数学前沿

Gradient clipping is a key tool for obtaining high-probability convergence guarantees for stochastic first-order methods under heavy-tailed noise. Existing analyses typically assume a bounded central moment of order $\alpha\in(1,2]$, which excludes distributions with tails of order $u^{-\alpha}$ at the critical exponent. In this paper, we consider smooth convex optimization under affine heavy-tailed noise, whose scale may grow with the gradient norm. We derive high-probability convergence guarantees for clipped SGD (Clip-SGD) and its accelerated variant Clip-SSTM under a weak-$L^\alpha$ tail condition, which is strictly weaker than the corresponding moment bound. For $1<\alpha<2$, the clipping bounds have the same powers of the clipping level as in the moment case, and both methods attain the additive stochastic rate $K^{-(\alpha-1)/\alpha}$. The gradient-dependent noise contributes a faster-decaying term of order $K^{-2(\alpha-1)/\alpha}$, up to logarithmic factors. At $\alpha=2$, we characterize the additional logarithmic factor on the Lorentz scale and establish its necessity for the additive error at fixed confidence. We also analyze Clip-SGD with prescribed clipping levels and derive finite-horizon guarantees for reaching a neighborhood determined by the clipping bias. For additive noise, matching lower bounds show that this neighborhood is unavoidable, up to constants and the natural error cap. Numerical experiments with Pareto noise illustrate the theoretical results.

Federated learning (FL) is an emerging distributed machine learning paradigm that enables local devices to jointly train a global model while keeping data decentralized and private. We propose a variance-reduction based algorithm, VRA-FedSGD, for FL in the presence of heavy-tailed gradient noise and communication noise, where these noises are prevalent in large-scale machine learning over wireless networks and Internet of Things deployments. VRA-FedSGD employs a momentum variance reduction technique together with a nonlinear mapping to mitigate heavy-tailed gradient noise, and uses a variance-reduced aggregation mechanism to suppress heavy-tailed communication noise. In the mean sense, VRA-FedSGD achieves a convergence rate of {\small$\mathcal{O}\left(K^{-(p-1)/(2p-1)}\right)$} for nonconvex objective functions, where $p$ is the tail index of heavy-tailed noise. In the almost sure sense, VRA-FedSGD achieves a convergence rate of $\tilde{\mathcal{O}}\left(K^{-(1-1/(p-\epsilon))}\right)$ for strongly convex objective functions, where $\epsilon$ is an arbitrarily small constant. Simulated experiments on a logistic regression problem with real-world data verify the effectiveness of VRA-FedSGD.

Modern distributed optimization methods mostly rely on traditional synchronous approaches, despite substantial recent progress in asynchronous optimization. We revisit Synchronous SGD and its robust variant, called $m$-Synchronous SGD, which can be interpreted as a method with a backup-worker strategy, and theoretically show that they are nearly optimal in many heterogeneous computation and communication scenarios, which is somewhat unexpected. We analyze the synchronous methods under random times and adversarial partial participation of workers, and prove that their time complexities are optimal in many practical regimes, up to logarithmic factors. While synchronous methods are not universal solutions and there exist tasks where asynchronous methods may be necessary, we show theoretically that they are sufficient for many modern heterogeneous computation and communication scenarios.

Federated learning (FL) is increasingly used in safety-critical and fairness-sensitive applications, where learning objectives must be optimized subject to functional constraints. This setting is challenging due to costly projection onto the resulting feasible set, stochastic constraint information, partial client participation, and limited communication at edge devices. We propose a federated stochastic switching framework based on primal-only updates, avoiding dual-variable tuning for functionally constrained federated optimization. For convex objectives, the framework further supports compression with error feedback to control compression-induced bias. We establish high-probability convergence and feasibility guarantees under stochastic gradients and partial client participation, and extend the analysis to weakly convex objectives through proximal-stationarity guarantees. Our bounds isolate the effects of stochastic gradients, constraint-evaluation noise, and partial participation, and show that direct switching can retain a nonvanishing constraint-estimation error even with exact local constraint evaluations. To overcome this limitation, we develop variance-reducing constraint estimators that yield vanishing feasibility and stationarity guarantees with fixed per-round client participation and constraint-evaluation batch sizes. Experiments on Neyman-Pearson classification, fairness-constrained learning on the ADULT dataset, and constrained deep reinforcement learning demonstrate the effectiveness of the proposed framework.

