Overview

Generalization bounds estimate how well a model trained on finite data will perform on new examples, helping explain when learning is reliable. Common tools include VC and Rademacher complexity, covering numbers, PAC-Bayes and information-theoretic bounds, and algorithmic stability. For deep networks and modern settings, analyses increasingly use learned feature geometry, optimization dynamics, and structure in data, while extending guarantees to distribution shift, generative models, and dependent or distributed samples. A central tension is that broad worst-case guarantees can be too loose to be useful, whereas tighter, non-vacuous bounds often rely on assumptions about the algorithm, data geometry, or shift between training and test distributions.

Latest papers

172 papers12 papers in the last four weeks, down 50% on the four weeks before. 0.1% of all new papers.

Jul 1, 2026

Unveiling the Non-Monotonic Effect of Privacy on Generalization under Byzantine Robustness

Recent work has established a fundamental trilemma between Byzantine robustness, local differential privacy (LDP), and optimization error in distributed learning. We show that this trilemma does not universally extend to generalization error, but instead depends critically on the privacy regime. Specifically, in the high-noise regime (strong privacy), we prove that increasing privacy reduces the generalization error, i.e., there is no tension between robustness and privacy. In the low-noise regime (weaker privacy), however, the tension between robustness and privacy reappears and increasing privacy indeed degrades generalization. Our theory explains this surprising non-monotonic behavior of the generalization error via matching lower and upper bounds on the algorithmic stability of Byzantine-robust distributed learning under LDP constraints. We corroborate and further analyze these theoretical findings with empirical evaluations.
Jul 1, 2026

Function-Counting Theory for Low-Dimensional Data Structures

The success of deep learning models in classification and regression is widely attributed to the low-dimensional structure that real-world data tend to exhibit, despite their high-dimensional representation. This work attempts to provide a mathematical framework for binary classification on low-dimensional data, building on Cover's (1965) function-counting theory. With our framework, we aim to address the question of how the low-dimensional structure of the data affects the classification capabilities of learning models. Cover's theory relies on a general position assumption that blinds it to the underlying data structure. We refine this assumption to account for the low-dimensionality of the data and derive dichotomy counts that reflect the data structure. We further extend Cover's separation capacity and problem of generalization to the low-dimensional setting, enabling the impact of the underlying data structure on both to be analyzed.
Jun 27, 2026

When Can Conformal Risk Control Certify LLM Outputs? Bounds, Impossibility, and Adaptation for Structured Generation

Large language models (LLMs) deployed for structured generation (NER, JSON extraction, QA, and classification) lack formal reliability guarantees, and standard heuristic abstention policies miss user-specified risk targets by 7.5--12.5%. We characterize when conformal risk control (CRC) can certify structured LLM outputs and when it provably cannot. First, we prove an impossibility result: when the base risk (μ> α), any distribution-free method must abstain on at least ((μ-α)/(1-α)) examples, yielding a closed-form feasibility test: one can check whether CRC will work before running it. Second, we analyze a certification hierarchy across Hoeffding, empirical Bernstein, and a betting-based e-CRC bound, with strict gains in low-variance/large-sample regimes: the Hoeffding-to-Bernstein step delivers the largest gain (+37% certified configurations), while e-CRC adds value when calibration data is scarce (10% certification at 20% data versus 0% for Hoeffding). Third, we validate adaptive conformal inference (ACI) under cross-dataset shift, reducing risk-target violations from 71% to 21%, with residual failures concentrated exactly where the impossibility bound predicts. Across six open-weight models (3B--72B parameters), eight datasets, four tasks, and six nonconformity scores, hard NER/QA/CLS configurations are uncertifiable at (α= 0.10); relaxing to (α= 0.30--0.40) unlocks practical certification (47% NER, 40% QA, 60% CLS). The framework gives a three-step deployment recipe: check feasibility, select the bound and score, then mitigate shift.
Jun 25, 2026

A Generalization Theory for JEPA-Based World Models

Joint Embedding Predictive Architectures (JEPAs) have recently emerged as a promising paradigm for world modeling by learning predictive dynamics in a latent space rather than generating future observations at the input level. Despite their empirical success, the theoretical understanding of JEPA-based world models remains limited. In this paper, we develop the first generalization theory for JEPA-based world models. We formulate JEPA pretraining as a conditional spectral graph learning problem and show that the JEPA objective is equivalent to a low-rank factorization of an action-conditioned co-occurrence matrix. Building on this characterization, we establish a connection between JEPA pretraining error and downstream planning regret, leading to a finite-sample generalization bound for JEPA-based world models. Our analysis reveals an inherent trade-off between approximation and sample errors with respect to the latent dimension, providing theoretical insights into the advantages and limitations of latent predictive models compared with input-level predictive approaches.
Jun 22, 2026

