan The last two are drafts and can be obtained online. 2.1 Convex Problems A convex problem is an optimization problem of the form min f(x) where f andC are x2C convex. First, we will debunk the idea that convex problems are easy by showing that virtually all optimization problems can be written as a convex problem. We can rewrite an optimization problem as follows. min f(x) , min t , min t X 2X t≥f(x);x2X (x;t)2epi(f) where the epigraph of a function is defined by epi(f)=f(x; t) 2X 2 ×IR : t ≥f(x)g Figure 1: An example of an epigraph. Source: https://en.wikipedia.org/wiki/Epigraph_(mathematics) 1 Now we observe that for linear functions, min c>x = min c>x x2D x2conv(D) where the convex hull is defined N N X X conv(D) = fy : 9 N 2 Z+; x1; : : : ; xN 2 D; αi ≥ 0; αi = 1; y = αixig i=1 i=1 To prove this, we know that the left side is a least as big as the right side since D ⊂ conv(D).
    [Show full text]
  • Lecture 10: Recurrent Neural Networks
    Lecture 10: Recurrent Neural Networks Fei-Fei Li, Ranjay Krishna, Danfei Xu Lecture 10 - 1 May 7, 2020 Administrative: Midterm - Midterm next Tue 5/12 take home - 1H 20 Mins (+20m buffer) within a 24 hour time period. - Will be released on Gradescope. - See Piazza for detailed information - Midterm review session: Fri 5/8 discussion section - Midterm covers material up to this lecture (Lecture 10) Fei-Fei Li, Ranjay Krishna, Danfei Xu Lecture 10 - 2 May 7, 2020 Administrative - Project proposal feedback has been released - Project milestone due Mon 5/18, see Piazza for requirements ** Need to have some baseline / initial results by then, so start implementing soon if you haven’t yet! - A3 will be released Wed 5/13, due Wed 5/27 Fei-Fei Li, Ranjay Krishna, Danfei Xu Lecture 10 - 3 May 7, 2020 Last Time: CNN Architectures GoogLeNet AlexNet Fei-Fei Li, Ranjay Krishna, Danfei Xu Lecture 10 - 4 May 7, 2020 Last Time: CNN Architectures ResNet SENet Fei-Fei Li, Ranjay Krishna, Danfei Xu Lecture 10 - 5 May 7, 2020 Comparing complexity... An Analysis of Deep Neural Network Models for Practical Applications, 2017. Figures copyright Alfredo Canziani, Adam Paszke, Eugenio Culurciello, 2017. Reproduced with permission. Fei-Fei Li, Ranjay Krishna, Danfei Xu Lecture 10 - 6 May 7, 2020 Efficient networks... MobileNets: Efficient Convolutional Neural Networks for Mobile Applications [Howard et al. 2017] - Depthwise separable convolutions replace BatchNorm standard convolutions by Pool BatchNorm factorizing them into a C2HW Pointwise Conv (1x1, C->C) convolutions depthwise convolution and a Pool 1x1 convolution BatchNorm 9C2HW Conv (3x3, C->C) - Much more efficient, with Pool little loss in accuracy Stanford network Conv (3x3, C->C, Depthwise 2 9CHW - Follow-up MobileNetV2 work Total compute:9C HW groups=C) convolutions in 2018 (Sandler et al.) MobileNets - ShuffleNet: Zhang et al, Total compute:9CHW + C2HW CVPR 2018 Fei-Fei Li, Ranjay Krishna, Danfei Xu Lecture 10 - May 7, 2020 Meta-learning: Learning to learn network architectures..