Gradient normalization can stabilize deep-learning optimization, and spectral normalizations are especially natural for matrix-shaped parameter blocks; Muon is the motivating example. We study an idealized deterministic, continuous-time, vanishing-momentum version of this idea in the mean-field regime, where wide models are represented by probability measures on parameter space. Starting from normalized matrix flows, we introduce Spectral Wasserstein distances indexed by norms $\gamma$ on positive semidefinite matrices: the trace norm recovers the classical $W_2$ distance, the operator norm gives the Muon geometry, and Schatten norms interpolate between them. We develop a static Kantorovich formulation, a max--min robust-cost representation, and Gaussian reductions extending the Bures formula. For monotone norms, we prove equivalence with a Benamou--Brenier formulation and obtain a gradient-flow interpretation of normalized mean-field training dynamics. Numerical experiments illustrate the resulting flows for MMD objectives, Gaussian reductions, two-layer ReLU models, and shallow attention.

Despite having an asymptotically optimal rate of convergence on convex quadratics, the acceleration of Polyak's heavy-ball method and a characterization of its dynamics within more general settings have remained long standing open problems. Recently, the work of Goujaud et al. [Math. Program. (2025)] made progress regarding acceleration in the context of smooth, strongly convex minimization: for a small enough inverse condition number, either the method does not accelerate on quadratics, or there exists a smooth strongly convex function such that heavy-ball cycles over the roots of unity. The constructed functions have two variables, thus leaving the question of acceleration over univariate functions open. In this work, we use a Fourier-based analysis to i) provide a proof for the non-acceleration of the heavy-ball method over the class of univariate smooth strongly convex functions. We also ii) provide a strict enlargement of the previous root-of-unity cycling region, with insights regarding the "fractional" sets of Goujaud et al. [arXiv:2502.19916], and iii) give a higher dimensional cycling instance which is not covered by root-of-unity cycling.

Bayesian optimization (BO) is widely used for expensive black-box optimization, yet observation noise makes each evaluation less informative and slows posterior learning. Standard BO learns from costly outcomes gathered during optimization. In many applications, however, low-cost auxiliary data are available beforehand. We introduce Auxiliary-Informed Gaussian Process Bandit (AI-GPB), a variance-reduction method that combines an offline estimate of the auxiliary mean with a correction coefficient learned from paired online observations. Notably, it exploits correlation between paired observation errors without requiring the auxiliary mean to approximate the objective. We establish finite-sample GP confidence bounds and regret guarantees, including explicit cumulative and simple regret rates for Mat\'ern kernels. Numerical experiments across continuous BO benchmarks with varied landscapes and dimensions show consistent reductions in cumulative and simple regret relative to vanilla GP-UCB, even when within-pair correlation is moderate and the auxiliary mean is estimated from noisy offline data.

Bilevel optimization provides a framework for machine learning tasks such as hyperparameter optimization, data reweighting, and meta-learning. Existing Newton-based bilevel methods typically rely on the classical hypergradient formula, which is generally unavailable when lower-level solutions are nonunique and the lower-level Hessian is singular. We develop two Newton-based methods beyond lower-level strong convexity by exploiting curvature shared by the lower-level and multiplier updates. Newton-based GALET handles potentially singular lower-level curvature under a positive spectral-gap condition, exploiting its range structure to avoid the squared-Hessian multiplier system used in GALET. Newton-based BAMM uses a strongly convex aggregation model when the lower-level objective is convex and the upper-level objective is strongly convex in the lower-level variable. We establish convergence guarantees and improved complexity bounds relative to GALET and sl-BAMM under the corresponding assumptions. We further introduce GNBO, a generalized Newton-based framework that captures the shared-curvature structure of the proposed methods and several existing Newton and quasi-Newton bilevel methods. Experiments on synthetic problems, data hyper-cleaning, and GPT-2 data reweighting evaluate the proposed methods across different bilevel learning tasks.

In large language model (LLM) serving, each request accumulates persistent graphics processing unit (GPU) memory during service as its key-value cache grows with every generated token. Under high concurrency, aggregate memory usage therefore increases endogenously over time: the service process itself creates future capacity pressure. When memory capacity is exceeded, systems evict active requests, discarding cached state and restarting them later, which wastes computation and reduces throughput. We develop a discrete-time dynamical model of memory-constrained LLM inference that captures admission, memory growth, and eviction under continuous batching. In the saturated-input regime, the system admits both eviction-free fixed points and limit cycles with evictions. For homogeneous workloads, we show that the eviction-free equilibrium is unstable and that, except for a Lebesgue-measure-zero exact-capture set, the system converges to a unique worst-case limit cycle that is asymptotically stable outside this exceptional set, with throughput losses as large as 50%. For heterogeneous workloads, we prove a stability criterion in the two-class common-input setting and explain how the survival-polynomial mechanism generalizes to multiple classes and heterogeneous-input lengths. Under an input-dominated scaling regime, coprime decoding lengths stabilize the eviction-free equilibrium, while non-coprime lengths create synchronized modes that drive instability. These results characterize when workload heterogeneity desynchronizes completions and helps stabilize memory-constrained serving. More broadly, we identify service-induced congestion as a structural instability mechanism and derive scheduling design principles for sustaining high throughput.

