NeurIPS 2024 Papers

The paper investigates the fundamental convergence properties of Sharpness-Aware Minimization (SAM), a recently proposed gradient-based optimization method (Foret et al., 2021) that significantly improves the generalization of deep neural networks. The convergence properties including the stationarity of accumulation points, the convergence of the sequence of gradients to the origin, the sequence of function values to the optimal value, and the sequence of iterates to the optimal solution are established for the method. The universality of the provided convergence analysis based on inexact gradient descent frameworks (Khanh et al., 2023b) allows its extensions to the normalized versions of SAM such as F-SAM (Li et al. 2024), VaSSO (Li & Giannakis, 2023), RSAM (Liu et al., 2022), and to the unnormalized versions of SAM such as USAM (Andriushchenko & Flammarion, 2022). Numerical experiments are conducted on classification tasks using deep learning models to confirm the practical aspects of our analysis.

Teaching is a potentially effective approach for understanding interactions among multiple intelligences. Previous explorations have convincingly shown that teaching presents additional opportunities for observation and demonstration within the learning model, such as data distillation and selection. However, the underlying optimization principles and convergence of interactive teaching lack theoretical analysis, and in this regard co-teaching serves as a notable prototype. In this paper, we discuss its role as a reduction of the larger loss landscape derived from Sharpness-Aware Minimization (SAM). Then, we classify it as an iterative parameter estimation process using Expectation-Maximization. The convergence of this typical interactive teaching is achieved by continuously optimizing a variational lower bound on the log marginal likelihood. This lower bound represents the expected value of the log posterior distribution of the latent variables under a scaled, factorized variational distribution. To further enhance interactive teaching's performance, we incorporate SAM's strong generalization information into interactive teaching, referred as Sharpness Reduction Interactive Teaching (SRIT). This integration can be viewed as a novel sequential optimization process. Finally, we validate the performance of our approach through multiple experiments.

A common source of anxiety for the computational neuroscience student is the question “will my recurrent neural network (RNN) model finally learn that task?”. Unlike in machine learning where any architectural modification of an RNN (e.g. GRU or LSTM) is acceptable if it speeds up training, the RNN models trained as _models of brain dynamics_ are subject to plausibility constraints that fundamentally exclude the usual machine learning hacks. The “vanilla” RNNs commonly used in computational neuroscience find themselves plagued by ill-conditioned loss surfaces that complicate training and significantly hinder our capacity to investigate the brain dynamics underlying complex tasks. Moreover, some tasks may require very long time horizons which backpropagation cannot handle given typical GPU memory limits. Here, we develop SOFO, a second-order optimizer that efficiently navigates loss surfaces whilst _not_ requiring backpropagation. By relying instead on easily parallelized batched forward-mode differentiation, SOFO enjoys constant memory cost in time. Morever, unlike most second-order optimizers which involve inherently sequential operations, SOFO's effective use of GPU parallelism yields a per-iteration wallclock time essentially on par with first-order gradient-based optimizers. We show vastly superior performance compared to Adam on a number of RNN tasks, including a difficult double-reaching motor task and the learning of an adaptive Kalman filter algorithm trained over a long horizon.

This paper studies simple bilevel problems, where a convex upper-level function is minimized over the optimal solutions of a convex lower-level problem. We first show the fundamental difficulty of simple bilevel problems, that the approximate optimal value of such problems is not obtainable by first-order zero-respecting algorithms. Then we follow recent works to pursue the weak approximate solutions. For this goal, we propose a novel method by reformulating them into functionally constrained problems. Our method achieves near-optimal rates for both smooth and nonsmooth problems. To the best of our knowledge, this is the first near-optimal algorithm that works under standard assumptions of smoothness or Lipschitz continuity for the objective functions.

Mean-field Langevin dynamics (MLFD) is a class of interacting particle methods that tackle convex optimization over probability measures on a manifold, which are scalable, versatile, and enjoy computational guarantees. However, some important problems -- such as risk minimization for infinite width two-layer neural networks, or sparse deconvolution -- are originally defined over the set of signed, rather than probability, measures. In this paper, we investigate how to extend the MFLD framework to convex optimization problems over signed measures. Among two known reductions from signed to probability measures -- the lifting and the bilevel approaches -- we show that the bilevel reduction leads to stronger guarantees and faster rates (at the price of a higher per-iteration complexity). In particular, we investigate the convergence rate of MFLD applied to the bilevel reduction in the low-noise regime and obtain two results. First, this dynamics is amenable to an annealing schedule, adapted from [Suzuki et al., 2023], that results in polynomial convergence rates to a fixed multiplicative accuracy. Second, we investigate the problem of learning a single neuron with the bilevel approach and obtain local exponential convergence rates that depend polynomially on the dimension and noise level (to compare with the exponential dependence that would result from prior analyses).

