DOCSLIB.ORG

Optimizing Millions of Hyperparameters by Implicit Differentiation

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Optimizing Millions of Hyperparameters by Implicit Differentiation Optimizing Millions of Hyperparameters by Implicit Differentiation Jonathan Lorraine Paul Vicol David Duvenaud University of Toronto, Vector Institute {lorraine, pvicol, duvenaud}@cs.toronto.edu Abstract high-dimensional HO problems locally with gradient- We propose an algorithm for inexpensive based optimizers, but this is difficult because we must gradient-based hyperparameter optimization differentiate through the optimized weights as a func- that combines the implicit function theorem tion of the hyperparameters. In other words, we must (IFT) with efficient inverse Hessian approx- approximate the Jacobian of the best-response function imations. We present results on the rela- of the parameters to the hyperparameters. tionship between the IFT and differentiat- We leverage the Implicit Function Theorem (IFT) to ing through optimization, motivating our al- compute the optimized validation loss gradient with gorithm. We use the proposed approach to respect to the hyperparameters—hereafter denoted the train modern network architectures with mil- hypergradient. The IFT requires inverting the training lions of weights and millions of hyperparame- Hessian with respect to the NN weights, which is infea- ters. We learn a data-augmentation network— sible for modern, deep networks. Thus, we propose an where every weight is a hyperparameter tuned approximate inverse, motivated by a link to unrolled for validation performance—that outputs aug- differentiation [3] that scales to Hessians of large NNs, mented training examples; we learn a distilled is more stable than conjugate gradient [22, 7], and only dataset where each feature in each datapoint requires a constant amount of memory. is a hyperparameter; and we tune millions of regularization hyperparameters. Jointly Finally, when fitting many parameters, the amount of tuning weights and hyperparameters with our data can limit generalization. There are ad hoc rules approach is only a few times more costly in for partitioning data into training and validation sets— memory and compute than standard training. e.g., using 10% for validation. Often, practitioners re-train their models from scratch on the combined 1 Introduction training and validation partitions with optimized hy- perparameters, which can provide marginal test-time The generalization of neural networks (NNs) depends performance increases. We verify empirically that stan- crucially on the choice of hyperparameters. Hyperpa- dard partitioning and re-training procedures perform rameter optimization (HO) has a rich history [1, 2], and well when fitting few hyperparameters, but break down achieved recent success in scaling due to gradient-based when fitting many. When fitting many hyperparame- optimizers [3–11]. There are dozens of regularization ters, we need a large validation partition, which makes techniques to combine in deep learning, and each may re-training our model with optimized hyperparameters have multiple hyperparameters [12]. If we can scale HO vital for strong test performance. to have as many—or more—hyperparameters as param- eters, there are various exciting regularization strategies Contributions arXiv:1911.02590v1 [cs.LG] 6 Nov 2019 to investigate. For example, we could learn a distilled • We propose a stable inverse Hessian approximation dataset with a hyperparameter for every feature of with constant memory cost. each input [4, 13], weights on each loss term [14–16], or augmentation on each input [17, 18]. • We show that the IFT is the limit of differentiating When the hyperparameters are low-dimensional— through optimization. e.g., 1-5 dimensions—simple methods, like random • We scale IFT-based hyperparameter optimization