The Clarke subdifferential is not suited to tackle nonsmooth deep learning issues: backpropagation, mini-batches and steady states are not properly modelled. As a remedy, we introduce set valued conservative fields as surrogates to standard subdifferential mappings. We study their properties and provide elements of a calculus. Functions having a conservative field are called path differentiable. Convex/concave, semi-algebraic, or Clarke regular Lipschitz continuous functions are path differentiable as their corresponding subdifferentials are conservative. Another concrete and considerable class of examples of conservative fields, which are not subdifferential mappings, is given by the automatic differentiation oracle, as for instance the "subgradients" provided by the backpropagation algorithm in deep learning. Our differential model is eventually used to ensure subsequential convergence for nonsmooth stochastic gradient methods in the tame Lipschitz continuous setting offering the possibility of using mini-batches, the actual backpropagation oracle and $o(1/\log k)$ stepsizes.

A term "Deep learning" refers to training neural networks and sometimes very large neural networks. They are massively used for problems of classification and regression. The main goal is to optimize parameters \( w \) and \( b \) ( both in logistic regression and in neural networks). Gradient Descent achieved amazing results in solving these optimization tasks. Not so new, developed in the 1940s, 1950s and 1960s.

Excessive computational cost for learning large data and streaming data can be alleviated by using stochastic algorithms, such as stochastic gradient descent and its variants. Recent advances improve stochastic algorithms on convergence speed, adaptivity and structural awareness. However, distributional aspects of these new algorithms are poorly understood, especially for structured parameters. To develop statistical inference in this case, we propose a class of generalized regularized dual averaging (gRDA) algorithms with constant step size, which improves RDA (Xiao, 2010; Flammarion and Bach, 2017). Weak convergence of gRDA trajectories are studied, and as a consequence, for the first time in the literature, the asymptotic distributions for online l1 penalized problems become available. These general results apply to both convex and non-convex differentiable loss functions, and in particular, recover the existing regret bound for convex losses (Nemirovski et al., 2009). As important applications, statistical inferential theory on online sparse linear regression and online sparse principal component analysis are developed, and are supported by extensive numerical analysis. Interestingly, when gRDA is properly tuned, support recovery and central limiting distribution (with mean zero) hold simultaneously in the online setting, which is in contrast with the biased central limiting distribution of batch Lasso (Knight and Fu, 2000). Technical devices, including weak convergence of stochastic mirror descent, are developed as by-products with independent interest. Preliminary empirical analysis of modern image data shows that learning very sparse deep neural networks by gRDA does not necessarily sacrifice testing accuracy.

Despite the development of numerous adaptive optimizers, tuning the learning rate of stochastic gradient methods remains a major roadblock to obtaining good practical performance in machine learning. Rather than changing the learning rate at each iteration, we propose an approach that automates the most common hand-tuning heuristic: use a constant learning rate until "progress stops," then drop. We design an explicit statistical test that determines when the dynamics of stochastic gradient descent reach a stationary distribution. This test can be performed easily during training, and when it fires, we decrease the learning rate by a constant multiplicative factor. Our experiments on several deep learning tasks demonstrate that this statistical adaptive stochastic approximation (SASA) method can automatically find good learning rate schedules and match the performance of hand-tuned methods using default settings of its parameters. The statistical testing helps to control the variance of this procedure and improves its robustness.

We introduce a framework to study the transformation of problems with manifold constraints into unconstrained problems through parametrizations in terms of a Euclidean space. We call these parametrizations "trivializations". We prove conditions under which a trivialization is sound in the context of gradient-based optimization and we show how two large families of trivializations have overall favorable properties, but also suffer from a performance issue. We then introduce "dynamic trivializations", which solve this problem, and we show how these form a family of optimization methods that lie between trivializations and Riemannian gradient descent, and combine the benefits of both of them. We then show how to implement these two families of trivializations in practice for different matrix manifolds. To this end, we prove a formula for the gradient of the exponential of matrices, which can be of practical interest on its own. Finally, we show how dynamic trivializations improve the performance of existing methods on standard tasks designed to test long-term memory within neural networks.