    [Show full text]
  • Stochastic Gradient Descent in Machine Learning
    DEGREE PROJECT IN TECHNOLOGY, FIRST CYCLE, 15 CREDITS STOCKHOLM, SWEDEN 2019 Stochastic Gradient Descent in Machine Learning CHRISTIAN L. THUNBERG NIKLAS MANNERSKOG KTH ROYAL INSTITUTE OF TECHNOLOGY SCHOOL OF ENGINEERING SCIENCES Abstract Some tasks, like recognizing digits and spoken words, are simple for humans to complete yet hard to solve for computer programs. For instance the human intuition behind recognizing the number eight, "8 ", is to identify two loops on top of each other and it turns out this is not easy to represent as an algorithm. With machine learning one can tackle the problem in a new, easier, way where the computer program learns to recognize patterns and make conclusions from them. In this bachelor thesis a digit recognizing program is implemented and the parameters of the stochastic gradient descent optimizing algorithm are analyzed based on how their effect on the computation speed and accuracy. These parameters being the learning rate ∆t and batch size N. The implemented digit recognizing program yielded an accuracy of around 95 % when tested and the time per iteration stayed constant during the training session and increased linearly with batch size. Low learning rates yielded a slower rate of convergence while larger ones yielded faster but more unstable con- vergence. Larger batch sizes also improved the convergence but at the cost of more computational power. Keywords: Stochastic Gradient Descent, Machinelearning, Neural Networks, Learning Rate, Batch Size, MNIST i DEGREE PROJECT IN TEKNIK, FIRST CYCLE, 15 CREDITS STOCKHOLM, SWEDEN 2019 Stochastic Gradient Descent inom Maskininlärning CHRISTIAN L. THUNBERG NIKLAS MANNERSKOG KTH ROYAL INSTITUTE OF TECHNOLOGY SKOLAN FÖR TEKNIKVETENSKAP Sammanfattning Vissa problem som f¨orm¨anniskor ¨arenkla att l¨osa,till exempel: att k¨annaigen siffror och sagda ord, ¨arsv˚artatt implementera i datorprogram.
    [Show full text]
  • Lecture 6: Stochastic Gradient Descent Sanjeev Arora Elad Hazan
    COS 402 – Machine Learning and Artificial Intelligence Fall 2016 Lecture 6: stochastic gradient descent Sanjeev Arora Elad Hazan Admin • Exercise 2 (implementation) this Thu, in class • Exercise 3 (written), this Thu, in class • Movie – “Ex Machina” + discussion panel w. Prof. Hasson (PNI) Wed Oct. 5th 19:30 tickets from Bella; room 204 COS • Today: special guest - Dr. Yoram Singer @ Google Recap • Definition + fundamental theorem of statistical learning, motivated efficient algorithms/optimization • Convexity and it’s computational importance • Local greedy optimization – gradient descent Agenda • Stochastic gradient descent • Dr. Singer on opt @ google & beyond Mathematical optimization Input: function �: � ↦ �, for � ⊆ �' Output: point � ∈ �, such that � � ≤ � � ∀ � ∈ � What is Optimization But generally speaking... We’re screwed. ! Local (non global) minima of f0 Prefer Convex Problems! All kinds of constraints (even restricting to continuous functions): h(x)=sin(2πx)=0 250 200 150 100 50 0 −50 3 2 3 1 2 0 1 −1 0 −1 −2 −2 −3 −3 Duchi (UC Berkeley) Convex Optimization for Machine Learning Fall 2009 7 / 53 6 Convex functions: local à global Sum of convex functions à also convex Convex Functions and Sets n A function f : R R is convex if for x, y domfand any a [0, 1], ! 2 2 f(ax +(1 a)y) af(x)+(1 a)f(y) − − Convex Sets Definition AsetC Rn is convex if for x, y C and any α [0, 1], ⊆ ∈ ∈ αx +(1 α)y C. − ∈ n AsetC R is convex if for x, y C and any a [0, 1], ✓ 2 2 y ax +(1 a)y C x − 2 8 Duchi (UC Berkeley) Convex Optimization for Machine Learning Fall 2009 9 / 53 Special case: optimization for linear classification Given a sample � = �0, �0 ,… , �3,�3 , find hyperplane (through the origin w.l.o.g) such that: � min # of mistakes 7 = min ℓ(� �>,�>) for a convex loss function 7 90 Convex relaxation for 0-1 loss 1.
    [Show full text]