PyTorchKR Papers

최신 논문

총 278편
태그·기간 등 상세 필터

카테고리 탐색

arXiv:2606.04757v2math.OC

Near-Optimal Decentralized Stochastic Convex Optimization over Networks

We study decentralized stochastic smooth convex optimization, where $M$ workers minimize an average objective using local stochastic gradients and neighbor-only communication over a fixed gossip network. A central question in this setting is to determine the largest number of workers that can be used under a total budget of $N$ gradient samples while still preserving the centralized $O(1/\sqrt N)$ statistical rate. We introduce an accelerated decentralized method that preserves this rate for up to $\smash{M\lesssim \sqrt{\rho}\,N^{3/4}}$ workers, where $\rho$ is the spectral gap of the gossip network, improving the best prior maximal scaling of $\smash{M\lesssim \rho\sqrt N}$. The method is based on a one-step-delayed stochastic acceleration scheme that enables workers to interleave minibatching with accelerated gossip while controlling residual disagreement, and its guarantee depends only logarithmically on the optimum-local heterogeneity. We also establish a matching lower bound for linear-span decentralized first-order methods, showing that the method is optimal up to logarithmic factors.

arXiv:2605.08866v2stat.ML

Tight Generalization Bounds for Noiseless Inverse Optimization

Inverse optimization (IO) seeks to infer the parameters of a decision-maker's objective from observed context--action data. We study noiseless IO, where demonstrations are generated by a ground-truth objective. We provide a high-probability ${O}(\frac{d}{T})$ generalization bound for the induced action set, where $d$ is the number of unknown parameters and $T$ is the size of the training dataset. We strengthen this guarantee under additional conditions that ensure uniqueness of the chosen action, bringing our IO guarantees in line with best-arm identification results in the bandit literature. We further show that the ${O}(\frac{d}{T})$ rate is tight over all consistent estimators considered here, and extend the result to both instantaneous and cumulative regret. Notably, the resulting regret lower bound matches the corresponding upper bounds in the adversarial setting, indicating that the stochastic IO setting is effectively adversarial for the class of estimators studied here. Finally, we propose a parameter-free algorithm with lower per-iteration complexity than generic solvers. Experiments validate the predicted rates and illustrate the tightness of our bounds.

arXiv:2605.07104v3cs.LG

Almost Sure Convergence Rates of Stochastic Approximation and Reinforcement Learning via a Poisson-Moreau Drift

Establishing almost sure convergence rates for stochastic approximation and reinforcement learning under Markovian noise is a fundamental theoretical challenge. We make progress towards this challenge for a class of stochastic approximation algorithms whose expected updates are contractive, a setting that arises in many reinforcement learning algorithms such as $Q$-learning and linear temporal difference learning. Specifically, for a power-law learning rate $O(n^{-\eta})$ with $\eta \in (1/2, 1)$, we obtain an almost sure convergence rate arbitrarily close to $o(n^{1 - 2\eta})$. For the harmonic learning rate $\alpha_n=\alpha/(n+1)$, the almost sure squared-error rate is $o(n^{-\zeta})$ for every $\zeta<\min\{1,2(1-\kappa)\alpha\}$, where $\kappa$ is the contraction factor. Thus exponents arbitrarily close to one are available when $2(1-\kappa)\alpha\geq1$. This approaches the $n^{-1}$ polynomial exponent of classical law-of-the-iterated-logarithm benchmarks. Key to our analysis is a novel Lyapunov drift construction that applies a Poisson-equation based correction for Markovian noise to the well-established Moreau-envelope smoothing for the contractive mapping.