Empirical averages are fundamental statistical primitives. On nonlinear spaces, their generalization, the Fr\'echet mean, generally lacks a closed-form solution, and its computation requires solving an optimization problem. We study the computation of intrinsic Fr\'echet means for data supported on an unknown compact $d$-dimensional Riemannian manifold $M_{1}$ embedded in an ambient metric space $(X,d_{\operatorname{obs}})$. Such settings arise when observations are better represented in non-Euclidean spaces, like probability distributions in Wasserstein space. We assume access only to a local sampler on $M_{1}$ and the observation metric $d_{\operatorname{obs}}$, without knowledge of its intrinsic geometry. We propose a graph-based iterative algorithm that adaptively samples the manifold and approximates geodesic interpolation. Under mild assumptions, we establish an $O(1/T)$ expected objective gap guarantee using at most $\widetilde{O}(T^{4d+3})$ sampled points. As a byproduct of our analysis we extend the metric equivalence guarantees of ISOMAP from Euclidean to Riemannian ambient spaces, which may be of independent interest. Numerical experiments demonstrate the effectiveness of our method for computing intrinsic averages on unknown data manifolds.

Next Best Action (NBA) systems must select among customer interventions (e.g., discounts, promotions, and lifecycle actions), maximize long-term economic value, satisfy complex operational constraints, and remain causally grounded. Existing approaches lack a rigorous framework for meeting these requirements jointly. We address this gap by formulating NBA as a Constrained Markov Decision Process (CMDP). We then integrate off-policy causal correction of the estimated customer dynamics directly into the Linear Programming (LP) formulation of the CMDP. The resulting policy selects actions based on customer states to maximize long-term value while explicitly enforcing business guardrails. Furthermore, the LP formulation yields shadow prices that quantify the economic trade-offs of operational constraints (e.g., fairness and user welfare), enhancing policy interpretability. We validate our proposed offline Reinforcement Learning (RL) framework on an industry case study involving rideshare coupon targeting, where customer states are defined by Recency, Frequency, and Monetary (RFM) stages. We evaluate performance using model-based off-policy evaluation, with causal dynamics estimated from a temporal holdout. Under the observed coupon budget and RFM state representation, our optimized policy improves customer lifetime value (LTV) by 19% over the estimated behavior policy and 8% over a myopic baseline. Although the demonstrated action space is binary, the proposed framework is action-agnostic, yielding a general-purpose decision module for targeting systems that manage multiple interventions.

We study the fundamental problem of one-step prediction of a marginally stable unknown nonlinear dynamical system. We describe an algorithm for this problem, based on the technique of spectral filtering, which learns a mapping from past observations to the next based on a spectral representation of the system. Using techniques from online convex optimization, we prove vanishing prediction error for any nonexpansive nonlinear dynamical system with finitely many marginally stable modes, with rates governed by a novel quantitative control-theoretic notion of learnability. The main technical component of our method is a new spectral filtering algorithm for linear dynamical systems, which incorporates past observations and applies to general noisy and marginally stable systems. This generalizes the original spectral filtering algorithm to both asymmetric dynamics as well as incorporating noise correction, and is of independent interest.

The block generalized-power method pairs threshold-based support selection with inexpensive Stiefel updates, but the standard block formulation conflates two nonequivalent objectives and leaves inactive components, threshold ties, and post-selection refitting ill-specified. We rebuild the model from the primal. Eliminating each loading column from the normalized loading-ball problem gives a sum of soft-thresholded Euclidean norms (A-SUM); the squared-threshold objective in common use arises from a quadratic lift (A-SQ). We prove both identities, exhibit a two-component instance whose maximizers disagree, and specify explicit conventions for zero-response columns, ties, rank deficiency, and mask-before-refit. For A-SQ we separate the historical full-polar iteration from a safeguarded Riemannian ascent and prove, for the latter, whole-sequence convergence via a sufficient-ascent lemma, a relative-error lemma, and the Kurdyka-Lojasiewicz property, which applies because the soft-threshold objective and the Stiefel constraint are semi-algebraic. A strict threshold margin then forces finite identification of the support and sign pattern. The post-identification map is ordinary power iteration on the active covariance for a rank-one hard threshold, and the Frechet derivative of a coupled polar map for component-specific supports; a radius-based safe-screening rule follows from the same local analysis. Planted-spike experiments match the predicted identification time and local rates. On modern-AI panels (GPT-2 residual stream, ViT CLS stream, Pitprops) the thresholded block-sparse components match dense PCA on reconstruction and improve on it in monosemanticity and class specialisation. On three native-label scRNA datasets (Zheng17 FACS-sorted PBMCs, 68k donor-A with FACS-correlated labels, Tabula Muris) A-SQ beats dense PCA on cluster-recovery AMI by +0.023 to +0.044 at matched sparsity.