Non-asymptotic estimates of the minimal risk in statistical learning

In this paper we prove some concentration inequalities for two types of error probabilities in the Empirical Risk Principle (ERP) in statistical learning, which provide a lower bound and an upper bound for the minimal risk (in terms of the minimal empirical risk) with non-asymptotic high confidence. The usual boundedness condition of the empirical risk function is relaxed to the Gaussian or exponential integrability condition. The confidence of the lower bound of the minimal risk is shown to be independent of the number of training parameters and the dimension of the input vectors, allowing one to detect the deficiency of a learning machine efficiently; and the confidence of the upper bound of the minimal risk is proved to be high provided that the sample size nn is much greater than the box dimension of the parameter set ΘΘ in the Orlicz metric dψ1d_{ψ_1} associated with the risk functions. Our work is based on Talagrand's concentration inequalities (the sharp versions by Bousquet and Klein-Rio), transport-entropy inequalities and the recent progress in the theory of empirical processes and statistical learning.
Jun 18, 2026

Formalizing and Estimating Task-Space Complexity for Zero-shot Generalization

Generalization is crucial for modern control, but learning across tasks can be costly. Recent work shows that zero-shot transfer, in which a controller designed for a source task is directly deployed to a target task, can achieve strong performance. We formalize task-space complexity, which characterizes the difficulty of achieving sufficiently good performance across a collection of tasks, by casting source selection as a covering problem. We provide bounds on the complexity and demonstrate its applicability.
Jun 17, 2026

On Local Population-Risk Certificates

We develop finite-sample certificates for local population-risk increments Pδv=R(θ0+v)−R(θ0)Pδ_v=R(θ_0+v)-R(θ_0), v∈Dv\in\mathcal D. The primitive object is an expected-valid upper endpoint U^D\widehat{\mathsf U}_{\mathcal D} satisfying Esup⁡v∈D{Pδv−U^D(v)}≤0\mathbb E\sup_{v\in\mathcal D} \{Pδ_v-\widehat{\mathsf U}_{\mathcal D}(v)\}\le0. This uniform criterion certifies any measurable update selected from the same sample and allows penalties to depend on empirical geometry. The main construction is a cross-fitted ridge calibration for linear feature classes. A pilot fold learns the ridge metric, the complementary fold calibrates the squared mean error in that metric, and complete split averaging recovers the full empirical covariance in the directional quadratic form q^X,λ\widehat q_{X,λ}. The optimized diagnostic scale is {q^X,λ(h)r^X,np,λcf/n}1/2\{\widehat q_{X,λ}(h) \widehat r_{X,n_{\rm p},λ}^{\rm cf}/n\}^{1/2}, and the calibrated trace factor r^X,np,λcf\widehat r_{X,n_{\rm p},λ}^{\rm cf} is compared with the ordinary ridge effective dimension r^X,λ\widehat r_{X,λ}. For nonsmooth losses, an exact fixed-mask decomposition δv=Jv0+Rv∘+Cvδ_v=J_v^0+R_v^\circ+C_v separates frozen Taylor fluctuations, good-path remainders, and interface crossings. Applying the linear and composite certificates componentwise yields endpoints for same-sample expected local search and concentrated release rules.
Jun 16, 2026

From Sparse Features to Trustworthy Proxies: Certifying SAE-Based Interpretability

Sparse autoencoders (SAEs) are increasingly used to extract interpretable features from language models (LMs), yet a central question remains: when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM We study this through a post-hoc generalization framework that certifies the LM via a sparse proxy, obtained by replacing a native hidden activation with its pretrained SAE reconstruction. Our framework derives an upper bound on the base model's expected risk using four measurable quantities: proxy risk, SAE reconstruction gap, concept-pool mismatch, and sparse complexity. We interpret this certificate as an operational criterion for explanatory faithfulness. In particular, a non-vacuous bound indicates that the extracted sparse features retain meaningful predictive information, while small reconstruction and mismatch errors indicate that the proxy remains behaviorally close to the original model. Empirically, we show that the bound becomes non-vacuous on GPT-2 Small, Gemma-2B, and Llama-3-8B at practical sample sizes. A detailed layerwise analysis of Llama-3-8B reveals a strong depth dependence, with later layers becoming much easier to certify, associated with both stronger local fidelity and weaker downstream error amplification. Finally, through feature-shuffling ablations, we show that the decomposition distinguishes genuine semantic alignment from mere statistical sparsity, providing a useful diagnostic for when SAE-based explanations become less reliable.
Jun 16, 2026