search, work; however, these break down for medium- to modern, large neural architectures, including dimensional HO—e.g., 5-100 dimensions. We may AlexNet and LSTM-based language models. use more scalable algorithms like Bayesian Optimiza- tion [19–21], but this often breaks down for high- • We demonstrate several uses for fitting hyperpa- dimensional HO—e.g., >100 dimensions. We can solve rameters almost as easily as weights, including per- parameter regularization, data distillation, and learned-from-scratch data augmentation methods. • We explore how training-validation splits should Submitted to AISTATS 2020. change when tuning many hyperparameters. Jonathan Lorraine, Paul Vicol, David Duvenaud ) ) λ , (λ, w∗(λ)) w ( ∗(λ) w . L ( L d n i w∗(λ) = i l a a r V Implict T ∗ = L L L λ best response s s s − s w o ∗ function o + = L L L L λ w λ Hypergradient w∗ = λ λ r λ Implicit derivative te r e te (λ , w) m e 0 a m L r a a r λ0, w∗(λ0) Pa p a ra r Pa p me e ra r te p me e r w y ter p H w y H Figure 1: Overview of gradient-based hyperparameter optimization (HO). Left: a training loss manifold; Right: a validation loss manifold. The implicit function w∗(λ) is the best-response of the weights to the hyperparameters ∗ and shown in blue projected onto the (λ; w)-plane. We get our desired objective function LV(λ) when the best-response is put into the validation loss, shown projected on the hyperparameter axis in red. The validation loss does not depend directly on the hyperparameters, as is typical in hyperparameter optimization. Instead, the hyperparameters only affect the validation loss by changing the weights’ response. We show the best-response Jacobian in blue, and the hypergradient in red. 2 Overview of Proposed Algorithm 3. We can estimate the implicit best-response There are four essential components to understanding with the IFT: We approximate the best-response our proposed algorithm. Further background is pro- Jacobian—how the optimal weights change with respect vided in AppendixA, and notation is shown in Table5. to the hyperparameters—using the IFT (Thm.1). We present the complete statement in AppendixC, but 1. HO is nested optimization: Let LT and LV highlight the key assumptions and results here. denote the training and validation losses, w the NN Theorem 1 (Cauchy, Implicit Function Theorem). If weights, and λ the hyperparameters. We aim to find op- for some (λ0; w0); @ LT j = 0 and regularity condi- timal hyperparameters λ∗ such that the NN minimizes @w λ0;w0 tions are satisfied, then surrounding (λ0; w0) there is a the validation loss after training: ∗ @ LT function w (λ) s.t. @w jλ;w∗(λ) = 0 and we have: ∗ ∗ λ :=arg min LV(λ) where (1) ∗ h 2 i−1 2 λ @ w @ LT @ LT @λ =− T × T (IFT) ∗ ∗ ∗ λ0 @w@w @w@λ λ0;w∗(λ0) LV(λ):=LV(λ;w (λ)) and w (λ):=arg min LT(λ;w) (2) | {z } | {z } w training Hessian training mixed partials ∗ Our implicit function is w (λ), which is the best-response @ LT 0 0 The condition @w jλ0;w0 = 0 is equivalent to λ ; w be- of the weights to the hyperparameters. We assume ing a fixed point of the training gradient field. Since unique solutions to arg min for simplicity. w∗(λ0) is a fixed point of the training gradient field, 2. Hypergradients have two terms: For gradient- we can leverage the IFT to evaluate the best-response ∗ @ LV(λ) Jacobian locally. We only have access to an approxi- based HO we want the hypergradient @λ , which decomposes into: mation of the true best-response—denoted wc∗—which we can find with gradient descent. ∗ ∗ @ LV(λ) @ LV @ LV @ w @λ = @λ + @w @λ = 4. Tractable inverse Hessian approximations: | {z } λ;w∗(λ) hypergradient To exactly invert a general m × m Hessian, we often 3 hyperparam indirect grad. (3) require O(m ) operations, which is intractable for the ∗ z ∗ }| ∗ { matrix in Eq. IFT in modern NNs. We can efficiently @ LV(λ;w (λ)) @ LV(λ;w (λ)) @ w (λ) @λ + @w∗(λ) × @λ approximate the inverse with the Neumann