We revisit the convergence analysis of constant stepsize stochastic approximation (SA) with decision-dependent Markovian noise, with a focus on characterizing the stationary bias against the root of the mean-field equation. We first establish the finite-time p-th moment bounds for the SA iterates in a general decision-dependent setting, which serve as a stability foundation for the subsequent analysis. Building on this foundation, and leveraging a local regularity condition termed Poisson–Gateaux differentiability (WD^∗) for the solution to Poisson equation induced by the decision-dependent Markov kernel, we show that the stationary bias is of the order 𝒪(α) for a broad class of decision-dependent settings. Additionally, we establish geometric weak convergence of the joint SA process towards a unique stationary distribution, and a functional central limit theorem. Our relaxed regularity condition enables us to cover cases of non-smooth kernels such as acceptance–rejection mechanisms, projected Langevin dynamics, and clipped state dynamics.

Learning graph structures from data is a fundamental problem that spans a wide range of signal processing and machine learning tasks. While significant effort has been made to tackle the problem, existing research has largely evolved along two parallel directions. The first seeks to infer the topology of an individual graph from observations supported on it, whereas the second seeks to learn a generative distribution from observed graph instances, enabling the sampling of new graphs. This review presents a unified framework that connects these formulations by viewing them as inverse problems of a common generation process for graph data. We review the major methodologies within this framework, highlight their relationships, strengths, and limitations, and identify opportunities for integrating ideas across paradigms. By bridging graph topology learning and graph generation, this review provides a broader cross-disciplinary perspective on the field and outlines promising directions for future research.

This paper introduces a robust two-timescale compressed primal-dual (TiCoPD) algorithm tailored for decentralized optimization under bandwidth-limited and unreliable channels. By integrating the majorization-minimization approach with the primal-dual optimization framework, the TiCoPD algorithm strategically compresses the difference term shared among agents to enhance communication efficiency and robustness against noisy channels without compromising convergence stability. The method incorporates a mirror sequence for agent consensus on nonlinearly compressed terms updated on a fast timescale, together with a slow timescale primal-dual recursion for optimizing the objective function. Our analysis demonstrates that the proposed algorithm converges to a stationary solution when the objective function is smooth but possibly non-convex. Numerical experiments corroborate the conclusions of this paper.

Learning the graph underlying a networked system from nodal signals is crucial to downstream tasks in graph signal processing and machine learning. The presence of hidden nodes whose signals are not observable might corrupt the estimated graph. While existing works proposed various robustifications of vanilla graph learning objectives by explicitly accounting for the presence of these hidden nodes, a robustness analysis of "naive", hidden-node agnostic approaches is still underexplored. This work demonstrates that vanilla graph topology learning methods are implicitly robust to partial observations of low-pass filtered graph signals. We achieve this theoretical result through extending the restricted isometry property (RIP) to the Dirichlet energy function used in graph learning objectives. We show that smoothness-based graph learning formulation (e.g., the GL-SigRep method) on partial observations can recover the ground truth graph topology corresponding to the observed nodes. Synthetic and real data experiments corroborate our findings.

Learning graph topology of complex networks is challenging due to limited data availability and imprecise data models. Different from prior works that focus on structural priors with explicit control on macroscopic properties such as sparsity, this paper proposes a novel functional prior approach for graph topology learning. We postulate that complex networks are inherently optimized to perform a certain task (e.g., social networks specialize at optimizing a welfare function, biological networks are resilient towards node/edge deletion), which can be incorporated as a regularizer to assist in graph learning. Mathematically, we formulate a bilevel optimization problem where the lower-level problem solves the associated task on a candidate graph topology and the upper-level problem trades off between data fitting and task performance. We design a two-timescale gradient descent (TTGD) algorithm and show that under verifiable conditions, it finds a stationary point to the bilevel graph learning problem with a sublinear convergence rate. We provide theoretical insights on the graph topology learned from the functional priors and show that the resulting regularizers subsume a broad class of graph filter regularizers, including polynomial graph regularizers as special cases. We show via extensive experiments on synthetic and real datasets that the proposed formulation gives rise to reliable estimates of graph topology, even with insufficient data.