Generalization Guarantees for Multi-Input Neural Operator Learning in Sobolev Spaces

We develop approximation and generalization error estimates for multi-input neural operators, with the output error measured in Sobolev norms. In contrast to standard operator-learning settings with a single input function, our framework allows multiple input functions defined on possibly different domains, with different dimensions and Sobolev regularities. The derived rates explicitly quantify the contribution of each input space to the final error bound. In particular, in the balanced regime, the approximation and generalization rates are governed by the interaction between the input dimensions, regularities, and Sobolev orders, while the dependence on the model complexity retains a log⁡log⁡/log⁡\log\log/\log-type structure. Our analysis provides a general theoretical framework for multi-input operator learning, including Sobolev training, and is applicable to operator learning problems arising from partial differential equations and scientific computing.
Jun 15, 2026

Upper Bounds on the Generalization Error of Deep Learning Models via Local Robustness and Stability

Generalization is a critical property of data-driven models, particularly deep learning models deployed in safety-critical applications. Robustness-based generalization bounds have gained attention as a principled way to link robustness properties to generalization performance, often in a data-dependent manner. However, most existing bounds suffer from vacuousness in practical settings, yielding loose upper bounds that greatly exceed the actual error rates and limiting their usefulness for real-world evaluation. While this issue is often attributed to the uncertainty term, a substantial part of the problem originates from the robustness term itself, particularly for the 0-1 loss. Existing approaches typically treat the robustness term as a global measure, ignoring its variation across different sub-regions of the input space. In this work, we propose a generalization bound that addresses this limitation by scaling the robustness term according to the number of stable and unstable samples within each sub-region. Our bounds incorporate both data- and model-dependent factors while maintaining practical relevance (yielding tighter upper bounds on true error). Experiments on models trained on the ImageNet dataset show that our bounds remain consistently non-vacuous and achieve the tightest estimates among existing methods, closely aligning with empirical performance across a range of robust deep neural networks.
Jun 14, 2026

Brownian Kernel Ladders

Constructing mathematically tractable function spaces that capture hierarchical compositional representations remains a central challenge in statistical learning theory. We introduce Brownian kernel ladders (BKLs), a recursively defined hierarchy of integral reproducing kernel Hilbert spaces generated through Brownian-kernel integral constructions. Starting from linear functionals, each layer is obtained by integrating Brownian kernels over probability measures supported on subsets of the previous layer, yielding a recursive function-space model in which depth is encoded directly through the hierarchy. Based on this framework, we define canonical BKL spaces together with an associated complexity functional. We establish several analytical and statistical properties of these spaces. In particular, we show that BKL spaces form quasi-Banach spaces, satisfy depth-dependent Hölder regularity estimates, and exhibit strict monotonicity with respect to depth. We further prove existence results for regularized empirical risk minimization and derive Gaussian complexity bounds that remain uniformly controlled with respect to both the ambient dimension and the hierarchy depth. A key ingredient of the analysis is a combinatorial proof technique based on recursive subset decompositions and Brownian-kernel threshold representations. These estimates yield excess-risk guarantees of near-parametric order for regularized empirical risk minimization over BKL spaces. Our results provide a mathematically tractable hierarchical function-space framework for studying compositional representations in deep learning.
Jun 14, 2026

The Data Manifold under the Microscope

A significant gap exists between theory and practice in deep learning. Generalization and approximation error bounds are often derived for simplified models or are too loose to be informative. Many rely on the manifold hypothesis and on geometric regularity such as intrinsic dimension, curvature, and reach. Progress requires insight into data-manifold geometry and suitable benchmarks, yet existing options are polarized: analytic manifolds with known geometry but limited applicability, or real-world datasets where geometry is only coarsely estimable. We introduce a benchmarking framework for studying data geometry. We repurpose and extend dSprites and COIL-20 with additional transformation dimensions and dense, axis-aligned sampling, and pair them with finite-difference estimators that recover curvature, reach, and volume at near-ground-truth accuracy in a regime where general-purpose estimators are unreliable or difficult to deploy. The framework is intended as a controlled testbed, useful as a calibration environment for geometric estimators and a sandbox for probing theoretical assumptions. To illustrate its use, we present two application studies, namely assessing the scaling behavior of the bounds of Genovese et al. and Fefferman et al., and tracking the layer-wise geometry of a ββ-VAE, highlighting the behavior of current bounds and the value of controlled benchmarks for guiding and validating future theory. A reference implementation is available at https://github.com/koulakis/manifold-microscope.
Jun 11, 2026

Beyond LoRA: Is Sparsity-Induced Adaptation Better?