series: | {z } | {z } hyperparam direct grad. | {z } best-response Jacobian parameter direct grad. i h 2 i−1 h 2 ij @ LT X @ LT T = lim I − T (4) The direct gradient is easy to compute, but the indirect @w@w i!1 @w@w gradient is difficult to compute because we must ac- j=0 count for how the optimal weights change with respect @ w∗(λ) In Section4 we show that unrolling differentiation for ∗ to the hyperparameters (i.e., @λ ). In HO the direct i steps around locally optimal weights w is equivalent gradient is often identically 0, necessitating an approx- to approximating the inverse with the first i terms in imation of the indirect gradient to make any progress the Neumann series. We then show how to use this (visualized in Fig.1). approximation without instantiating any matrices by using efficient vector-Jacobian products. Jonathan Lorraine, Paul Vicol, David Duvenaud @ w∗ @λ @ L∗ V @ LV @ LV z}|{ @λ @λ @w Early work applied the IFT to regularization by ex- z}|{ = z}|{ + z }| { plicitly computing the Hessian (or Gauss-Newton) in- verse [23, 2]. In [24], the identity matrix is used to ap- @2L −1 2 proximate the inverse Hessian in the IFT. HOAG [30] − T @ LT @w@wT @w@λT uses conjugate gradient (CG) to invert the Hessian @ LV @ LV z }| { z}|{ @λ @w approximately and provides convergence results given z}|{ z }| { = + tolerances on the optimal parameter and inverse. In iMAML [9], a center to the weights is fit to perform | {z } vector-inverse Hessian product well on multiple tasks—contrasted with our use of vali- 2 @ LT @w@λT dation loss. In DEQ [31], implicit differentiation is used @ L @ L @2L −1 V V ×− T z}|{ @λ @w @w@wT to add differentiable fixed-point methods into NN ar- = z}|{ + z }| { chitectures. We use a Neumann approximation for the inverse-Hessian, instead of CG [30,9] or the identity. | {z } vector-Jacobian product Approximate inversion algorithms. CG is diffi- Figure 2: Hypergradient computation. The entire cult to scale to modern, deep NNs.
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]
  • Deep Learning Based Computer Generated Face Identification Using
    applied sciences Article Deep Learning Based Computer Generated Face Identification Using Convolutional Neural Network L. Minh Dang 1, Syed Ibrahim Hassan 1, Suhyeon Im 1, Jaecheol Lee 2, Sujin Lee 1 and Hyeonjoon Moon 1,* 1 Department of Computer Science and Engineering, Sejong University, Seoul 143-747, Korea; [email protected] (L.M.D.); [email protected] (S.I.H.); [email protected] (S.I.); [email protected] (S.L.) 2 Department of Information Communication Engineering, Sungkyul University, Seoul 143-747, Korea; [email protected] * Correspondence: [email protected] Received: 30 October 2018; Accepted: 10 December 2018; Published: 13 December 2018 Abstract: Generative adversarial networks (GANs) describe an emerging generative model which has made impressive progress in the last few years in generating photorealistic facial images. As the result, it has become more and more difficult to differentiate between computer-generated and real face images, even with the human’s eyes. If the generated images are used with the intent to mislead and deceive readers, it would probably cause severe ethical, moral, and legal issues. Moreover, it is challenging to collect a dataset for computer-generated face identification that is large enough for research purposes because the number of realistic computer-generated images is still limited and scattered on the internet. Thus, a development of a novel decision support system for analyzing and detecting computer-generated face images generated by the GAN network is crucial. In this paper, we propose a customized convolutional neural network, namely CGFace, which is specifically designed for the computer-generated face detection task by customizing the number of convolutional layers, so it performs well in detecting computer-generated face images.