In this paper, we investigate the problem of decentralized online resource allocation in the presence of Byzantine attacks. In this problem setting, some agents may be compromised due to external manipulations or internal failures, causing them to behave maliciously and disrupt the resource allocation process by sending incorrect messages to their neighbors. Given the non-consensual nature of the resource allocation problem, we formulate it under a primal-dual optimization framework, in which the dual variables are aggregated among the agents, enabling the incorporation of robust aggregation rules to mitigate Byzantine attacks. By leveraging the classical Byzantine attack model, we propose a class of Byzantine-resilient decentralized online resource allocation algorithms that judiciously integrate the adaptive robust clipping technique with the existing robust aggregation rules to filter out malicious messages. We establish theoretical guarantees, showing that the proposed algorithms achieve tight linear dynamic regret and accumulative constraint violation bounds, where the constants depend on the properties of robust aggregation rules. Numerical experiments on decentralized online economic dispatch validate the effectiveness of our approach and support our theoretical results.

We consider the distributed optimization problem with data dispersed across multiple workers under the orchestration of a parameter server. In distributed environments, variations in computation speeds and network conditions across workers often lead to significant idle times in synchronous training. Although asynchronous training has been widely explored to reduce the synchronization overhead, existing methods either assume bounded dissimilarity among workers' local data, which hampers performance under high data heterogeneity, or rely on worker scheduling strategies that limit system asynchrony. This work proposes the dual-delayed stochastic gradient descent (DuDe-SGD) algorithm to overcome the above limitations. Through a server-side buffer architecture, DuDe-SGD makes use of stale stochastic gradients from all workers to neutralize the effects of data heterogeneity while maintaining full asynchrony and per-iteration computation cost on par with traditional asynchronous stochastic gradient descent (SGD) algorithms. Our analysis demonstrates that DuDe-SGD achieves a comparable convergence rate for smooth nonconvex problems as state-of-the-art asynchronous SGD algorithms, even with arbitrarily heterogeneous data without adopting any worker scheduling schemes. Numerical experiments demonstrate the favorable performance of DuDe-SGD compared with existing synchronous and asynchronous SGD-based algorithms, especially in scenarios with highly heterogeneous data.

This paper aims to accelerate decentralized optimization by strategically designing the edge weights used in the agent-to-agent message exchanges. We propose a Dynamic Directed Decentralized Gradient (D3GD) framework and show that the proposed data-dependent framework is a practical alternative to the classical directed DGD (Di-DGD) algorithm for learning on directed graphs. To obtain a strategy for edge weights refinement, we derive a design function inspired by the cost-to-go function in a new convergence analysis for Di-DGD. This results in a data-dependent dynamical design for the edge weights. A fully decentralized version of D3GD is developed such that each agent refines its communication strategy using only neighbor's information. Numerical experiments show that D3GD accelerates convergence towards stationary solution by 30-40% over Di-DGD, and learns edge weights that adapt to data similarity.

We study asynchronous optimization for finite-sum eigenspace computation in heterogeneous distributed systems. The theoretical foundations for asynchronous eigenspace computation remain scarce, with existing approaches offering limited coverage of dynamics directly on the Grassmannian under stale information. In this paper, we propose a Grassmannian incremental aggregation method that refreshes only arriving components and reuses cached gradients, retaining low per-update cost without global synchronization. The method employs an extrinsic polar update that preserves the intrinsic subspace geometry without requiring parallel transport of stale tangent vectors. Our analysis establishes a tight angle-dependent gradient-dominance characterization of the objective and a basin-invariance property for stale aggregated updates. These yield two-phase linear convergence, comprising an explicit broad-basin regime and a sharper local regime, with constants controlled by component spectral spreads. Experiments on serial and distributed PCA demonstrate improved sample efficiency and wall-clock convergence over representative baselines.

Lookahead-based acceleration methods, such as Nesterov's momentum, are widely used in optimization, but they often become unreliable in deep learning training mainly due to stochastic gradient noise and non-convex loss landscapes. In particular, standard lookahead relies on short-horizon update signals (e.g., differences between consecutive iterates), which are inherently noisy and can lead to unstable extrapolation directions. This work revisits Nesterov's acceleration from a trajectory perspective and argues that effective acceleration in deep learning should harness the low-frequency trends of optimization trajectories rather than extrapolating noisy one-step updates. Leveraging this insight, we propose EMA-Nesterov, a simple modification that replaces the standard Nesterov's lookahead direction with an exponential moving average (EMA) of parameter updates. This yields a stabilized lookahead direction that captures and harnesses the evolving trend of the training trajectory through a low-pass filter, while remaining adaptive to progressive changes via the geometric weighting structure of EMA. We show that EMA-Nesterov retains a theoretical accelerated convergence rate in convex problems that is analogous to Nesterov's accelerated gradient method. Furthermore, we provide empirical evidence on language model pre-training to verify that EMA-Nesterov is broadly applicable across a range of fine-tuned base optimizers, including Adam, SOAP, Muon, as well as complex optimizers that achieve state-of-the-art performance on optimization benchmarks (NanoGPT). Compared to prior lookahead methods, EMA-Nesterov achieves better performance by avoiding the instability of short-horizon lookahead and the non-adaptivity of long-horizon lookahead.