Low-rank adaptation (LoRA) and its variants provide a memory- and compute-efficient alternative to full fine-tuning of pre-trained models. However, questions remain about the comparative generalizability of these approaches and how the structural restrictions on low-rank updates preserve effective adaptation performance. We present a historical framing, covering the past (full fine-tuning and original LoRA), the present (different variants of LoRA), and propose simpler, cheaper, parameter-efficient extensions by inducing sparsity within existing LoRA variants: Cheap LoRA (cLA), training a single low-rank factor with the other fixed (deterministically or, in its randomized variant, stochastically), and the chained circulant variant, c3{c}^3LA. We frame cLA as a structured instance of asymmetric LoRA, serving as a controlled column-subspace restriction of full fine-tuning. We derive information-theoretic generalization error bounds for these variants, marking one of the first endeavors in this area. Empirically, we evaluate 11 fine-tuning methods across 10 pre-trained models and 14 datasets, analyzing the fine-tuned models' performance and generalization using tools such as loss landscapes and spectral analysis. Despite the sensitivity of fine-tuned models to the pre-trained model, datasets, and other factors, our study suggests that restricting LoRA-based PEFT methods' adaptation to a sparse, structured column space remains competitive across tasks with their parameter-matched baselines while reducing up to 10% training time and peak GPU memory up to 15%, even with a naïve, non-optimized, sparse implementation. Our theoretical and empirical generalization measures provide a more consistent and principled approach to their cost-effective adaptation than commonly used analytical tools. Overview and code are available at: https://elicaden.github.io/Beyond_LoRA/.
Jun 9, 2026

Bidirectional Random Projections

This paper analyzes bidirectional random projections for ordinary least squares (OLS) regression under the fixed design setting. Let (X,Y)∈Rn×p×Rn(X, Y) \in \mathbb{R}^{n\times p}\times\mathbb{R}^{n} be a sample and W∈Rn1×nW \in \mathbb{R}^{n_1\times n}, R∈Rp×p1R \in \mathbb{R}^{p\times p_1} be two random projections. We develop an expected excess loss bound for the OLS estimator built on (WXR,WY)(WXR, WY), extending an established bound for the OLS estimator built on (XR,Y)(XR, Y). Our analysis shows that our bound is larger and the gap between the two bounds decays at a rate of approximately O(1/n1)O(1/n_1). These findings are confirmed by numerical results.
Jun 7, 2026

Generalization in Nonlinear Least Squares via Learned Feature Geometry

We study the generalization of ridge-regularized nonlinear least-squares models via on-average algorithmic stability, deriving error bounds for local minimizers in terms of a data-dependent effective dimension that reflects the geometry of the gradient model at the trained parameters, through the empirical Jacobian Gram matrix and a residual-curvature term. In the linear case, where the curvature term vanishes, this recovers the classical effective dimension of the Jacobian kernel covariance, but evaluated at the trained model rather than at initialization as is typical in neural tangent kernel analyses. We further bound this effective dimension via covering complexity of the gradient features, leading to guarantees that depend on learned geometry rather than parameter count. In particular, for manifold-supported data and piecewise Lipschitz Jacobians, the bounds scale with intrinsic dimension, while for one-hidden-layer ReLU networks, the mechanism can be made explicit through counts of activation-stable regions. Experiments on synthetic manifolds, clustered distributions, and benchmark datasets illustrate trained-Jacobian compression, the tightness of the residual-curvature linearization, and agreement between the stability bound and observed generalization gaps. A key feature of our bounds is the simplicity of their derivation, which follows from first principles using the Brascamp-Lieb inequality under strongly log-concave noise.
Jun 7, 2026