arXiv:2605.06152v5cs.LG

Grokking or Glitching? How Low-Precision Drives Slingshot Loss Spikes

Deep neural networks exhibit periodic loss spikes during unregularized long-term training, a phenomenon known as the "Slingshot Mechanism." Existing work usually attributes this to intrinsic optimization dynamics, but its triggering mechanism remains unclear. This paper proves that this phenomenon is a result of floating-point arithmetic precision limits. As training enters a high-confidence stage, the difference between the correct-class logit and the other logits may exceed the absorption-error threshold. Then during backpropagation, the gradient of the correct class is rounded exactly to zero, while the gradients of the incorrect classes remain nonzero. This breaks the zero-sum constraint of gradients across classes and introduces a systematic drift in the parameter update of the classifier layer. We prove that this drift forms a positive feedback loop with the feature, causing the global classifier mean and the global feature mean to grow exponentially. We call this mechanism Numerical Feature Inflation (NFI). This mechanism explains the rapid norm growth before a Slingshot spike, the subsequent reappearance of gradients, and the resulting loss spike. We further show that NFI is not equivalent to an observed loss spike: in more practical tasks, partial absorption may not produce visible spikes, but it can still break the zero-sum constraint and drive rapid growth of parameter norms. Our results reinterpret Slingshot as a numerical dynamic of finite-precision training, and provide a testable explanation for abnormal parameter growth and logit divergence in late-stage training.

arXiv:2604.17213v2math.OC

Symplectic Inductive Bias for Data-Driven Target Reachability in Hamiltonian Systems

Inductive bias refers to restrictions on the hypothesis class that enable a learning method to generalize effectively from limited data. A canonical example in control is linearity, which underpins low sample-complexity guarantees for stabilization and optimal control. For general nonlinear dynamics, by contrast, guarantees often rely on smoothness assumptions (e.g., Lipschitz continuity) which, when combined with covering arguments, can lead to data requirements that grow exponentially with the ambient dimension. In this paper we argue that data-efficient nonlinear control demands exploiting inductive bias embedded in nature itself, namely, structure imposed by physical laws. Focusing on Hamiltonian systems, we leverage symplectic geometry and intrinsic recurrence on energy level sets to solve target reachability problems. Our approach combines the recurrence property with a recently proposed class of policies, called chain policies, which composes locally certified trajectory segments extracted from demonstrations to achieve target reachability. We provide sufficient conditions for reachability under this construction and show that the resulting data requirements depend on explicit geometric and recurrence properties of the Hamiltonian rather than the state dimension.

arXiv:2511.03087v3stat.ME

Beyond Maximum Likelihood: Variational Inequality Estimation for Generalized Linear Models

Generalized linear models (GLMs) are fundamental tools for statistical modeling, with maximum likelihood estimation (MLE) serving as the classical approach for parameter inference. While MLE performs well for canonical GLMs, it can become computationally challenging in more general settings with non-canonical, non-smooth, or saturating link functions, where the resulting optimization landscape may be ill-conditioned, non-convex, or non-differentiable. In this paper, we study an alternative estimation framework based on variational inequalities (VIs), which formulates GLM estimation through an operator-based equilibrium condition rather than likelihood minimization. We analyze the VI estimator from a statistical perspective and establish finite-sample error bounds and asymptotic normality, together with convergence guarantees for fixed-point and stochastic approximation algorithms. The framework accommodates a broad class of link functions, including non-canonical, non-smooth, and non-monotone cases satisfying a strong Minty-type condition, and extends naturally to generalized additive models (GAMs) via basis expansion. Numerical experiments demonstrate that under a fixed iteration budget, the VI approach achieves competitive estimation accuracy and faster, more stable convergence than MLE, particularly in GLMs and GAMs with non-canonical or non-smooth link functions.

arXiv:2509.14039v2stat.ML

On the Rate of Gaussian Approximation for Online Linear Regression Problems

In this paper, we consider the problem of Gaussian approximation for the online linear regression task. We derive the corresponding rates for the setting of a constant stepsize and study the explicit dependence of the convergence rate on the problem dimension $d$ and quantities related to the design matrix. When the number of iterations $n$ is known in advance, our results yield a normal approximation rate of order $\sqrt{\log{n}/n}$, provided that the sample size $n$ is large enough.

arXiv:2506.03757v3cs.LG

PPO in the Fisher-Rao geometry

Proximal Policy Optimization (PPO) is widely used in reinforcement learning due to its strong empirical performance, yet it lacks formal guarantees for policy improvement and convergence. PPO's clipped surrogate objective is motivated by a lower bound on linearization of the value function in flat geometry setting. We derive a tighter surrogate objective and introduce Fisher-Rao PPO (FR-PPO) by leveraging the Fisher-Rao (FR) geometry. Our scheme provides strong theoretical guarantees, including monotonic policy improvement. In the direct parametrization setting, we show that FR-PPO achieves sub-linear convergence, and for parametrized policies we further obtain sub-linear convergence up to the compatible function approximation error. Finally, although our primary focus is theoretical, we also demonstrate empirically that FR-PPO can be implemented as a drop-in replacement for PPO and achieves performance comparable to PPO across a range of standard reinforcement learning tasks.