    [Show full text]
  • Matrix Calculus
    Appendix D Matrix Calculus From too much study, and from extreme passion, cometh madnesse. Isaac Newton [205, §5] − D.1 Gradient, Directional derivative, Taylor series D.1.1 Gradients Gradient of a differentiable real function f(x) : RK R with respect to its vector argument is defined uniquely in terms of partial derivatives→ ∂f(x) ∂x1 ∂f(x) , ∂x2 RK f(x) . (2053) ∇ . ∈ . ∂f(x) ∂xK while the second-order gradient of the twice differentiable real function with respect to its vector argument is traditionally called the Hessian; 2 2 2 ∂ f(x) ∂ f(x) ∂ f(x) 2 ∂x1 ∂x1∂x2 ··· ∂x1∂xK 2 2 2 ∂ f(x) ∂ f(x) ∂ f(x) 2 2 K f(x) , ∂x2∂x1 ∂x2 ··· ∂x2∂xK S (2054) ∇ . ∈ . .. 2 2 2 ∂ f(x) ∂ f(x) ∂ f(x) 2 ∂xK ∂x1 ∂xK ∂x2 ∂x ··· K interpreted ∂f(x) ∂f(x) 2 ∂ ∂ 2 ∂ f(x) ∂x1 ∂x2 ∂ f(x) = = = (2055) ∂x1∂x2 ³∂x2 ´ ³∂x1 ´ ∂x2∂x1 Dattorro, Convex Optimization Euclidean Distance Geometry, Mεβoo, 2005, v2020.02.29. 599 600 APPENDIX D. MATRIX CALCULUS The gradient of vector-valued function v(x) : R RN on real domain is a row vector → v(x) , ∂v1(x) ∂v2(x) ∂vN (x) RN (2056) ∇ ∂x ∂x ··· ∂x ∈ h i while the second-order gradient is 2 2 2 2 , ∂ v1(x) ∂ v2(x) ∂ vN (x) RN v(x) 2 2 2 (2057) ∇ ∂x ∂x ··· ∂x ∈ h i Gradient of vector-valued function h(x) : RK RN on vector domain is → ∂h1(x) ∂h2(x) ∂hN (x) ∂x1 ∂x1 ··· ∂x1 ∂h1(x) ∂h2(x) ∂hN (x) h(x) , ∂x2 ∂x2 ··· ∂x2 ∇ .
    [Show full text]
  • Differentiability in Several Variables: Summary of Basic Concepts
    DIFFERENTIABILITY IN SEVERAL VARIABLES: SUMMARY OF BASIC CONCEPTS 3 3 1. Partial derivatives If f : R ! R is an arbitrary function and a = (x0; y0; z0) 2 R , then @f f(a + tj) ¡ f(a) f(x ; y + t; z ) ¡ f(x ; y ; z ) (1) (a) := lim = lim 0 0 0 0 0 0 @y t!0 t t!0 t etc.. provided the limit exists. 2 @f f(t;0)¡f(0;0) Example 1. for f : R ! R, @x (0; 0) = limt!0 t . Example 2. Let f : R2 ! R given by ( 2 x y ; (x; y) 6= (0; 0) f(x; y) = x2+y2 0; (x; y) = (0; 0) Then: @f f(t;0)¡f(0;0) 0 ² @x (0; 0) = limt!0 t = limt!0 t = 0 @f f(0;t)¡f(0;0) ² @y (0; 0) = limt!0 t = 0 Note: away from (0; 0), where f is the quotient of differentiable functions (with non-zero denominator) one can apply the usual rules of derivation: @f 2xy(x2 + y2) ¡ 2x3y (x; y) = ; for (x; y) 6= (0; 0) @x (x2 + y2)2 2 2 2. Directional derivatives. If f : R ! R is a map, a = (x0; y0) 2 R and v = ®i + ¯j is a vector in R2, then by definition f(a + tv) ¡ f(a) f(x0 + t®; y0 + t¯) ¡ f(x0; y0) (2) @vf(a) := lim = lim t!0 t t!0 t Example 3. Let f the function from the Example 2 above. Then for v = ®i + ¯j a unit vector (i.e.
    [Show full text]
  • Lipschitz Recurrent Neural Networks
    Published as a conference paper at ICLR 2021 LIPSCHITZ RECURRENT NEURAL NETWORKS N. Benjamin Erichson Omri Azencot Alejandro Queiruga ICSI and UC Berkeley Ben-Gurion University Google Research [email protected] [email protected] [email protected] Liam Hodgkinson Michael W. Mahoney ICSI and UC Berkeley ICSI and UC Berkeley [email protected] [email protected] ABSTRACT Viewing recurrent neural networks (RNNs) as continuous-time dynamical sys- tems, we propose a recurrent unit that describes the hidden state’s evolution with two parts: a well-understood linear component plus a Lipschitz nonlinearity. This particular functional form facilitates stability analysis of the long-term behavior of the recurrent unit using tools from nonlinear systems theory. In turn, this en- ables architectural design decisions before experimentation. Sufficient conditions for global stability of the recurrent unit are obtained, motivating a novel scheme for constructing hidden-to-hidden matrices. Our experiments demonstrate that the Lipschitz RNN can outperform existing recurrent units on a range of bench- mark tasks, including computer vision, language modeling and speech prediction tasks. Finally, through Hessian-based analysis we demonstrate that our Lipschitz recurrent unit is more robust with respect to input and parameter perturbations as compared to other continuous-time RNNs. 1 INTRODUCTION Many interesting problems exhibit temporal structures that can be modeled with recurrent neural networks (RNNs), including problems in robotics, system identification, natural language process- ing, and machine learning control. In contrast to feed-forward neural networks, RNNs consist of one or more recurrent units that are designed to have dynamical (recurrent) properties, thereby enabling them to acquire some form of internal memory.