A Joint Finite-Sample Certificate for Adaptive Selective Conformal Risk Control

Selective predictors answer on confident inputs and abstain elsewhere; deploying one safely needs a single finite-sample certificate that simultaneously upper-bounds the selected risk, lower-bounds the acceptance probability \pacc\pacc above a floor \pmin\pmin, and lower-bounds the deployment utility. This certificate must be valid under adaptive threshold selection from a finite grid of mm pairs on \ncert\ncert samples. We give such a certificate for bounded, possibly non-monotone losses by treating the selected risk directly as a ratio rather than through a Hoeffding-style range bound. The construction couples three confidence bounds: a variance-adaptive empirical-Bernstein bound on the ratio risk, a Clopper--Pearson bound on acceptance, and a two-sided closeness bound on utility. Together they lower-bound the certified policy's utility absolutely and to within 2\gammau2\gammau of the best over the certified set, both non-vacuous whenever feasible; a regime-scoped third leg matches an external oracle, informative only where the risk margin \gammar<α\gammar < α and vacuous at the headline operating points. Relative to the range-only Hoeffding-ratio construction this sharpens the acceptance-floor dependence from 1/\pmin1/\pmin to 1/\pmin1/\sqrt{\pmin}, and a closed-form corollary identifies a per-pair regime in which our risk bound dominates a Hoeffding conformal risk control (Hoeffding--CRC) selective bound. Empirically, on ImageNet (three ResNets) and COCO val 2017 panoptic, the certificate opens a +22+22 pp certified-acceptance frontier over Hoeffding--CRC and is ≈10×{\approx}10{\times} tighter than a non-vacuous matched-valid baseline; these gains are regime-scoped, not universal, and absent on ADE20K. The certifier runs in O(\ncertm)O(\ncert m) time.
Jun 5, 2026

Uniform Stability and Generalization Error of GD and SGD on Fixed-Point Parameters

We analyze generalization error, uniform stability, and uniform argument stability of gradient descent (GD) and stochastic gradient descent (SGD) over discrete parameter spaces, where each update involves deterministic or stochastic rounding. We show that deterministic rounding degrades the generalization error of GD on convex, Lipschitz, and smooth loss functions, increasing the rate from O(T/n)O(T/n) to O(T/n)O(T/\sqrt{n}), and establish matching lower bounds. We further prove that uniform stability of GD becomes Ω(T)Ω(T), showing that stability-based generalization bounds are vacuous in this setting. In contrast, for the same losses, stochastic gradient descent with deterministic rounding admits nontrivial uniform stability guarantees, which differ qualitatively from the real-valued case and exhibit distinct dependencies on the number of iterations and the dimension: we prove tight bounds O(T/n)O(T/n) for one dimension and O(T2/n)O(T^2/n) for higher dimensions. We also show that stochastic rounding can introduce generalization error that increases with the dimension; such a phenomenon is absent in standard real-valued optimization and in the deterministic rounding case. Finally, we provide upper bounds on uniform argument stability for stochastic rounding schemes and show that these bounds are tight when the loss can be represented as a sum of coordinate-wise functions.
Jun 5, 2026

Stability beyond Bounded Differences: Sharp Generalization Bounds under Finite LpL_p Moments

While algorithmic stability is a central tool for understanding generalization of learning algorithms, existing high-probability guarantees typically rely on uniform boundedness or sub-Gaussian/sub-Weibull tail assumptions, which can be overly restrictive for modern settings with heavy-tailed or unbounded losses. We develop a stability-based framework that requires only a finite LpL_p moment condition. Our first contribution is sharp concentration inequalities for functions of independent random variables under LpL_p constraints, extending McDiarmid's bounded-differences techniques beyond the classical regime. Leveraging these results, we derive sharp high-probability generalization bounds across a range of learning paradigms, including empirical risk minimization, transductive regression, and meta-learning. These guarantees show that LpL_p stability suffices for robust generalization even when boundedness fails, substantially weakening the standard assumptions in the stability literature.
Jun 4, 2026

Generalization in Deep Neural Networks: Minimax Rates for Gradient Methods

Understanding the generalization performance of over-parameterized neural networks has become a central topic in deep learning theory. While recent advances, particularly works under the Neural Tangent Kernel (NTK) regime, have shed light on the behavior of shallow architectures, the statistical generalization properties of deep neural networks (DNNs), especially in regression tasks, remain far less understood. In this paper, we make significant progress toward closing this gap by providing a comprehensive generalization analysis of DNNs trained using gradient-based methods. First, we establish, for the first time, a crucial connection between the learning dynamics of a DNN with smooth activation functions trained via gradient-based methods and those of kernel methods, showing that gradient-based methods on over-parameterized DNNs can fully inherit the favorable learning dynamics of their kernel counterparts. Building on this connection and the well-established optimality of kernel methods, we derive the first known minimax-optimal rates for the excess population risk of both gradient descent (GD) and stochastic gradient descent (SGD), under the assumption that network width scales polynomially with the sample size. Our results demonstrate that, with sufficient width, DNNs trained by GD or SGD can achieve generalization performance comparable to kernel-based methods.
Jun 1, 2026

Measurement-Budget Allocation in Quantum Learning with Finite-Shot Generalization Guarantees