The software and data in this repository are associated with the paper: Xiaolu Wang, Yuchang Sun, Hoi-To Wai, and Jun Zhang Tackling Arbitrarily Heterogeneous Data in Asynchronous SGD without Worker Scheduling

This paper investigates decentralized online economic dispatch for smart grids in the presence of Byzantine attacks. In smart grids, some generation stations are subject to external manipulations or internal damages, such that they behave maliciously and send wrong messages to neighboring generation stations, thereby disrupting the online economic dispatch optimization process. We utilize the classical Byzantine attack model to characterize these malicious behaviors and propose a class of Byzantine-resilient decentralized online economic dispatch algorithms. The proposed algorithms employ a variety of existing robust aggregation rules to effectively filter out wrong messages, and are proved to achieve linear dynamic regret and accumulative constraint violation. Numerical experiments are conducted to validate our theoretical results.

This paper considers a finite-sum optimization problem under first-order queries and investigates the benefits of strategic querying on stochastic gradient-based methods compared to uniform querying strategy. We first introduce Oracle Gradient Querying (OGQ), an idealized algorithm that selects one user's gradient yielding the largest possible expected improvement (EI) at each step. However, OGQ assumes oracle access to the gradients of all users to make such a selection, which is impractical in real-world scenarios. To address this limitation, we propose Strategic Gradient Querying (SGQ), a practical algorithm that has better transient-state performance than SGD while making only one query per iteration. For smooth objective functions satisfying the Polyak-Lojasiewicz condition, we show that under the assumption of EI heterogeneity, OGQ enhances transient-state performance and reduces steady-state variance, while SGQ improves transient-state performance over SGD. Our numerical experiments validate our theoretical findings.

This work proposes and studies the distributed resource allocation problem in asynchronous and stochastic settings. We consider a distributed system with multiple workers and a coordinating server with heterogeneous computation and communication times. We explore an approximate stochastic primal-dual approach with the aim of 1) adhering to the resource budget constraints, 2) allowing for the asynchronicity between the workers and the server, and 3) relying on the locally available stochastic gradients. We analyze our Asynchronous stochastic Primal-Dual (Asyn-PD) algorithm and prove its convergence in the second moment to the saddle point solution of the approximate problem at the rate of O(1/t), where t is the iteration number. Furthermore, we verify our algorithm numerically to validate the analytically derived convergence results, and demonstrate the advantages of utilizing our asynchronous algorithm rather than deploying a synchronous algorithm where the server must wait until it gets update from all workers.

Reinforcement learning (RL) aims to estimate the action to take given a (time-varying) state, with the goal of maximizing a cumulative reward function. Predominantly, there are two families of algorithms to solve RL problems: value-based and policy-based methods, with the latter designed to learn a probabilistic parametric policy from states to actions. Most contemporary approaches implement this policy using a neural network (NN). However, NNs usually face issues related to convergence, architectural suitability, hyper-parameter selection, and underutilization of the redundancies of the state-action representations (e.g. locally similar states). This paper postulates multi-linear mappings to efficiently estimate the parameters of the RL policy. More precisely, we leverage the PARAFAC decomposition to design tensor low-rank policies. The key idea involves collecting the policy parameters into a tensor and leveraging tensor-completion techniques to enforce low rank. We establish theoretical guarantees of the proposed methods for various policy classes and validate their efficacy through numerical experiments. Specifically, we demonstrate that tensor low-rank policy models reduce computational and sample complexities in comparison to NN models while achieving similar rewards.