arXiv:2407.04259v4math.OC공식 코드 · Papers with Code

Q-Learning under Finite Model Uncertainty

We propose a robust Q-learning algorithm for Markov decision processes under model uncertainty when each state--action pair is associated with a finite ambiguity set of candidate transition kernels. This finite-measure framework enables user-designed uncertainty models, including competing fitted models and stress scenarios. {The learner observes successor states and their deterministic rewards from every candidate simulator, without requiring explicit knowledge of the transition probabilities or the reward function.} We establish almost sure convergence of the learned Q-function to the robust optimum, and derive a time-uniform high-probability error certificate that separates an observable empirical Bellman residual from transition-kernel estimation error. We also give quantitative finite-ambiguity approximation results for Wasserstein and parametric uncertainty models on finite state spaces. Numerical examples illustrate the effect of scenario selection and compare the tabular algorithm with robust value iteration.

arXiv:2402.08106v4math.OC

Mirror Descent-Ascent for mean-field min-max problems

We study two variants of the mirror descent-ascent (MDA) algorithm for solving min-max problems on the space of measures: simultaneous and alternating. We work under assumptions of convexity-concavity and relative smoothness of the payoff function with respect to a suitable Bregman divergence, defined on the space of measures via flat derivatives. We establish non-asymptotic convergence rates to mixed Nash equilibria, measured in the Nikaid\^{o}-Isoda error, proving an $\mathcal{O}(N^{-1/2})$ rate for simultaneous MDA and an improved $\mathcal{O}(N^{-2/3})$ rate for alternating MDA. The main technical contribution is an infinite-dimensional dual space analysis that relates Bregman divergences on measures to dual Bregman divergences on spaces of bounded continuous functions, allowing us to control asymmetric commutator terms created by alternating updates. The results substantially generalize prior analyses restricted to bilinear objectives and also apply to nonlinear convex-concave problems on measure spaces, thereby providing a unified theoretical foundation for MDA in mean-field min-max optimization.

arXiv:2611.06879v1cs.LG

Adaptive Multi Mode: Budget-Aware Routing of Optimizers By Online Landscape Probes

The best classical optimizer for a continuous black-box problem depends on its budget and landscape. Existing routers decide from the dimension and budget, past improvement or features fitted to one benchmark, and few test an optimizer's implicit assumptions online. AMM (Adaptive Multi Mode) routes a known budget among quasi-Newton, trust-region, coordinate, multistart and evolution-strategy optimization engines with online probes of curvature, plateaus, separability, kinks, basins and fractal structure. Each probe has a default, a dimension-by-budget validity range and a cost charged to the budget. In its pre-registered development loop on bbob, 32 of 36 predictions held over the last 16 confirmations. AMM was then tested once on held-out bbob instances 101-110 (D=2-40, 10*D-1000*D) against up to seven competitors. It has the highest target share in 382 of 600 function-dimension-budget cells (236 outright, 146 tied) and, by a descriptive posterior rule, is credibly behind in 35, mostly at 10*D and 30*D. At every budget it also leads the controls built from its own engines (the best engine alone, the best engine per dimension and budget, and a greedy schedule), narrowly at 10*D and 30*D, and nevergrad's NgIohTuned. In pre-registered runs its lead over IPOP-CMA-ES holds under affine and box transformations, on box-constrained bbob and, pooled, on rotated f2-f4, but rotating the separable f3 and f4 removes most of the lead there. On held-out MA-BBOB its final share at 2000*D is level with IPOP-CMA-ES's and its anytime area lower. We published AMM as an open-source Python package (pip install amm-optimizer; https://github.com/BjBodner/amm-optimizer).

arXiv:2611.06021v1cs.LG

Attraction-Ranking Mismatch in Out-of-Distribution Detection: Theory and Evidence from Hamiltonian Feature Dynamics