Performing eigendecomposition during neural network training is essential for tasks such as dimensionality reduction, network compression, image denoising, and graph learning. However, eigendecomposition is computationally expensive as it is orders of magnitude slower than other neural network operations. To address this challenge, we propose a novel approach called "amortized eigendecomposition" that relaxes the exact eigendecomposition by introducing an additional loss term called eigen loss. Our approach offers significant speed improvements by replacing the computationally expensive eigendecomposition with a more affordable QR decomposition at each iteration. Theoretical analysis guarantees that the desired eigenpair is attained as optima of the eigen loss. Empirical studies on nuclear norm regularization, latent-space principal component analysis, and graphs adversarial learning demonstrate significant improvements in training efficiency while producing nearly identical outcomes to conventional approaches. This novel methodology promises to integrate eigendecomposition efficiently into neural network training, overcoming existing computational challenges and unlocking new potential for advanced deep learning applications.

In minimax optimization, the extragradient (EG) method has been extensively studied because it outperforms the gradient descent-ascent method in convex-concave (C-C) problems. Yet, stochastic EG (SEG) has seen limited success in C-C problems, especially for unconstrained cases. Motivated by the recent progress of shuffling-based stochastic methods, we investigate the convergence of shuffling-based SEG in unconstrained finite-sum minimax problems, in search of convergent shuffling-based SEG. Our analysis reveals that both random reshuffling and the recently proposed flip-flop shuffling alone can suffer divergence in C-C problems. However, with an additional simple trick called anchoring, we develop the SEG with flip-flop anchoring (SEG-FFA) method which successfully converges in C-C problems. We also show upper and lower bounds in the strongly-convex-strongly-concave setting, demonstrating that SEG-FFA has a provably faster convergence rate compared to other shuffling-based methods.

Bayesian optimization is an effective technique for black-box optimization, but its applicability is typically limited to low-dimensional and small-budget problems due to the cubic complexity of computing the Gaussian process (GP) surrogate. While various approximate GP models have been employed to scale Bayesian optimization to larger sample sizes, most suffer from overly-smooth estimation and focus primarily on problems that allow for large online samples. In this work, we argue that Bayesian optimization algorithms with sparse GPs can more efficiently allocate their representational power to relevant regions of the search space. To achieve this, we propose focalized GP, which leverages a novel variational loss function to achieve stronger local prediction, as well as FocalBO, which hierarchically optimizes the focalized GP acquisition function over progressively smaller search spaces. Experimental results demonstrate that FocalBO can efficiently leverage large amounts of offline and online data to achieve state-of-the-art performance on robot morphology design and to control a 585-dimensional musculoskeletal system.

We consider nonconvex stochastic optimization problems in the asynchronous centralized distributed setup where the communication times from workers to a server can not be ignored, and the computation and communication times are potentially different for all workers. Using an unbiassed compression technique, we develop a new method—Shadowheart SGD—that provably improves the time complexities of all previous centralized methods. Moreover, we show that the time complexity of Shadowheart SGD is optimal in the family of centralized methods with compressed communication. We also consider the bidirectional setup, where broadcasting from the server to the workers is non-negligible, and develop a corresponding method.

In this paper, we show that applying adaptive methods directly to distributed minimax problems can result in non-convergence due to inconsistency in locally computed adaptive stepsizes. To address this challenge, we propose D-AdaST, a Distributed Adaptive minimax method with Stepsize Tracking. The key strategy is to employ an adaptive stepsize tracking protocol involving the transmission of two extra (scalar) variables. This protocol ensures the consistency among stepsizes of nodes, eliminating the steady-state error due to the lack of coordination of stepsizes among nodes that commonly exists in vanilla distributed adaptive methods, and thus guarantees exact convergence. For nonconvex-strongly-concave distributed minimax problems, we characterize the specific transient times that ensure time-scale separation of stepsizes and quasi-independence of networks, leading to a near-optimal convergence rate of $\tilde{\mathcal{O}} \left( \epsilon ^{-\left( 4+\delta \right)} \right)$ for any small $\delta > 0$, matching that of the centralized counterpart. To our best knowledge, D-AdaST is the *first* distributed adaptive method achieving near-optimal convergence without knowing any problem-dependent parameters for nonconvex minimax problems. Extensive experiments are conducted to validate our theoretical results.

