ICLR 2025 Papers

The \textbf{i}nexact \textbf{s}tochastic \textbf{p}roximal \textbf{p}oint \textbf{a}lgorithm (isPPA) is popular for solving stochastic composite optimization problems with many applications in machine learning. While the convergence theory of the (inexact) PPA has been well established, the known convergence guarantees of isPPA require restrictive assumptions. In this paper, we establish the stability and almost sure convergence of isPPA under mild assumptions, where smoothness and (restrictive) strong convexity of the objective function are not required. Imposing a local Lipschitz condition on component functions and a quadratic growth condition on the objective function, we establish last-iterate iteration complexity bounds of isPPA regarding the distance to the solution set and the Karush–Kuhn–Tucker (KKT) residual. Moreover, we show that the established iteration complexity bounds are tight up to a constant by explicitly analyzing the bounds for the regularized Fr\'echet mean problem. We further validate the established convergence guarantees of isPPA by numerical experiments.

Incremental gradient and incremental proximal methods are a fundamental class of optimization algorithms used for solving finite sum problems, broadly studied in the literature. Yet, without strong convexity, their convergence guarantees have primarily been established for the ergodic (average) iterate. We establish the first nonasymptotic convergence guarantees for the last iterate of both incremental gradient and incremental proximal methods, in general convex smooth (for both) and convex Lipschitz (for the proximal variants) settings. Our oracle complexity bounds for the last iterate nearly match (i.e., match up to a square-root-log or a log factor) the best known oracle complexity bounds for the average iterate, for both classes of methods. We further obtain generalizations of our results to weighted averaging of the iterates with increasing weights and for randomly permuted ordering of updates. We study last iterate convergence of the incremental proximal method as a mathematical abstraction of forgetting in continual learning and prove a lower bound that certifies that a large amount of regularization is crucial to mitigating catastrophic forgetting---one of the key considerations in continual learning. Our results generalize last iterate guarantees for incremental methods compared to state of the art, as such results were previously known only for overparameterized linear models, which correspond to convex quadratic problems with infinitely many solutions.

Training language models currently requires pre-determining a fixed compute budget because the typical cosine learning rate schedule depends on the total number of steps. In contrast, the Warmup-Stable-Decay (WSD) schedule uses a constant learning rate to produce a main branch of iterates that can in principle continue indefinitely without a pre-specified compute budget. Then, given any compute budget, one can branch out from the main branch at a proper time with a rapidly decaying learning rate to produce a strong model. Empirically, WSD generates an intriguing, non-traditional loss curve: the loss remains elevated during the stable phase but sharply declines during the decay phase. Towards explaining this phenomenon, we conjecture that pretraining loss exhibits a river valley landscape, which resembles a deep valley with a river at its bottom. Under this assumption, we show that during the stable phase, the iterate undergoes large oscillations due to the high learning rate, yet it progresses swiftly along the river. During the decay phase, the rapidly dropping learning rate minimizes the iterate’s oscillations, moving it closer to the river and revealing true optimization progress. Therefore, the sustained high learning rate phase and fast decaying phase are responsible for progress in the river and the mountain directions, respectively, and are both critical. Our analysis predicts phenomenons consistent with empirical observations and shows that this landscape can naturally emerge from pretraining on a simple bi-gram dataset. Inspired by the theory, we introduce WSD-S, a variant of WSD that reuses previous checkpoints’ decay phases and keeps only one main branch, where we resume from a decayed checkpoint. WSD-S empirically outperforms WSD and Cyclic-Cosine in obtaining multiple pretrained language model checkpoints across various compute budgets in a single run for parameters scaling from 0.1B to 1.2B.

We analyze two variants of Local Gradient Descent applied to distributed logistic regression with heterogeneous, separable data and show convergence at the rate $O(1/KR)$ for $K$ local steps and sufficiently large $R$ communication rounds. In contrast, all existing convergence guarantees for Local GD applied to any problem are at least $\Omega(1/R)$, meaning they fail to show the benefit of local updates. The key to our improved guarantee is showing progress on the logistic regression objective when using a large stepsize $\eta \gg 1/K$, whereas prior analysis depends on $\eta \leq 1/K$.

