Despite the vast empirical success of neural networks, theoretical understanding of the training procedures remains limited, especially in providing performance guarantees of testing performance due to the non-convex nature of the optimization problem. The current paper investigates an alternative approach of neural network training by reducing to another problem with convex structure -- to solve a monotone variational inequality (MVI) -- inspired by a recent work of (Juditsky & Nemirovsky, 2019). The solution to MVI can be found by computationally efficient procedures, and importantly, this leads to performance guarantee of $\ell_2$ and $\ell_{\infty}$ bounds on model recovery and prediction accuracy under the theoretical setting of training a single-layer linear neural network. In addition, we study the use of MVI for training multi-layer neural networks and propose a practical algorithm called \textit{stochastic variational inequality} (SVI), and demonstrate its applicability in training fully-connected neural networks and graph neural networks (GNN) (SVI is completely general and can be used to train other types of neural networks). We demonstrate the competitive or better performance of SVI compared to widely-used stochastic gradient descent methods on both synthetic and real network data prediction tasks regarding various performance metrics, especially in the improved efficiency in the early stage of training.
相關內容
The goal of this survey is to present an explanatory review of the approximation properties of deep neural networks. Specifically, we aim at understanding how and why deep neural networks outperform other classical linear and nonlinear approximation methods. This survey consists of three chapters. In Chapter 1 we review the key ideas and concepts underlying deep networks and their compositional nonlinear structure. We formalize the neural network problem by formulating it as an optimization problem when solving regression and classification problems. We briefly discuss the stochastic gradient descent algorithm and the back-propagation formulas used in solving the optimization problem and address a few issues related to the performance of neural networks, including the choice of activation functions, cost functions, overfitting issues, and regularization. In Chapter 2 we shift our focus to the approximation theory of neural networks. We start with an introduction to the concept of density in polynomial approximation and in particular study the Stone-Weierstrass theorem for real-valued continuous functions. Then, within the framework of linear approximation, we review a few classical results on the density and convergence rate of feedforward networks, followed by more recent developments on the complexity of deep networks in approximating Sobolev functions. In Chapter 3, utilizing nonlinear approximation theory, we further elaborate on the power of depth and approximation superiority of deep ReLU networks over other classical methods of nonlinear approximation.
As opaque predictive models increasingly impact many areas of modern life, interest in quantifying the importance of a given input variable for making a specific prediction has grown. Recently, there has been a proliferation of model-agnostic methods to measure variable importance (VI) that analyze the difference in predictive power between a full model trained on all variables and a reduced model that excludes the variable(s) of interest. A bottleneck common to these methods is the estimation of the reduced model for each variable (or subset of variables), which is an expensive process that often does not come with theoretical guarantees. In this work, we propose a fast and flexible method for approximating the reduced model with important inferential guarantees. We replace the need for fully retraining a wide neural network by a linearization initialized at the full model parameters. By adding a ridge-like penalty to make the problem convex, we prove that when the ridge penalty parameter is sufficiently large, our method estimates the variable importance measure with an error rate of $O(\frac{1}{\sqrt{n}})$ where $n$ is the number of training samples. We also show that our estimator is asymptotically normal, enabling us to provide confidence bounds for the VI estimates. We demonstrate through simulations that our method is fast and accurate under several data-generating regimes, and we demonstrate its real-world applicability on a seasonal climate forecasting example.
In recent years, change point detection for high dimensional data has become increasingly important in many scientific fields. Most literature develop a variety of separate methods designed for specified models (e.g. mean shift model, vector auto-regressive model, graphical model). In this paper, we provide a unified framework for structural break detection which is suitable for a large class of models. Moreover, the proposed algorithm automatically achieves consistent parameter estimates during the change point detection process, without the need for refitting the model. Specifically, we introduce a three-step procedure. The first step utilizes the block segmentation strategy combined with a fused lasso based estimation criterion, leads to significant computational gains without compromising the statistical accuracy in identifying the number and location of the structural breaks. This procedure is further coupled with hard-thresholding and exhaustive search steps to consistently estimate the number and location of the break points. The strong guarantees are proved on both the number of estimated change points and the rates of convergence of their locations. The consistent estimates of model parameters are also provided. The numerical studies provide further support of the theory and validate its competitive performance for a wide range of models. The developed algorithm is implemented in the R package LinearDetect.
Zeroth-order optimization methods are developed to overcome the practical hurdle of having knowledge of explicit derivatives. Instead, these schemes work with merely access to noisy functions evaluations. The predominant approach is to mimic first-order methods by means of some gradient estimator. The theoretical limitations are well-understood, yet, as most of these methods rely on finite-differencing for shrinking differences, numerical cancellation can be catastrophic. The numerical community developed an efficient method to overcome this by passing to the complex domain. This approach has been recently adopted by the optimization community and in this work we analyze the practically relevant setting of dealing with computational noise. To exemplify the possibilities we focus on the strongly-convex optimization setting and provide a variety of non-asymptotic results, corroborated by numerical experiments, and end with local non-convex optimization.