On near-term quantum hardware, estimating a Born probability requires repeated circuit executions. A quantum learning experiment with a fixed measurement budget BB must therefore decide how many distinct training states nn to use and how many shots SS to spend on each. We study this trade-off for binary quantum classifiers whose measurement operator MM is learned from the training data and evaluated with fresh measurement outcomes. We prove a distribution-free generalization bound for bounded, LL-Lipschitz losses with a sample term of order d/n\sqrt{d/n} and a shot term of order (log⁡n)/S\sqrt{(\log n)/S}; under B=nSB=nS the two terms move in opposite directions. Minimizing a closed-form surrogate of the bound gives the allocation rule n∗=22dB/log⁡(4B/δ)n^* = 2\sqrt{2dB/\log(4B/δ)}, S∗=B/n∗S^* = B/n^*, whose bound value is within 0.1% of the exact numerical minimum on all budgets we evaluate, and a worst-case rate of B−1/4B^{-1/4}. PennyLane simulations with 2- and 4-qubit variational circuits on nine synthetic benchmarks show that the empirical gap stays below the bound in all configurations, although the bound is vacuous at the simulated budgets and becomes non-vacuous only for larger budgets (e.g., B≳5000B \gtrsim 5000 for d=4d=4). For a loss that is concave in the score, the empirical gap is U-shaped under a fixed budget and the measured shot contribution matches its analytical value; for the absolute loss, shot noise is unbiased and no shot-limited regime appears. The rule can be used to plan finite-shot evaluations before an experiment is run.
Jun 1, 2026

Regime-Arrival Uncertainty in Generalization Bounds under Distribution Shift

The standard generalization bounds assume that the training and deployment distributions are the same, or are static, and don't consider regime switching environments where the ratio of calm vs crisis states is different. This paper proposes a framework that generalizes regime-aware models by quantifying the extra risk due to regime composition mismatch, when distribution shifts are Markov-switching. We obtain an exact decomposition, separating regime mismatch from regime sensitivity; we extend the bound to beta-mixing data using the effective sample size corrected for the spectral gap; and we show a minimax lower bound for synthetic data and on 25 years of global equity indices. The proposed penalty is an ex post realized generalization gap, whereas the training-only estimator does not show significant correlation: the feature geometry of crises can be detected, but not the temporal arrival. Thus, the framework is not a forecast machine. Forecasting the composition of the future regime is an open question in the rare cases of regime change.
Jun 1, 2026

A Note on Stability for Orthogonalized Matrix Momentum with Client Sampling

We study finite-sample generalization for a client-sampled distributed optimization scheme with matrix-valued parameters and orthogonalized momentum updates. The central quantity is the gap between the population and empirical objectives at the returned model when only a subset of clients participates in each round. Under independent heterogeneous client data, unequal local sample counts, and fixed aggregation weights, we derive a finite-round upper-tail guarantee from a coupled-neighbor stability recursion and a weighted concentration step. The bound keeps the client-selection counts through the amplification factor Yi(C)Y_i(\mathcal C); in the uniform full-participation full-batch regime, it yields O~(n−1+n−1/2)\widetilde{\mathcal O}(n^{-1}+n^{-1/2}) scaling whenever the horizon-dependent amplification terms are controlled. The matrix-orthogonalization rule is required to be Lipschitz along paired trajectories, a condition satisfied by regularized polar-type maps and normalized finite-step Newton--Schulz orthogonalizers. For the unregularized matrix sign, the same argument requires coupled spectral separation, whereas Gaussian smoothing gives a finite-round smoothed variant. A one-dimensional counterexample shows why a gap, smoothing, or regularity condition is necessary.
May 30, 2026

A Practical Upper Bound on Selection Bias Effects in Medical Prediction Models

Selection bias is a common and often unavoidable aspect of real-world data that challenges the generalizability of machine learning models. When models trained on biased data are deployed in the broader target population, poor model generalization may lead to real harm, particularly in high-risk settings such as healthcare. This risk highlights the need for practitioners to reliably assess model generalizability prior to deployment. However, existing methods for predicting model performance rely on unrealistic access to the target distribution or knowledge of the selection mechanism causing bias. To address these limitations, we propose a novel upper bound on the worst-case model performance on the target population under the realistic setting where the selection mechanism and the target population data are only partially observed. We demonstrate the validity and practical utility of our method through experiments on fully synthetic data, semi-synthetic data derived from the All of Us Research Program, and real-world selection bias in MIMIC-IV. Our work offers a principled and practical tool to estimate the impact of selection bias in an otherwise intractable setting, thereby enabling practitioners to build safer and more generalizable models in healthcare and beyond.
May 27, 2026