Imitation learning is a data-driven approach to learning policies from expert behavior, but it is prone to unreliable outcomes in out-of-sample (OOS) regions. While previous research relying on stable dynamical systems guarantees convergence to a desired state, it often overlooks transient behavior. We propose a framework for learning policies modeled by contractive dynamical systems, ensuring that all policy rollouts converge regardless of perturbations, and in turn, enable efficient OOS recovery. By leveraging recurrent equilibrium networks and coupling layers, the policy structure guarantees contractivity for any parameter choice, which facilitates unconstrained optimization. We also provide theoretical upper bounds for worst-case and expected loss to rigorously establish the reliability of our method in deployment. Empirically, we demonstrate substantial OOS performance improvements for simulated robotic manipulation and navigation tasks. See [sites.google.com/view/contractive-dynamical-policies](https://sites.google.com/view/contractive-dynamical-policies) for our codebase and highlight of the results.

While backpropagation effectively trains models, it presents challenges related to bio-plausibility, resulting in high memory demands and limited parallelism. Recently, Hinton (2022) proposed the Forward-Forward (FF) algorithm for high-parallel local updates. FF leverages squared sums as the local update target, termed goodness, and decouples goodness by normalizing the vector length to extract new features. However, this design encounters issues with feature scaling and deactivated neurons, limiting its application mainly to shallow networks. This paper proposes a novel goodness design utilizing **layer normalization** and **mean goodness** to overcome these challenges, demonstrating performance improvements even in 17-layer CNNs. Experiments on CIFAR-10, MNIST, and Fashion-MNIST show significant advantages over existing FF-based algorithms, highlighting the potential of FF in deep models. Furthermore, the model parallel strategy is proposed to achieve highly efficient training based on the property of local updates.

Decentralized bilevel optimization (DBO) provides a powerful framework for multi-agent systems to solve local bilevel tasks in a decentralized fashion without the need for a central server. However, most existing DBO methods rely on lower-level strong convexity (LLSC) to guarantee unique solutions and a well-defined hypergradient for stationarity measure, hindering their applicability in many practical scenarios not satisfying LLSC. To overcome this limitation, we introduce a new single-loop DBO algorithm called diminishing quadratically-regularized bilevel decentralized optimization (DUET), which eliminates the need for LLSC by introducing a diminishing quadratic regularization to the lower-level (LL) objective. We show that DUET achieves an iteration complexity of $O(1/T^{1-5p-\frac{11}{4}\tau})$ for approximate KKT-stationary point convergence under relaxed assumptions, where $p$ and $\tau $ are control parameters for LL learning rate and averaging, respectively. In addition, our DUET algorithm incorporates gradient tracking to address data heterogeneity, a key challenge in DBO settings. To the best of our knowledge, this is the first work to tackle DBO without LLSC under decentralized settings with data heterogeneity. Numerical experiments validate the theoretical findings and demonstrate the practical effectiveness of our proposed algorithms.

The success of modern machine learning is due in part to the adaptive optimization methods that have been developed to deal with the difficulties of training large models over complex datasets. One such method is gradient clipping: a practical procedure with limited theoretical underpinnings. In this work, we study clipping in a least squares problem under streaming SGD. We develop a theoretical analysis of the learning dynamics in the limit of large intrinsic dimension—a model and dataset dependent notion of dimensionality. In this limit we find a deterministic equation that describes the evolution of the loss and demonstrate that this equation predicts the path of clipped SGD on synthetic, CIFAR10, and Wikitext2 data. We show that with Gaussian noise clipping cannot improve SGD performance. Yet, in other noisy settings, clipping can provide benefits with tuning of the clipping threshold. We propose a simple heuristic for near optimal scheduling of the clipping threshold which requires the tuning of only one hyperparameter. We conclude with a discussion about the links between high-dimensional clipping and neural network training.