Computational optimal transport (OT) has received massive interests in the machine learning community, and great advances have been gained in the direction of entropic-regularized OT. The Sinkhorn algorithm, as well as its many improved versions, has become the *de facto* solution to large-scale OT problems. However, most of the existing methods behave like first-order methods, which typically require a large number of iterations to converge. More recently, Newton-type methods using sparsified Hessian matrices have demonstrated promising results on OT computation, but there still remain a lot of unresolved open questions. In this article, we make major new progresses towards this direction: first, we propose a novel Hessian sparsification scheme that promises a strict control of the approximation error; second, based on this sparsification scheme, we develop a *safe* Newton-type method that is guaranteed to avoid singularity in computing the search directions; third, the developed algorithm has a clear implementation for practical use, avoiding most hyperparameter tuning; and remarkably, we provide rigorous global and local convergence analysis of the proposed algorithm, which is lacking in the prior literature. Various numerical experiments are conducted to demonstrate the effectiveness of the proposed algorithm in solving large-scale OT problems.

In this paper, we study a class of non-smooth non-convex problems in the form of $\min_{x}[\max_{y\in\mathcal Y}\phi(x, y) - \max_{z\in\mathcal Z}\psi(x, z)]$, where both $\Phi(x) = \max_{y\in\mathcal Y}\phi(x, y)$ and $\Psi(x)=\max_{z\in\mathcal Z}\psi(x, z)$ are weakly convex functions, and $\phi(x, y), \psi(x, z)$ are strongly concave functions in terms of $y$ and $z$, respectively. It covers two families of problems that have been studied but are missing single-loop stochastic algorithms, i.e., difference of weakly convex functions and weakly convex strongly-concave min-max problems. We propose a stochastic Moreau envelope approximate gradient method dubbed SMAG, the first single-loop algorithm for solving these problems, and provide a state-of-the-art non-asymptotic convergence rate. The key idea of the design is to compute an approximate gradient of the Moreau envelopes of $\Phi, \Psi$ using only one step of stochastic gradient update of the primal and dual variables. Empirically, we conduct experiments on positive-unlabeled (PU) learning and partial area under ROC curve (pAUC) optimization with an adversarial fairness regularizer to validate the effectiveness of our proposed algorithms.

In modern deep learning, it is common to warm up the learning rate $\eta$, often by a linear schedule between $\eta_{\text{init}} = 0$ and a predetermined target $\eta_{\text{trgt}}$. In this paper, we show through systematic experiments with SGD and Adam that the overwhelming benefit of warmup arises from allowing the network to tolerate larger $\eta_{\text{trgt}}$ by forcing the network to more well-conditioned areas of the loss landscape. The ability to handle larger target learning rates in turn makes hyperparameter tuning more robust while improving the final performance of the network. We uncover different regimes of operation during the warmup period, depending on whether the network training starts off in a progressive sharpening or sharpness reduction phase, which in turn depends on the initialization and parameterization. Using these insights, we show how $\eta_{\text{init}}$ can be properly chosen by utilizing the loss catapult mechanism, which saves on the number of warmup steps, in some cases completely eliminating the need for warmup. We also suggest an initialization for the variance in Adam, which provides benefits similar to warmup.

We propose adaptive, line-search-free second-order methods with optimal rate of convergence for solving convex-concave min-max problems. By means of an adaptive step size, our algorithms feature a simple update rule that requires solving only one linear system per iteration, eliminating the need for line-search or backtracking mechanisms. Specifically, we base our algorithms on the optimistic method and appropriately combine it with second-order information. Moreover, distinct from common adaptive schemes, we define the step size recursively as a function of the gradient norm and the prediction error in the optimistic update. We first analyze a variant where the step size requires knowledge of the Lipschitz constant of the Hessian. Under the additional assumption of Lipschitz continuous gradients, we further design a parameter-free version by tracking the Hessian Lipschitz constant locally and ensuring the iterates remain bounded. We also evaluate the practical performance of our algorithm by comparing it to existing second-order algorithms for minimax optimization.

First-order optimization (FOO) algorithms are pivotal in numerous computational domains, such as reinforcement learning and deep learning. However, their application to complex tasks often entails significant optimization inefficiency due to their need of many sequential iterations for convergence. In response, we introduce first-order optimization expedited with approximately parallelized iterations (OptEx), the first general framework that enhances the time efficiency of FOO by leveraging parallel computing to directly mitigate its requirement of many sequential iterations for convergence. To achieve this, OptEx utilizes a kernelized gradient estimation that is based on the history of evaluated gradients to predict the gradients required by the next few sequential iterations in FOO, which helps to break the inherent iterative dependency and hence enables the approximate parallelization of iterations in FOO. We further establish theoretical guarantees for the estimation error of our kernelized gradient estimation and the iteration complexity of SGD-based OptEx, confirming that the estimation error diminishes to zero as the history of gradients accumulates and that our SGD-based OptEx enjoys an effective acceleration rate of Θ(√N ) over standard SGD given parallelism of N, in terms of the sequential iterations required for convergence. Finally, we provide extensive empirical studies, including synthetic functions, reinforcement learning tasks, and neural network training on various datasets, to underscore the substantial efficiency improvements achieved by our OptEx in practice.