Stochastic Gradient Descent with Momentum is Algorithmically Stable

Stochastic gradient descent with momentum (SGDM) is one of the most widely used optimization algorithms in machine learning. While optimization properties of SGDM have been extensively studied in the literature, it remains insufficiently understood whether and when SGDM can generalize well to unseen data. In particular, it has been conjectured that while momentum accelerates training, it may degrade generalization. In this paper, we close this gap by developing a comprehensive generalization analysis of SGDM through the lens of algorithmic stability. More specifically, we introduce a generalized SGDM framework that encompasses both Polyak's and Nesterov's momentum schemes, and establish tight on-average model stability bounds for smooth and convex problems. Notably, the obtained bounds exploit small optimization error bounds along the trajectory, apply to any momentum parameter in the interval [0,1)[0, 1), and do not require the commonly assumed Lipschitzness of loss functions. We further derive optimization error bounds for the generalized SGDM, and combine them with our generalization analyses to obtain optimal excess population risk bounds for SGDM with both Polyak's and Nesterov's momentum.
May 27, 2026

Learning Theory of the SVRG: Generalization and Convergence Analysis

Variance reduction (VR) methods employ stochastic gradients with decreasing variance, and they have been widely applied to solve large-scale optimization problems in machine learning because of their efficiency. Existing theoretical studies of VR methods are mainly focused on the convergence analysis, leaving the generalization behavior largely unexplored. In this paper, we bridge this gap by developing the first non-vacuous generalization analysis of the representative VR method: Stochastic Variance Reduced Gradient (SVRG), through the lens of algorithmic stability. In particular, we establish sharp stability bounds of the SVRG in both convex and strongly convex settings by exploiting its algorithmic structure. The obtained bounds are data-dependent, because the training errors are incorporated along the trajectory. Our analysis clarifies the interplay between optimization and generalization, leading to optimal excess population risk bounds in both settings. Our approach differs substantially from existing analyses of stochastic algorithms in the sense that we decompose the SVRG update as an SGD-like step plus a zero-mean correction term and then introduce novel Lyapunov functions to absorb the additional gradient terms induced by the reference points. Our analytical framework can be generalized to other VR methods, and we demonstrate the generalization by the well-known Stochastic Average Gradient Accelerated (SAGA) method.
May 25, 2026

A PAC-Bayesian View of Generalisation for Physics-Informed Machine Learning

Physics-informed machine learning (PIML) integrates mechanistic knowledge, typically in the form of partial differential equations (PDE), into data-driven models. Despite strong empirical performance, its statistical generalisation properties remain poorly understood, particularly in the regression setting with unbounded losses. Existing analyses rely on approximation or stability arguments and do not fully capture how physical structure influences generalisation from finite data. In this work, we develop a PAC-Bayesian framework for PIML that provides high-probability generalisation guarantees in the presence of unbounded losses. We adopt a multi-task perspective that jointly treats data fidelity, PDE residuals, initial and boundary conditions, avoiding the looseness induced by standard union-bound approaches. Our analysis leverages the structure of physics-informed objectives to derive novel bounds where the complexity scales with input-gradient norms of the losses, revealing a direct link between physical regularity and generalisation. We instantiate this framework under Sobolev and Poincaré-type assumptions, yielding two classes of bounds that trade off statistical complexity and smoothness in different regimes. Building on these results, we propose a self-bounding-aware learning algorithm that directly optimises tractable surrogates of the derived bounds, along with a practical procedure to estimate the associated constants in realistic settings. Empirical evaluations on standard PDE benchmarks demonstrate that our bounds are non-vacuous, significantly tighter than union-bound baselines, and can be effectively minimised during training. Overall, our results provide a principled statistical foundation for the generalisation of physics-informed models.
May 20, 2026

On the Regularity and Generalization of One-Step Wasserstein-guided Generative Models for PDE-Induced Measures

Despite the remarkable empirical success of generative models, the available theory on their statistical accuracy in scientific computing remains largely pessimistic. This paper develops a theoretical framework for understanding the regularity of transport maps and the generalization properties of one-step Wasserstein-guided generative models for PDE-induced probability measures. We consider normalized target densities associated with linear elliptic and parabolic equations on bounded domains, as well as diffusion and Fokker--Planck equations on the torus. Under standard structural assumptions, we prove that these target measures satisfy doubling conditions. By combining this fact with regularity theory for optimal transport between doubling measures, we show that the optimal transport map from a uniform source measure to the target measure is Hölder continuous. This regularity yields an approximation-theoretic justification for one-step generative models that learn PDE-induced distributions via a single pushforward map. As a representative instance, we study DeepParticle and derive excess-risk bounds characterizing the discrepancy between the learned map and the population-optimal map. We also establish a robustness estimate under target shift and illustrate the theory with experiments which support the derived rates.
May 19, 2026