Beyond the worst-case analysis of algorithms, the learning-augmented framework considers that an algorithm can leverage possibly imperfect predictions about the unknown variables to have guarantees tied to the prediction quality. We consider online non-clairvoyant scheduling to minimize the max-stretch under this framework, where the scheduler can access job size predictions. We present a family of algorithms: Relaxed-Greedy (RG) with an $O(\eta^3 \cdot \sqrt{P})$ competitive ratio, where $\eta$ denotes the prediction error for job sizes and $P$ the maximum job size ratio; Adaptive Relaxed-Greedy with an $O(\lambda^{0.5} \cdot \eta^{2.5} \cdot \sqrt{P})$ competitive ratio, where $\lambda$ denotes the error for the minimum job size; Predictive Relaxed-Greedy with an $O(\lambda^{0.5} \cdot \varphi^{0.5} \cdot \eta \cdot \max \\\{ \eta, \varphi \\\} \cdot \sqrt{P})$ competitive ratio, where $\varphi$ denotes the error for the maximum job size. We also present *${RG}^x$*, an algorithm that represents a trade-off between consistency and smoothness, with an $O(\eta^{2+2x} \cdot P^{1-x})$ competitive ratio. We introduce a general method using resource augmentation to bound robustness, resulting in *RR*-augmented *RG*, with a $(1 + \epsilon)$-speed $O(\min \\\{ \eta^3 \sqrt{P}, \frac{n}{\epsilon} \\\})$ competitive ratio. Finally, we conduct simulations on synthetic and real-world datasets to evaluate the practical performance of these algorithms.

We consider online scheduling with class constraints. That is, we are given $m$ machines, each with $k$ class slots. Upon receiving a job $j$ with class $c_j$, an algorithm needs to allocate $j$ on some machine $i$. The goal is to minimize the makespan while not assigning more than $k$ different classes onto each machine. While the offline case is well understood and even (E)PTAS results are known [Jansen, Lassota, Maack SPAA'20, Chen Jansen Luo Zhang COCOA'16], the online case admits strong impossibility results in classical competitive analysis [Epstein, Lassota, Levin, Maack, Rohwedder STACS'22]. We overcome these daunting results by investigating the problem in a learning-augmented setting where an algorithm can access possibly erroneous predictions. We present new algorithms with competitive ratios independent of $m$ and tight lower bounds for several classical and problem-specific prediction models. We thereby give a structured overview of what additional information helps in the design of better scheduling algorithms.

Adversarial training is a widely-applied approach to training deep neural networks to be robust against adversarial perturbation. However, although adversarial training has achieved empirical success in practice, it still remains unclear why adversarial examples exist and how adversarial training methods improve model robustness. In this paper, we provide a theoretical understanding of adversarial examples and adversarial training algorithms from the perspective of feature learning theory. Specifically, we focus on a multiple classification setting, where the structured data can be composed of two types of features: the robust features, which are resistant to perturbation but sparse, and the non-robust features, which are susceptible to perturbation but dense. We train a two-layer smoothed ReLU convolutional neural network to learn our structured data. First, we prove that by using standard training (gradient descent over the empirical risk), the network learner primarily learns the non-robust feature rather than the robust feature, which thereby leads to the adversarial examples that are generated by perturbations aligned with negative non-robust feature directions. Then, we consider the gradient-based adversarial training algorithm, which runs gradient ascent to find adversarial examples and runs gradient descent over the empirical risk at adversarial examples to update models. We show that the adversarial training method can provably strengthen the robust feature learning and suppress the non-robust feature learning to improve the network robustness. Finally, we also empirically validate our theoretical findings with experiments on real-image datasets, including MNIST, CIFAR10 and SVHN.

This paper studies second-order methods for convex-concave minimax optimization. Monteiro & Svaiter (2012) proposed a method to solve the problem with an optimal iteration complexity of $\mathcal{O}(\epsilon^{-3/2})$ to find an $\epsilon$-saddle point. However, it is unclear whether the computational complexity, $\mathcal{O}((N+ d^2) d \epsilon^{-2/3})$, can be improved. In the above, we follow Doikov et al. (2023) and assume the complexity of obtaining a first-order oracle as $N$ and the complexity of obtaining a second-order oracle as $dN$. In this paper, we show that the computation cost can be reduced by reusing Hessian across iterations. Our methods take the overall computational complexity of $\tilde{\mathcal{O}}( (N+d^2)(d+ d^{2/3}\epsilon^{-2/3}))$, which improves those of previous methods by a factor of $d^{1/3}$. Furthermore, we generalize our method to strongly-convex-strongly-concave minimax problems and establish the complexity of $\tilde{\mathcal{O}}((N+d^2) (d + d^{2/3} \kappa^{2/3}) )$ when the condition number of the problem is $\kappa$, enjoying a similar speedup upon the state-of-the-art method. Numerical experiments on both real and synthetic datasets also verify the efficiency of our method.