Information over-squashing is a phenomenon of inefficient information propagation between distant nodes on networks. It is an important problem that is known to significantly impact the training of graph neural networks (GNNs), as the receptive field of a node grows exponentially. To mitigate this problem, a preprocessing procedure known as rewiring is often applied to the input network. In this paper, we investigate the use of discrete analogues of classical geometric notions of curvature to model information flow on networks and rewire them. We show that these classical notions achieve state-of-the-art performance in GNN training accuracy on a variety of real-world network datasets. Moreover, compared to the current state-of-the-art, these classical notions exhibit a clear advantage in computational runtime by several orders of magnitude.
We consider the problem of parameter estimation in slowly varying regression models with sparsity constraints. We formulate the problem as a mixed integer optimization problem and demonstrate that it can be reformulated exactly as a binary convex optimization problem through a novel exact relaxation. The relaxation utilizes a new equality on Moore-Penrose inverses that convexifies the non-convex objective function while coinciding with the original objective on all feasible binary points. This allows us to solve the problem significantly more efficiently and to provable optimality using a cutting plane-type algorithm. We develop a highly optimized implementation of such algorithm, which substantially improves upon the asymptotic computational complexity of a straightforward implementation. We further develop a heuristic method that is guaranteed to produce a feasible solution and, as we empirically illustrate, generates high quality warm-start solutions for the binary optimization problem. We show, on both synthetic and real-world datasets, that the resulting algorithm outperforms competing formulations in comparable times across a variety of metrics including out-of-sample predictive performance, support recovery accuracy, and false positive rate. The algorithm enables us to train models with 10,000s of parameters, is robust to noise, and able to effectively capture the underlying slowly changing support of the data generating process.
This work characterizes the effect of depth on the optimization landscape of linear regression, showing that, despite their nonconvexity, deeper models have more desirable optimization landscape. We consider a robust and over-parameterized setting, where a subset of measurements are grossly corrupted with noise and the true linear model is captured via an $N$-layer linear neural network. On the negative side, we show that this problem \textit{does not} have a benign landscape: given any $N\geq 1$, with constant probability, there exists a solution corresponding to the ground truth that is neither local nor global minimum. However, on the positive side, we prove that, for any $N$-layer model with $N\geq 2$, a simple sub-gradient method becomes oblivious to such ``problematic'' solutions; instead, it converges to a balanced solution that is not only close to the ground truth but also enjoys a flat local landscape, thereby eschewing the need for "early stopping". Lastly, we empirically verify that the desirable optimization landscape of deeper models extends to other robust learning tasks, including deep matrix recovery and deep ReLU networks with $\ell_1$-loss.
Claiming causal inferences in network settings necessitates careful consideration of the often complex dependency between outcomes for actors. Of particular importance are treatment spillover or outcome interference effects. We consider causal inference when the actors are connected via an underlying network structure. Our key contribution is a model for causality when the underlying network is unobserved and the actor covariates evolve stochastically over time. We develop a joint model for the relational and covariate generating process that avoids restrictive separability assumptions and deterministic network assumptions that do not hold in the majority of social network settings of interest. Our framework utilizes the highly general class of Exponential-family Random Network models (ERNM) of which Markov Random Fields (MRF) and Exponential-family Random Graph models (ERGM) are special cases. We present potential outcome based inference within a Bayesian framework, and propose a simple modification to the exchange algorithm to allow for sampling from ERNM posteriors. We present results of a simulation study demonstrating the validity of the approach. Finally, we demonstrate the value of the framework in a case-study of smoking over time in the context of adolescent friendship networks.
Recently, neural networks have been widely applied for solving partial differential equations (PDEs). However, with current training algorithms the numerical convergence of neural networks when solving PDEs has not been empirically observed. The primary difficulty lies in solving the highly non-convex optimization problems resulting from the neural network discretization. Theoretically analyzing the optimization process presents significant difficulties and empirical experiments require extensive hyperparameter tuning to achieve acceptable results. In order to conquer this challenge, we develop a novel greedy training algorithm for shallow neural networks in this paper. We also analyze the resulting method and obtain a priori error bounds when solving PDEs from the function class defined by shallow networks. This rigorously establishes the convergence of the method as the network size increases. Finally, we test the algorithm on several benchmark examples, including high dimensional PDEs, to confirm the theoretical convergence rate and to establish its efficiency and robustness. An advantage of this method is its straightforward applicability to high-order equations on general domains.
Sampling methods (e.g., node-wise, layer-wise, or subgraph) has become an indispensable strategy to speed up training large-scale Graph Neural Networks (GNNs). However, existing sampling methods are mostly based on the graph structural information and ignore the dynamicity of optimization, which leads to high variance in estimating the stochastic gradients. The high variance issue can be very pronounced in extremely large graphs, where it results in slow convergence and poor generalization. In this paper, we theoretically analyze the variance of sampling methods and show that, due to the composite structure of empirical risk, the variance of any sampling method can be decomposed into \textit{embedding approximation variance} in the forward stage and \textit{stochastic gradient variance} in the backward stage that necessities mitigating both types of variance to obtain faster convergence rate. We propose a decoupled variance reduction strategy that employs (approximate) gradient information to adaptively sample nodes with minimal variance, and explicitly reduces the variance introduced by embedding approximation. We show theoretically and empirically that the proposed method, even with smaller mini-batch sizes, enjoys a faster convergence rate and entails a better generalization compared to the existing methods.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191