    [Show full text]
  • Neural Networks
    School of Computer Science 10-601B Introduction to Machine Learning Neural Networks Readings: Matt Gormley Bishop Ch. 5 Lecture 15 Murphy Ch. 16.5, Ch. 28 October 19, 2016 Mitchell Ch. 4 1 Reminders 2 Outline • Logistic Regression (Recap) • Neural Networks • Backpropagation 3 RECALL: LOGISTIC REGRESSION 4 Using gradient ascent for linear Recall… classifiers Key idea behind today’s lecture: 1. Define a linear classifier (logistic regression) 2. Define an objective function (likelihood) 3. Optimize it with gradient descent to learn parameters 4. Predict the class with highest probability under the model 5 Using gradient ascent for linear Recall… classifiers This decision function isn’t Use a differentiable function differentiable: instead: 1 T pθ(y =1t)= h(t)=sign(θ t) | 1+2tT( θT t) − sign(x) 1 logistic(u) ≡ −u 1+ e 6 Using gradient ascent for linear Recall… classifiers This decision function isn’t Use a differentiable function differentiable: instead: 1 T pθ(y =1t)= h(t)=sign(θ t) | 1+2tT( θT t) − sign(x) 1 logistic(u) ≡ −u 1+ e 7 Recall… Logistic Regression Data: Inputs are continuous vectors of length K. Outputs are discrete. (i) (i) N K = t ,y where t R and y 0, 1 D { }i=1 ∈ ∈ { } Model: Logistic function applied to dot product of parameters with input vector. 1 pθ(y =1t)= | 1+2tT( θT t) − Learning: finds the parameters that minimize some objective function. θ∗ = argmin J(θ) θ Prediction: Output is the most probable class. yˆ = `;Kt pθ(y t) y 0,1 | ∈{ } 8 NEURAL NETWORKS 9 Learning highly non-linear functions f: X à Y l f might be non-linear
    [Show full text]
  • Jointly Clustering with K-Means and Learning Representations
    Deep k-Means: Jointly clustering with k-Means and learning representations Maziar Moradi Fard Univ. Grenoble Alpes, CNRS, Grenoble INP – LIG [email protected] Thibaut Thonet Univ. Grenoble Alpes, CNRS, Grenoble INP – LIG [email protected] Eric Gaussier Univ. Grenoble Alpes, CNRS, Grenoble INP – LIG, Skopai [email protected] Abstract We study in this paper the problem of jointly clustering and learning representations. As several previous studies have shown, learning representations that are both faithful to the data to be clustered and adapted to the clustering algorithm can lead to better clustering performance, all the more so that the two tasks are performed jointly. We propose here such an approach for k-Means clustering based on a continuous reparametrization of the objective function that leads to a truly joint solution. The behavior of our approach is illustrated on various datasets showing its efficacy in learning representations for objects while clustering them. 1 Introduction Clustering is a long-standing problem in the machine learning and data mining fields, and thus accordingly fostered abundant research. Traditional clustering methods, e.g., k-Means [23] and Gaussian Mixture Models (GMMs) [6], fully rely on the original data representations and may then be ineffective when the data points (e.g., images and text documents) live in a high-dimensional space – a problem commonly known as the curse of dimensionality. Significant progress has been arXiv:1806.10069v2 [cs.LG] 12 Dec 2018 made in the last decade or so to learn better, low-dimensional data representations [13].