A Measure-Theoretic Analysis of Reasoning in Transformers: Structural Generalization and Approximation Limits

Scaling laws for reasoning in large language models (LLMs) are well documented, but little theory explains their out-of-distribution (OOD) generalization. We formalize reasoning with optimal transport: each search trajectory is mapped to twelve structural coordinates, such as tree depth and backtracking rate, and dataset shift is measured by the Wasserstein-1 distance. Kantorovich duality bounds the OOD risk through this distance, the architecture's Lipschitz constant and approximation limits. First, under absolute positional encoding (APE), shifting part of a trajectory to later positions almost surely changes the attention scores within that part; if this misroutes exact-match search, the Lipschitz constant and risk are Ω(1)Ω(1). Shift-invariant encodings such as rotary position embedding (RoPE) leave the scores unchanged and bound the error. Second, we map backtracking to a Dyck-kk language and, under an implicit-depth assumption, derive a risk floor for search trees deeper than a multiple of the number of layers. Enough layers remove this floor; more width only shrinks a separate Barron-space approximation term. Experiments with 54 trained Transformers on combinatorial search (Countdown) support these bounds. Appendix experiments measure structural shifts in public agent, embodied and robot trajectories, shift all position indices of 33 public language models, and split deeply nested tasks into separate model calls.
May 18, 2026

Pointwise Generalization in Deep Neural Networks

We address the fundamental question of why deep neural networks generalize by establishing a pointwise generalization theory for fully connected networks. This framework resolves long-standing barriers to characterizing the rich nonlinear feature-learning regime and builds a new statistical foundation for representation learning. For each trained model, we characterize the hypothesis via a pointwise Riemannian Dimension, derived from the eigenvalues of the learned feature representations across layers. This establishes a principled framework for deriving hypothesis-dependent, representation-aware generalization bounds. These bounds offer a systematic upgrade over approaches based on model size, products of norms, and infinite-width linearizations, yielding guarantees that are orders of magnitude tighter in both theory and experiment. Analytically, we identify the structural properties and mathematical principles that explain the tractability of deep networks. Empirically, the pointwise Riemannian Dimension exhibits substantial feature compression, decreases with increased over-parameterization, and captures the implicit bias of optimizers. Taken together, our results indicate that deep networks are mathematically tractable in practical regimes and that their generalization is sharply explained by pointwise, feature-spectrum-aware complexity.
May 18, 2026

Shallow ReLUs^s Networks in LpL^p-Type and Sobolev Spaces: Approximation and Path-Norm Controlled Generalization

This paper studies approximation by shallow ReLUs^s networks, σs(t)=max⁡{0,t}sσ_s(t)=\max\{0,t\}^s, together with their generalization behavior under ℓ1\ell_1 path-norm control. For the LpL^p-type integral spaces F~p,τd,s\widetilde{\mathcal{F}}_{p,τ_d,s}, 1≤p≤21\le p\le2, spherical harmonic analysis yields approximation bounds for shallow networks. In particular, when τdτ_d is the uniform measure and 1≤p<21\le p<2, the approximation rate is O ⁣(m−p(2s+2d+1)−2d2dp)O\!\left(m^{-\frac{p(2s+2d+1)-2d}{2dp}}\right) for 1≤p≤p∗1\le p\le p^* and O ⁣(m−p(4s+3d−1)−2d+24dp)O\!\left(m^{-\frac{p(4s+3d-1)-2d+2}{4dp}}\right) for p∗<p<2p^*<p<2, where p∗=2d+2d+3p^*=\frac{2d+2}{d+3}. Approximation bounds for Sobolev spaces Wα,pW^{α,p}, 1≤p<21\le p<2, are obtained through embeddings into spectral Barron spaces. For nonparametric regression with sub-Gaussian noise, path-norm-regularized shallow ReLUs^s networks achieve minimax-optimal rates O ⁣(n−d+2s+12d+2s+1log⁡n)O\!\left(n^{-\frac{d+2s+1}{2d+2s+1}}\log n\right) over Bs\mathscr{B}_s and O ⁣(n−2α2α+dlog⁡n)O\!\left(n^{-\frac{2α}{2α+d}}\log n\right) over Wα,∞W^{α,\infty}, with matching lower bounds up to logarithmic factors.