paper-with-me

홈 › Papers

Information-Theoretic Generalization Bounds for Sequential Decision Making

2026-05-12 · Futoshi Futami, Masahiro Fujisawa arxiv

Information-theoretic generalization bounds based on the supersample construction are a central tool for algorithm-dependent generalization analysis in the batch i.i.d.~setting. However, existing supersample conditional mutual information (CMI) bounds do not directly apply to sequential decision-making problems such as online learning, streaming active learning, and bandits, where data are revealed adaptively and the learner evolves along a causal trajectory. To address this limitation, we develop a sequential supersample framework that separates the learner filtration from a proof-side enlargement used for ghost-coordinate comparisons. Under a row-wise exchangeability assumption, the sequential generalization gap is controlled by sequential CMI, a sum of roundwise selector--loss information terms. We also establish a Bernstein-type refinement that yields faster rates under suitable variance conditions. The selector-SCMI proof strategy applies to online learning, streaming active learning with importance weighting, and stochastic multi-armed bandits.

📄 PDF Abstract BibTeX arXiv:2605.12190

Code (0)

등록된 구현이 없습니다.

Tasks

Multi-Armed BanditsDecision MakingActive Learning

Similar Papers 제목 키워드 기반

Information-Theoretic Generalization Bounds of Replay-based Continual Learning

2025-07-16 · Wen Wen, Tieliang Gong, Yunjiao Zhang, Zeyu Gao 외

Continual learning (CL) has emerged as a dominant paradigm for acquiring knowledge from sequential tasks while avoiding catastrophic forgetting. Although many CL methods have been proposed to show impressive empirical pe…

Continual LearningGeneralization Bounds

An Information-Theoretic Analysis of OOD Generalization in Meta-Reinforcement Learning

2025-10-27 · Xingtu Liu arxiv

In this work, we study out-of-distribution (OOD) generalization in meta-reinforcement learning from an information-theoretic perspective. We begin by establishing OOD generalization bounds for meta-supervised learning un…

Reinforcement Learning

Formal limitations of sample-wise information-theoretic generalization bounds

2022-05-13 · Hrayr Harutyunyan, Greg Ver Steeg, Aram Galstyan

Some of the tightest information-theoretic generalization bounds depend on the average information between the learned hypothesis and a single training example. However, these sample-wise bounds were derived only for exp…

Generalization Bounds

Generic Bounds on the Maximum Deviations in Sequential Prediction: An Information-Theoretic Analysis

2019-10-11 · Song Fang, Quanyan Zhu

In this paper, we derive generic bounds on the maximum deviations in prediction errors for sequential prediction via an information-theoretic approach. The fundamental bounds are shown to depend only on the conditional e…

Prediction

Information Theoretic Lower Bounds for Information Theoretic Upper Bounds

2023-02-09 · NeurIPS 2023 11 · Roi Livni

We examine the relationship between the mutual information between the output model and the empirical sample and the generalization of the algorithm in the context of stochastic convex optimization. Despite increasing in…

Generalization Bounds