This paper studies the convergence of clipped stochastic gradient descent (SGD) algorithms with decision-dependent data distribution. Our setting is motivated by privacy preserving optimization algorithms that interact with performative data where the prediction models can influence future outcomes. This challenging setting involves the non-smooth clipping operator and non-gradient dynamics due to distribution shifts. We make two contributions in pursuit for a performative stable solution using clipped SGD algorithms. First, we characterize the clipping bias with projected clipped SGD (PCSGD) algorithm which is caused by the clipping operator that prevents PCSGD from reaching a stable solution. When the loss function is strongly convex, we quantify the lower and upper bounds for this clipping bias and demonstrate a bias amplification phenomenon with the sensitivity of data distribution. When the loss function is non-convex, we bound the magnitude of stationarity bias. Second, we propose remedies to mitigate the bias either by utilizing an optimal step size design for PCSGD, or to apply the recent DiceSGD algorithm [Zhang et al., 2024]. Our analysis is also extended to show that the latter algorithm is free from clipping bias in the performative setting. Numerical experiments verify our findings.

This paper proposes a unified approach for designing stochastic optimization algorithms that robustly scale to the federated learning setting. Our work studies a class of Majorize-Minimization (MM) problems, which possesses a linearly parameterized family of majorizing surrogate functions. This framework encompasses (proximal) gradient-based algorithms for (regularized) smooth objectives, the Expectation Maximization algorithm, and many problems seen as variational surrogate MM. We show that our framework motivates a unifying algorithm called Stochastic Approximation Stochastic Surrogate MM (), which includes previous stochastic MM procedures as special instances. We then extend to the federated setting, while taking into consideration common bottlenecks such as data heterogeneity, partial participation, and communication constraints; this yields . The originality of is to learn locally and then aggregate information characterizing the surrogate majorizing function, contrary to classical algorithms which learn and aggregate the original parameter. Finally, to showcase the flexibility of this methodology beyond our theoretical setting, we use it to design an algorithm for computing optimal transport maps in the federated setting.

This paper studies the effect of data homogeneity on multi-agent stochastic optimization. We consider the decentralized stochastic gradient (DSGD) algorithm and perform a refined convergence analysis. Our analysis is explicit on the similarity between Hessian matrices of local objective functions which captures the degree of data homogeneity. We illustrate the impact of our analysis through studying the transient time, defined as the minimum number of iterations required for a distributed algorithm to achieve comparable performance as its centralized counterpart. When the local objective functions have similar Hessian, the transient time of DSGD can be as small as ${\cal O}(n^{2/3}/\rho ^{8/3})$ for smooth (possibly non-convex) objective functions, ${\cal O}(\sqrt{n}/\rho)$ for strongly convex objective functions, where $n$ is the number of agents and $\rho$ is the spectral gap of graph. These findings provide a theoretical justification for the empirical success of DSGD. Our analysis relies on a novel observation with higher-order Taylor approximation for gradient maps that can be of independent interest. Numerical simulations validate our findings.

This paper proposes a two-timescale compressed primal-dual (TiCoPD) algorithm for decentralized optimization with improved communication efficiency over prior works on primal-dual decentralized optimization. The algorithm is built upon the primal-dual optimization framework and utilizes a majorization-minimization procedure. The latter naturally suggests the agents to share a compressed difference term during the iteration. Furthermore, the TiCoPD algorithm incorporates a fast timescale mirror sequence for agent consensus on nonlinearly compressed terms, together with a slow timescale primal-dual recursion for optimizing the objective function. We show that the TiCoPD algorithm converges with a constant step size. It also finds an O(1 /T ) stationary solution after T iterations. Numerical experiments on decentralized training of a neural network validate the efficacy of TiCoPD algorithm.

Supervised fine-tuning is a standard method for adapting pre-trained large language models (LLMs) to downstream tasks. Quantization has been recently studied as a post-training technique for efficient LLM deployment. To obtain quantized fine-tuned LLMs, conventional pipelines would first fine-tune the pre-trained models, followed by post-training quantization. This often yields suboptimal performance as it fails to leverage the synergy between fine-tuning and quantization. To effectively realize low-bit quantization of weights, activations and KV caches in LLMs, we propose an algorithm named Rotated Straight-Through-Estimator (RoSTE), which combines quantization-aware supervised fine-tuning (QA-SFT) with an adaptive rotation strategy that identifies an effective rotation configuration to reduce activation outliers. We provide theoretical insights on RoSTE by analyzing its prediction error when applied to an overparameterized least square quantized training problem. Our findings reveal that the prediction error is directly proportional to the quantization error of the converged weights, which can be effectively managed through an optimized rotation configuration. Experiments on Pythia, Qwen and Llama models of different sizes demonstrate the effectiveness of RoSTE. Compared to existing post-SFT quantization baselines, our method consistently achieves superior performances across various tasks and different LLM architectures. Our code is available at https://github.com/OptimAI-Lab/RoSTE.