We consider learning a sparse model from linear measurements taken by a network of agents. Different from existing decentralized methods designed based on the LASSO regression with explicit $\ell_1$ norm regularization, we exploit the implicit regularization of decentralized optimization method applied to an over-parameterized nonconvex least squares formulation without penalization. Our first result shows that despite nonconvexity, if the network connectivity is good, the well-known decentralized gradient descent algorithm (DGD) with small initialization and early stopping can compute the statistically optimal solution. Sufficient conditions on the initialization scale, choice of step size, network connectivity, and stopping time are further provided to achieve convergence. Our result recovers the convergence rate of gradient descent in the centralized setting, showing its tightness. Based on the analysis of DGD, we further propose a communication-efficient version, termed T-DGD, by truncating the iterates before transmission. In the high signal-to-noise ratio (SNR) regime, we show that T-DGD achieves comparable statistical accuracy to DGD, while the communication cost is logarithmic in the number of parameters. Numerical results are provided to validate the effectiveness of DGD and T-DGD for sparse learning through implicit regularization.

Inverse Optimization (IO) is a framework for learning the unknown objective function of an expert decision-maker from a past dataset. In this paper, we extend the hypothesis class of IO objective functions to a reproducing kernel Hilbert space (RKHS), thereby enhancing feature representation to an infinite-dimensional space. We demonstrate that a variant of the representer theorem holds for a specific training loss, allowing the reformulation of the problem as a finite-dimensional convex optimization program. To address scalability issues commonly associated with kernel methods, we propose the Sequential Selection Optimization (SSO) algorithm to efficiently train the proposed Kernel Inverse Optimization (KIO) model. Finally, we validate the generalization capabilities of the proposed KIO model and the effectiveness of the SSO algorithm through learning-from-demonstration tasks on the MuJoCo benchmark.

We study online composite optimization under the Stochastically Extended Adversarial (SEA) model. Specifically, each loss function consists of two parts: a fixed non-smooth and convex regularizer, and a time-varying function which can be chosen either stochastically, adversarially, or in a manner that interpolates between the two extremes. In this setting, we show that for smooth and convex time-varying functions, optimistic composite mirror descent (OptCMD) can obtain an $\mathcal{O}(\sqrt{\sigma_{1:T}^2} + \sqrt{\Sigma_{1:T}^2})$ regret bound, where $\sigma_{1:T}^2$ and $\Sigma_{1:T}^2$ denote the cumulative stochastic variance and the cumulative adversarial variation of time-varying functions, respectively. For smooth and strongly convex time-varying functions, we establish an $\mathcal{O}((\sigma_{\max}^2 + \Sigma_{\max}^2)\log(\sigma_{1:T}^2 + \Sigma_{1:T}^2))$ regret bound, where $\sigma_{\max}^2$ and $\Sigma_{\max}^2$ denote the maximal stochastic variance and the maximal adversarial variation, respectively. For smooth and exp-concave time-varying functions, we achieve an $\mathcal{O}(d \log (\sigma_{1:T}^2 + \Sigma_{1:T}^2))$ bound where $d$ denotes the dimensionality. Moreover, to deal with the unknown function type in practical problems, we propose a multi-level \textit{universal} algorithm that is able to achieve the desirable bounds for three types of time-varying functions simultaneously. It should be noticed that all our findings match existing bounds for the SEA model without the regularizer, which implies that there is \textit{no price} in regret bounds for the benefits gained from the regularizer.

Protein optimization is a fundamental biological task aimed at enhancing theperformance of proteins by modifying their sequences. Computational methodsprimarily rely on evolutionary information (EI) encoded by protein languagemodels (PLMs) to predict fitness landscape for optimization. However, thesemethods suffer from a few limitations. (1) Evolutionary processes involve thesimultaneous consideration of multiple functional properties, often overshadowingthe specific property of interest. (2) Measurements of these properties tend to betailored to experimental conditions, leading to reduced generalizability of trainedmodels to novel proteins. To address these limitations, we introduce DenoisingProtein Language Models (DePLM), a novel approach that refines the evolutionaryinformation embodied in PLMs for improved protein optimization. Specifically, weconceptualize EI as comprising both property-relevant and irrelevant information,with the latter acting as “noise” for the optimization task at hand. Our approachinvolves denoising this EI in PLMs through a diffusion process conducted in therank space of property values, thereby enhancing model generalization and ensuringdataset-agnostic learning. Extensive experimental results have demonstrated thatDePLM not only surpasses the state-of-the-art in mutation effect prediction butalso exhibits strong generalization capabilities for novel proteins.