Attracting features toward learned in-distribution (ID) regions is an intuitive strategy for out-of-distribution (OOD) detection, but does such attraction improve detection? We investigate this question using class-conditioned Hamiltonian dynamics as a controlled testbed and identify an attraction-ranking mismatch: improving individual class-affinity scores need not improve the relative ordering of ID and OOD samples. Our analysis separates misalignment between the driving field and the detection score from ranking failure that persists even when they are aligned. An exact AUROC decomposition and a local boundary analysis identify relative score changes at the ID-OOD ranking boundary as the source of first-order AUROC change under regularity conditions. Experiments demonstrate that a score-aligned rule can improve performance on a held-out source validation split yet consistently degrade it after transfer, even without a single winning-class change. A separate perturbation study finds that near-boundary relative score changes track the direction of small AUROC changes in the tested settings. Finally, centroid approximation and shared scales connect the zero-step score to a static centroid family that includes the existing CTM detector. Together, these results explain why class attraction is insufficient to justify feature refinement and motivate evaluating its effect on the ranking boundary.

arXiv:2611.05869v1cs.LG

On Convergence of Gradient Temporal-Difference Learning with Regularized Corrections and Eligibility Traces

Temporal-difference (TD) learning underlies many prediction and control algorithms, but can diverge when combined with off-policy learning, bootstrapping, and function approximation. TD with Regularized Corrections (TDRC) is theoretically sound and practically simple, but existing convergence theory covers only the one-step case under i.i.d. sampling, while the Huber Bellman-error formulation has not been extended to TDRC or eligibility traces. In this work, we establish convergence of constrained TDRC($\lambda$) with diminishing stepsizes and non-i.i.d. off-policy data, for state-dependent and history-dependent $\lambda$-parameters. We also propose TDRC($\lambda$)-Huber, which preserves the TDRC($\lambda$) fixed point while reducing sensitivity to large Bellman errors. For both algorithms, we apply an ODE-based weak-convergence method from stochastic approximation theory. For the nonlinear mean dynamics of TDRC($\lambda$)-Huber, we give an integral small-gain condition that guarantees uniqueness and global asymptotic stability of the target equilibrium. We then establish convergence of the constrained algorithm, derive computable sufficient conditions for convergence, and identify a regularization-based convergence mechanism shared by linear TDRC($\lambda$) and TDRC($\lambda$)-Huber. When the mean TD matrix is positive stable, both constrained algorithms converge under sufficiently strong regularization without requiring positive definiteness.

arXiv:2611.05835v1cs.LG

Pathwise-Budgeted LLM Routing: Online Allocation with K-Bit Tail-Cost Learning

Firms using large language models under fixed budgets must decide which model to call and whether to retry a failed task. Because a call's charge is revealed only after completion, one long response can exhaust the remaining budget and leave later tasks unserved. Our objective is to maximize the number of tasks solved before service stops. We develop a contextual bandit policy, ADP-HT, which learns each model's task-specific success probability and constructs conservative cost estimates from coarsened response-length labels on a quantile grid with $2^K$ intervals. The policy weighs predicted success against these costs, adjusting the cost penalty to realized spending and the budget available per remaining task. Under calibration, local regularity, and grid-coverage conditions, we bound the excess allowance in the policy's conservative cost estimates. At a fixed sample size, increasing $K$ reduces the discretization term but increases the estimation allowance. At the tightest tested budgets, ADP-HT solves more than twice as many MATH500 and GPQA tasks as the best implemented baseline. It achieves the highest mean cumulative score at ten of fourteen RouterBench budgets. On LiveCodeBench, it outperforms a deployed commercial router in both default and cost-saving modes at the three lowest test budgets.

arXiv:2611.05594v1cs.LG

Relevance Propagation as Entropic Optimal Transport on Filippov Neural Networks: Sinkhorn Propagation and Region Relevance in Image Classifiers