Improving the sample efficiency in reinforcement learning has been a long-standing research problem. In this work, we aim to reduce the sample complexity of existing policy gradient methods. We propose a novel policy gradient algorithm called SRVR-PG, which only requires $O(1/\epsilon^{3/2})$ episodes to find an $\epsilon$-approximate stationary point of the nonconcave performance function $J(\boldsymbol{\theta})$ (i.e., $\boldsymbol{\theta}$ such that $\|\nabla J(\boldsymbol{\theta})\|_2^2\leq\epsilon$). This sample complexity improves the best known result $O(1/\epsilon^{5/3})$ for policy gradient algorithms by a factor of $O(1/\epsilon^{1/6})$. In addition, we also propose a variant of SRVR-PG with parameter exploration, which explores the initial policy parameter from a prior probability distribution. We conduct numerical experiments on classic control problems in reinforcement learning to validate the performance of our proposed algorithms.

The present paper develops a novel aggregated gradient approach for distributed machine learning that adaptively compresses the gradient communication. The key idea is to first quantize the computed gradients, and then skip less informative quantized gradient communications by reusing outdated gradients. Quantizing and skipping result in `lazy' worker-server communications, which justifies the term Lazily Aggregated Quantized gradient that is henceforth abbreviated as LAQ. Our LAQ can provably attain the same linear convergence rate as the gradient descent in the strongly convex case, while effecting major savings in the communication overhead both in transmitted bits as well as in communication rounds. Empirically, experiments with real data corroborate a significant communication reduction compared to existing gradient- and stochastic gradient-based algorithms.

Many problems in real life can be converted to combinatorial optimization problems (COPs) on graphs, that is to find a best node state configuration or a network structure such that the designed objective function is optimized under some constraints. However, these problems are notorious for their hardness to solve because most of them are NP-hard or NP-complete. Although traditional general methods such as simulated annealing (SA), genetic algorithms (GA) and so forth have been devised to these hard problems, their accuracy and time consumption are not satisfying in practice. In this work, we proposed a simple, fast, and general algorithm framework called Gumbel-softmax Optimization (GSO) for COPs. By introducing Gumbel-softmax technique which is developed in machine learning community, we can optimize the objective function directly by gradient descent algorithm regardless of the discrete nature of variables. We test our algorithm on four different problems including Sherrington-Kirkpatrick (SK) model, maximum independent set (MIS) problem, modularity optimization, and structural optimization problem. High-quality solutions can be obtained with much less time consuming compared to traditional approaches.

In this article, I will show you how to classify clothes from the Fashion MNIST data set using the python programming language and a machine learning technique called Artificial Neural Networks! If you prefer not to read this article and would like a video representation of it, you can check out the video below. It goes through everything in this article with a little more detail and will help make it easy for you to start programming your own Artificial Neural Network (ANN) model even if you don't have the programming language Python installed on your computer. Or you can use both the video and this article as supplementary materials for learning about ANN's! First I will write a description of what this program will do.

Covariance matrices have attracted attention for machine learning applications due to their capacity to capture interesting structure in the data. The main challenge is that one needs to take into account the particular geometry of the Riemannian manifold of symmetric positive definite (SPD) matrices they belong to. In the context of deep networks, several architectures for these matrices have recently been proposed. In our article, we introduce a Riemannian batch normalization (batchnorm) algorithm, which generalizes the one used in Euclidean nets. This novel layer makes use of geometric operations on the manifold, notably the Riemannian barycenter, parallel transport and non-linear structured matrix transformations. We derive a new manifold-constrained gradient descent algorithm working in the space of SPD matrices, allowing to learn the batchnorm layer. We validate our proposed approach with experiments in three different contexts on diverse data types: a drone recognition dataset from radar observations, and on emotion and action recognition datasets from video and motion capture data. Experiments show that the Riemannian batchnorm systematically gives better classification performance compared with leading methods and a remarkable robustness to lack of data.