We consider neural networks (NNs) where the final layer is down-scaled by a fixed hyperparameter $\gamma$. Recent work has identified $\gamma$ as controlling the strength of feature learning. As $\gamma$ increases, network evolution changes from "lazy" kernel dynamics to "rich" feature-learning dynamics, with a host of associated benefits including improved performance on common tasks. In this work, we conduct a thorough empirical investigation of the effect of scaling $\gamma$ across a variety of models and datasets in the online training setting. We first examine the interaction of $\gamma$ with the learning rate $\eta$, identifying several scaling regimes in the $\gamma$-$\eta$ plane which we explain theoretically using a simple model. We find that the optimal learning rate $\eta^*$ scales non-trivially with $\gamma$. In particular, $\eta^* \propto \gamma^2$ when $\gamma \ll 1$ and $\eta^* \propto \gamma^{2/L}$ when $\gamma \gg 1$ for a feed-forward network of depth $L$. Using this optimal learning rate scaling, we proceed with an empirical study of the under-explored ``ultra-rich'' $\gamma \gg 1$ regime. We find that networks in this regime display characteristic loss curves, starting with a long plateau followed by a drop-off, sometimes followed by one or more additional staircase steps. We find networks of different large $\gamma$ values optimize along similar trajectories up to a reparameterization of time. We further find that optimal online performance is often found at large $\gamma$ and could be missed if this hyperparameter is not tuned. Our findings indicate that analytical study of the large-$\gamma$ limit may yield useful insights into the dynamics of representation learning in performant models.

We propose Adam-mini, an optimizer that achieves on-par or better performance than AdamW with $50$% less memory footprint. Adam-mini reduces memory by cutting down the learning rate resources in Adam (i.e., $1/\sqrt{v}$). By delving into the Hessian structure of neural nets, we find Adam’s $v$ might not function at its full potential as effectively as we expected. We find that $\geq 99.9$% of these learning rates in $v$ could be harmlessly removed if we (1) carefully partition the parameters into blocks following our proposed principle on Hessian structure; (2) assign a single but good learning rate to each parameter block. We then provide one simple way to find good learning rates and propose Adam-mini. Empirically, we verify that Adam-mini performs on par or better than AdamW on various language models sized from 39M to 13B for pre-training, supervised fine-tuning, and RLHF. The reduced memory footprint of Adam-mini also alleviates communication overheads among GPUs, thereby increasing throughput. For instance, Adam-mini achieves $49.6$% higher throughput than AdamW when pre-training Llama 2-7B on $2\times$ A800-80GB GPUs, which saves 33% wall-clock time for pre-training.

Pretraining large language models often requires significant computational resources and memory due to their vast parameter amount. An effective approach to enhance parameter efficiency in both training and inference is to parameterize each full-size weight as the product of two trainable low-rank factors. While low-rank fine-tuning has achieved great success, low-rank pretraining remains challenging as it requires learning extensive knowledge from scratch under the restrictive low-rank parameterization. During standard low-rank pretraining, separately optimizing the low-rank factors introduces redundant information from the full gradient, which hinders the learning process. To achieve efficient yet effective low-rank pretraining, we propose a **Lo**w-rank **R**iemannian **O**ptimizer (**LORO**). At each LORO update step, the low-rank factor pairs are jointly updated to ensure their full-size product moves along the steepest descent direction on the low-rank manifold, without the need to compute any memory-intensive full-size matrices or gradients. Hence, our LORO finds low-rank models that achieve high performance comparable to full-size pretrained models, while significantly reducing memory usage and accelerating both training and inference. A LLaMA 1B model pretrained with LORO achieves a perplexity score of 2\% better than the full-size baseline, with a 54\% reduction in model memory, a $\times1.8$ speedup in training, and a $\times2.2$ speedup in inference. The code is available on https://github.com/mzf666/LORO-main.