    [Show full text]
  • Differentiability of the Convolution
    Differentiability of the convolution. Pablo Jimenez´ Rodr´ıguez Universidad de Valladolid March, 8th, 2019 Pablo Jim´enezRodr´ıguez Differentiability of the convolution Definition Let L1[−1; 1] be the set of the 2-periodic functions f so that R 1 −1 jf (t)j dt < 1: Given f ; g 2 L1[−1; 1], the convolution of f and g is the function Z 1 f ∗ g(x) = f (t)g(x − t) dt −1 Pablo Jim´enezRodr´ıguez Differentiability of the convolution Definition Let L1[−1; 1] be the set of the 2-periodic functions f so that R 1 −1 jf (t)j dt < 1: Given f ; g 2 L1[−1; 1], the convolution of f and g is the function Z 1 f ∗ g(x) = f (t)g(x − t) dt −1 Pablo Jim´enezRodr´ıguez Differentiability of the convolution Definition Let L1[−1; 1] be the set of the 2-periodic functions f so that R 1 −1 jf (t)j dt < 1: Given f ; g 2 L1[−1; 1], the convolution of f and g is the function Z 1 f ∗ g(x) = f (t)g(x − t) dt −1 Pablo Jim´enezRodr´ıguez Differentiability of the convolution Definition Let L1[−1; 1] be the set of the 2-periodic functions f so that R 1 −1 jf (t)j dt < 1: Given f ; g 2 L1[−1; 1], the convolution of f and g is the function Z 1 f ∗ g(x) = f (t)g(x − t) dt −1 Pablo Jim´enezRodr´ıguez Differentiability of the convolution If f is Lipschitz, then f ∗ g is Lipschitz.
    [Show full text]
  • The Simple Essence of Automatic Differentiation
    The Simple Essence of Automatic Differentiation CONAL ELLIOTT, Target, USA Automatic differentiation (AD) in reverse mode (RAD) is a central component of deep learning and other uses of large-scale optimization. Commonly used RAD algorithms such as backpropagation, however, are complex and stateful, hindering deep understanding, improvement, and parallel execution. This paper develops a simple, generalized AD algorithm calculated from a simple, natural specification. The general algorithm is then specialized by varying the representation of derivatives. In particular, applying well-known constructions to a naive representation yields two RAD algorithms that are far simpler than previously known. In contrast to commonly used RAD implementations, the algorithms defined here involve no graphs, tapes, variables, partial derivatives, or mutation. They are inherently parallel-friendly, correct by construction, and usable directly from an existing programming language with no need for new data types or programming style, thanks to use of an AD-agnostic compiler plugin. CCS Concepts: · Mathematics of computing → Differential calculus; · Theory of computation → Program reasoning; Program specifications; Additional Key Words and Phrases: automatic differentiation, program calculation, category theory ACM Reference Format: Conal Elliott. 2018. The Simple Essence of Automatic Differentiation. Proc. ACM Program. Lang. 2, ICFP, Article 70 (September 2018), 29 pages. https://doi.org/10.1145/3236765 1 INTRODUCTION Accurate, efficient, and reliable computation of derivatives has become increasingly important over the last several years, thanks in large part to the successful use of backpropagation in machine learning, including multi-layer neural networks, also known as łdeep learningž [Goodfellow et al. 2016; Lecun et al. 2015]. Backpropagation is a specialization and independent invention of the reverse mode of automatic differentiation (AD) and is used to tune a parametric model to closely match observed data, using gradient descent (or stochastic gradient descent).