Relevance propagation, which redistributes a network's prediction backward through the layers, is usually treated as a heuristic. We show that it is the solution of a single variational problem: Layer-wise Relevance Propagation (LRP) is the divergence-free flow on the active computational graph of zero relative entropy to the backward transition law of the forward activation flow; for a non-negative reference, this is a relative entropy between Markov measures on backward paths. The rules LRP$_0$, LRP$_{z^+}$ and LRP$_{\alpha,\beta}$ follow from this principle by the choice of a reference, and are instances of one rule, LRPN$_{\alpha,\beta}$, whose weights may vary from node to node, for instance with the node's activity, in the spirit of Hebb's principle. The principle rests on a measure-theoretic formalization of feed-forward networks as directed acyclic graphs of Filippov operators, which covers piecewise-smooth and discontinuous architectures (e.g. hard-routed Mixture-of-Experts) with a Clarke-Filippov generalized-gradient calculus; it also characterizes backpropagation as the adjoint equation of a Lagrangian relaxation. On networks of ReLU and max-pooling nodes, LRP$_0$ is Gradient$\times$Input, with an explicit jump across a kink where LRP$_{z^+}$ stays continuous. Taking the entropy edge by edge yields a new rule, Sinkhorn Relevance Propagation. Aggregated over an input region, the flow of a non-negative reference tells, with exact accounting, which hidden units carry its evidence, the conditioned flow being a Schr\"odinger bridge. On VGG-16, the conservation identities hold to round-off, and the rules are told apart by the sign of their maps, the share of the output left to the bias nodes, faithfulness, localization and class dependence; on ConvNeXt-Tiny, LRPN$_{\alpha,\beta}$ conserves the output and its Hebbian weights give the highest ABPC of the rules tested.

arXiv:2611.05527v1math.OC

Tight Lower Bounds under Matched High-Order H\"older Smoothness and Uniform Convexity

In this paper, we provide tight lower bounds for minimizing high-order H\"older-smooth and uniformly convex functions when their orders match. Specifically, we consider functions whose $p^{th}$-order derivatives are H\"older continuous with degree $\nu$ and parameter $H$, and that are uniformly convex with degree $q$ and parameter $\sigma$. When these orders match, namely, $q = p + \nu$, we show that any deterministic algorithm with access to derivatives up to order $p$ requires $\Omega\left(\left(\frac{H}{\sigma}\right)^{\frac{2}{3(p+\nu)-2}}\log\frac{\Delta_0}{\epsilon}\right)$ oracle calls to reduce an initial optimality gap of $\Delta_0$ to $\epsilon$. This matches the corresponding upper bound and establishes a higher-order uniformly convex counterpart to Nesterov's classical first-order lower bound for strongly convex functions. Rather than directly generalizing Nesterov's classical quadratic-chain construction, we develop a multiscale phase-wise construction that preserves previous oracle responses while maintaining global H\"older smoothness and uniform convexity. Our result resolves the matched case left open by existing lower bounds for the two asymmetric regimes.

arXiv:2610.29311v2cs.LG

Interpretable Cutting Plane Selection for Mixed-Integer Linear Programming Using Genetic Programming

Cutting planes (cuts) remove fractional solutions from mixed-integer linear programs (MILPs) without excluding any feasible integer solutions, making them a key ingredient in modern MILP solvers. Current state-of-the-art solvers rely on fine-tuned handcrafted cut strategies, which, despite their effectiveness, remain inherently static and can be suboptimal. More recently, learning-based approaches have been proposed to design adaptive cut strategies; however, most of these methods are based on neural networks, offering limited interpretability and providing no guarantees of convergence. In this paper, we propose \emph{Genetic Programming and Statistics-based Cut Selection} (GPSCS), the first learning-based framework that explicitly generalizes handcrafted cut selection heuristics into a fully interpretable policy. First, a genetic programming model evolves a symbolic scoring function that ranks candidate cuts using features derived from the cuts and the solving context. Second, a lightweight statistics-based mechanism adaptively determines the number of top-ranked cuts to add at each separation round by detecting significant score dropouts. We evaluate GPSCS within the SCIP solver on seven classes of synthetic and real-world benchmark instances, with a strong emphasis on generalization: evaluation instances are on average 41 times harder than those used for training. Experimental results show that GPSCS improves average solving time by $6\%$ compared to SCIP's default cut selection strategy and achieves a two-fold speedup over both a recent neural-network-based approach and a no-cut baseline, demonstrating strong robustness and generalization across heterogeneous MILP families.