Quadratic programming (QP) forms a crucial foundation in optimization, appearing in a broad spectrum of domains and serving as the basis for more advanced algorithms. Consequently, as the scale and complexity of modern applications continue to grow, the development of efficient and reliable QP algorithms becomes increasingly vital. In this context, this paper introduces a novel deep learning-aided distributed optimization architecture designed for tackling large-scale QP problems. First, we combine the state-of-the-art Operator Splitting QP (OSQP) method with a consensus approach to derive **DistributedQP**, a new method tailored for network-structured problems, with convergence guarantees to optimality. Subsequently, we unfold this optimizer into a deep learning framework, leading to **DeepDistributedQP**, which leverages learned policies to accelerate reaching to desired accuracy within a restricted amount of iterations. Our approach is also theoretically grounded through Probably Approximately Correct (PAC)-Bayes theory, providing generalization bounds on the expected optimality gap for unseen problems. The proposed framework, as well as its centralized version **DeepQP**, significantly outperform their standard optimization counterparts on a variety of tasks such as randomly generated problems, optimal control, linear regression, transportation networks and others. Notably, DeepDistributedQP demonstrates strong generalization by training on small problems and scaling to solve much larger ones (up to 50K variables and 150K constraints) using the same policy. Moreover, it achieves orders-of-magnitude improvements in wall-clock time compared to OSQP. The certifiable performance guarantees of our approach are also demonstrated, ensuring higher-quality solutions over traditional optimizers.

Neural pruning aims to compress and accelerate deep neural networks by identifying the optimal subnetwork within a specified sparsity budget. In this work, we study how to gradually sparsify the unpruned dense model to the target sparsity level with minimal performance drop. Specifically, we analyze the evolution of the population of optimal subnetworks under continuous sparsity increments from a thermodynamic perspective. We first reformulate neural pruning as an expected loss minimization problem over the mask distributions. Then, we establish an effective approximation for the sparsity evolution of the optimal mask distribution, termed the **S**parsity Evolutionary **F**okker-**P**lanck-**K**olmogorov Equation (**SFPK**), which provides closed-form, mathematically tractable guidance on distributional transitions for minimizing the expected loss under an infinitesimal sparsity increment. On top of that, we propose SFPK-pruner, a particle simulation-based probabilistic pruning method, to sample performant masks with desired sparsity from the destination distribution of SFPK. In theory, we establish the convergence guarantee for the proposed SFPK-pruner. Our SFPK-pruner exhibits competitive performance in various pruning scenarios. The code is available on https://github.com/mzf666/SFPK-main.

In this paper, we study the data-dependent convergence and generalization behavior of gradient methods for neural networks with smooth activation. Our first result is a novel bound on the excess risk of deep networks trained by the logistic loss via an alogirthmic stability analysis. Compared to previous works, our results improve upon the shortcomings of the well-established Rademacher complexity-based bounds. Importantly, the bounds we derive in this paper are tighter, hold even for neural networks of small width, do not scale unfavorably with width, are algorithm-dependent, and consequently capture the role of initialization on the sample complexity of gradient descent for deep nets. Specialized to noiseless data separable with margin $\gamma$ by neural tangent kernel (NTK) features of a network of width $\Omega(poly(\log(n)))$, we show the test-error rate $e^{O(L)}/{\gamma^2 n}$, where $n$ is the training set size and $L$ denotes the number of hidden layers. This results in an improvement in the test loss bound compared to previous works while maintaining the poly-logarithmic width conditions. We further investigate excess risk bounds for deep nets trained with noisy data, establishing that under a polynomial condition on the network width, gradient descent can achieve the optimal excess risk. Finally, we show that a large step-size significantly improves upon the NTK regime's results in classifying the XOR distribution. In particular, we show for a one-hidden layer neural network of constant width $m$ with quadratic activation and standard Gaussian initialization that SGD with linear sample complexity and with a large step-size $\eta=m$ reaches the perfect test accuracy after only $\lceil\log(d)\rceil$ iterations, where $d$ is the data dimension.

Black-box optimization algorithms have been widely used in various machine learning problems, including reinforcement learning and prompt fine-tuning. However, directly optimizing the training loss value, as commonly done in existing black-box optimization methods, could lead to suboptimal model quality and generalization performance. To address those problems in black-box optimization, we propose a novel Sharpness-Aware Black-box Optimization (SABO) algorithm, which applies a sharpness-aware minimization strategy to improve the model generalization. Specifically, the proposed SABO method first reparameterizes the objective function by its expectation over a Gaussian distribution. Then it iteratively updates the parameterized distribution by approximated stochastic gradients of the maximum objective value within a small neighborhood around the current solution in the Gaussian distribution space. Theoretically, we prove the convergence rate and generalization bound of the proposed SABO algorithm. Empirically, extensive experiments on the black-box prompt fine-tuning tasks demonstrate the effectiveness of the proposed SABO method in improving model generalization performance.