    [Show full text]
  • Expressivity of Deep Neural Networks∗
    Expressivity of Deep Neural Networks∗ Ingo G¨uhring yz Mones Raslan yz Gitta Kutyniokzx 10/07/2020 Abstract In this review paper, we give a comprehensive overview of the large variety of approximation results for neural networks. Approximation rates for classical function spaces as well as benefits of deep neural networks over shallow ones for specifically structured function classes are discussed. While the main body of existing results is for general feedforward architectures, we also review approximation results for convolutional, residual and recurrent neural networks. Keywords: approximation, expressivity, deep vs. shallow, function classes 1 Introduction While many aspects of the success of deep learning still lack a comprehensive mathematical explanation, the approximation properties1 of neural networks have been studied since around 1960 and are relatively well understood. Statistical learning theory formalizes the problem of approximating - in this context also called learning - a function from a finite set of samples. Next to statistical and algorithmic considerations, approximation theory plays a major role for the analysis of statistical learning problems. We will clarify this in the following by introducing some fundamental notions.2 Assume that X is an input space and Y a target space, L : Y × Y ! [0; 1] is a loss function and P(X ;Y) a (usually unknown) probability distribution on some σ-algebra of X × Y. We then aim at finding a minimizer of the risk functional3 Z X R : Y ! [0; 1]; f 7! L (f(x); y) dP(X ;Y)(x; y); X ×Y X induced by L and P(X ;Y) (where Y denotes the set of all functions from X to Y).
    [Show full text]
  • Differentiable Programming for Piecewise Polynomial
    Under review as a conference paper at ICLR 2021 DIFFERENTIABLE PROGRAMMING FOR PIECEWISE POLYNOMIAL FUNCTIONS Anonymous authors Paper under double-blind review ABSTRACT The paradigm of differentiable programming has significantly enhanced the scope of machine learning via the judicious use of gradient-based optimization. How- ever, standard differentiable programming methods (such as autodiff) typically require that the machine learning models be differentiable, limiting their applica- bility. We introduce a new, principled approach to extend gradient-based optimiza- tion to piecewise smooth models, such as k-histograms, splines, and segmentation maps. We derive an accurate form of the weak Jacobian of such functions and show that it exhibits a block-sparse structure that can be computed implicitly and efficiently. We show that using the redesigned Jacobian leads to improved per- formance in applications such as denoising with piecewise polynomial regression models, data-free generative model training, and image segmentation. 1 INTRODUCTION Motivation: Differentiable programming has been proposed as a paradigm shift in algorithm design. The main idea is to leverage gradient-based optimization to optimize the parameters of the algo- rithm, allowing for end-to-end trainable systems (such as deep neural networks) to exploit structure and patterns in data and achieve better performance. This approach has found use in a large va- riety of applications such as scientific computing (Innes, 2020; Innes et al., 2019; Schafer et al., 2020), image processing (Li et al., 2018b), physics engines (Degrave et al., 2017), computational simulations (Alnæs et al., 2015), and graphics (Li et al., 2018a; Chen et al., 2019). One way to leverage differential programming modules is to encode additional structural priors as “layers” in a larger machine learning model.
    [Show full text]
  • GAN and What It Is?
    Basics: Why GAN and what it is? The Optimization Problem of GANs How to Train GANs: An Odyssey How to Evaluate GANs? An Application Tour of GANs Summary and Open Challenges Outline Basics: Why GAN and what it is? Generative vs. Discriminative Models • Given a distribution of inputs X and labels Y • Discriminative: model the conditional distribution P(Y | X). • Generative networks model the joint distribution P(X , Y). • If the model understands the joint distribution P(X ,Y), then it • can calculate P(Y | X) using Bayes rule • can perform other tasks like P(X | Y), i.e., generating data from the label (called “conditional generation” in GANs) • understands the distribution better than a discriminative model, a scientific philosophy called “analysis by synthesis” “The analysis of the patterns generated by the world in any modality, with all their naturally occurring complexity and ambiguity, with the goal of reconstructing the processes, objects and events that produced them” - David Mumford Generative vs. Discriminative Models • Even if you only have X, you can still build a generative model • … making generative modeling amendable to unsupervised/semi-supervised representation learning • Not every problem is discriminative, but all problems could be generative! • However, generative modeling is harder! • Map from X to Y is typically many to one • Map from Y to X is typically one to many • Dimensionality of X typically >> dimensionality of Y • Hence compared to estimating P(Y | X), the estimation of P(X, Y) usually runs into the “curse of dimensionality” Setup of Generative Models Autoregressive Models RNNs for Autoregressive Language Modeling PixelRNN (van der Oord et al.
    [Show full text]