arXiv:2609.09502v2stat.ML

Recovery Theory for Projected Power Iterations in Permutation Synchronization

We study the projected power method for permutation synchronization: recovering $n$ unknown permutations of $m$ objects from pairwise relative permutations, observed independently with probability $p$ and independently corrupted by uniform permutations with probability $1-\pi_0$. Our main theorem establishes uniform one-step contraction for all estimates with optimally aligned block-error fraction at most $1/2-\epsilon$, for fixed $\epsilon>0$, including measurement-dependent estimates. When $np\ge C_0\log(n)$ and $m=o(np\pi_0^2)$, one update contracts the error by a factor $O(m/(np\pi_0^2))$, up to a floor of $O(e^{-c\min\{m\,np\pi_0^2,\,np\pi_0\}} + \log(n) / n)$. Thus, every initialization in this majority basin achieves $o(1)$ error after one update and maintains it uniformly over subsequent iterations. A reference-block spectral initializer enters the basin with probability $1-o(1)$, yielding end-to-end almost-exact recovery. If additionally $\min\{m\,np\pi_0^2,\,np\pi_0\} \ge C\log(n)$ for sufficiently large $C$, the method achieves exact recovery in finitely many iterations. We also establish prescribed-block recovery under weaker conditions for measurement-independent estimates and extend the guarantees to independent permutation-valued corruptions with uniform mean and to partial permutations with common visible support.

arXiv:2604.03566v2math.OC

Fr\'echet Regression on the Bures-Wasserstein Manifold

Fr\'echet regression, or conditional Barycenters, is a flexible framework for modeling relationships between covariates (usually Euclidean) and response variables on general metric spaces, e.g., probability distributions or positive definite matrices. However, in contrast to classical barycenter problems, computing conditional counterparts in many non-Euclidean spaces remains an open challenge, as they yield non-convex optimization problems with an affine structure. In this work, we study the existence and computation of conditional barycenters, specifically in the space of positive-definite matrices with the Bures-Wasserstein metric. We provide a sufficient condition for the existence of a minimizer of the conditional barycenter problem that characterizes the regression range of extrapolation. Moreover, we further characterize the optimization landscape, proving that under this condition, the objective is free of local maxima. Additionally, we develop a projection-free and provably correct algorithm for the approximate computation of first-order stationary points. Finally, we provide a stochastic reformulation that enables the use of off-the-shelf stochastic Riemannian optimization methods for large-scale setups. Numerical experiments validate the performance of the proposed methods on regression problems of real-world biological networks and on large-scale synthetic Diffusion Tensor Imaging problems.

arXiv:2512.20270v2math.OC

Optimality-Informed Neural Networks for Solving Parametric Optimization Problems

Many engineering tasks require solving families of nonlinear constrained optimization problems, parametrized in setting-specific variables. This is computationally demanding, particularly if solutions have to be computed across strongly varying parameter values, e.g., in real-time control or for model-based design. Thus, we propose to learn the mapping from parameters to the primal optimal solutions and to their corresponding duals using neural networks, giving a dense estimation in contrast to gridded approaches. Our approach, Optimality-Informed Neural Networks (OptINNs), combines (i) a KKT-residual loss that penalizes violations of the first-order optimality conditions under standard constraint qualification assumptions, and (ii) problem-specific output activations that enforce simple inequality constraints (e.g., box-type/positivity) by construction. This design reduces data requirements, allows the prediction of dual variables, and improves feasibility and closeness to optimality compared to penalty-only training. Taking quadratic penalties as a baseline, as previously proposed for this problem class in the literature, our method simplifies hyperparameter tuning and attains tighter adherence to optimality conditions. We evaluate OptINNs on different nonlinear optimization problems ranging from low to high dimensions and compare them with equally sized neural networks trained using the quadratic-penalty method. On small problems, OptINNs match a quadratic-penalty baseline in primal accuracy while additionally predicting dual variables with low error. On larger problems, OptINNs achieve lower constraint violations, lower primal error, and lower errors in the predicted costs. These results suggest that embedding feasibility and optimality into the network architecture and loss can make learning-based surrogates more accurate, feasible, and data-efficient for parametric optimization.