Learning under Distributional Drift: Prequential Reproducibility as an Intrinsic Statistical Resource

Statistical learning under distributional drift remains poorly characterized, especially in closed-loop settings where learning alters the data-generating law. We introduce an intrinsic drift budget $C_T$ that quantifies the cumulative information-geometric motion of the data distribution along the realized learner-environment trajectory, measured in Fisher-Rao distance (the Riemannian metric induced by Fisher information on a statistical manifold of data-generating laws). The budget decomposes this motion into exogenous change (environmental drift that would occur without intervention) and policy-sensitive feedback contributions (drift induced by the learner's actions through the closed loop). This yields a rate-based characterization: in prequential reproducibility bounds -- where performance on the realized stream is used to predict one-step-ahead performance under the next distribution -- the drift contribution enters through the average drift rate $C_T/T$, i.e., normalized cumulative Fisher-Rao motion per time step. We prove a drift--feedback bound of order $T^{-1/2} + C_T/T$ (up to a controlled second-order remainder) and establish a matching minimax lower bound on a canonical subclass, showing this dependence is tight up to constants. Consequently, when $C_T/T$ is nonnegligible, one-step-ahead reproducibility admits an irreducible accuracy floor of the same order. Finally, the framework places exogenous drift, adaptive data analysis, and performative feedback within a common geometric account of distributional motion.

• Publication year

  2025

• Venue

  Unknown venue

• Publication date

  2025-12-15

• Fields of study

  Mathematics, Computer Science

• Identifiers
  arXiv 2512.13506
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• Optimal Adaptive Experimental Design for Estimating Treatment Effect

  2024cited by this paper
• How to Learn when Data Reacts to Your Model: Performative Gradient Descent

  2021cited by this paper
• Adaptivity and Non-stationarity: Problem-dependent Dynamic Regret for Online Convex Optimization

  2021cited by this paper
• High‐dimensional Statistics: A Non‐asymptotic Viewpoint, Martin J.Wainwright, Cambridge University Press, 2019, xvii 552 pages, £57.99, hardback ISBN: 978‐1‐1084‐9802‐9

  2020influential reference
• Performative Prediction

  2020influential reference
• PAC-Bayesian Theory for Domain Adaptation

  2019cited by this paper
• Guaranteed Validity for Empirical Approaches to Adaptive Data Analysis

  2019cited by this paper
• Fisher Information and Natural Gradient Learning of Random Deep Networks

  2018cited by this paper
• Information-theoretic analysis of generalization capability of learning algorithms

  2017cited by this paper
• Domain Generalization by Marginal Transfer Learning

  2017cited by this paper
• Time series prediction and online learning

  2016cited by this paper
• Stochastic thermodynamics of learning

  2016cited by this paper
• Algorithmic stability for adaptive data analysis

  2015cited by this paper
• Generalization in Adaptive Data Analysis and Holdout Reuse

  2015cited by this paper
• Controlling Bias in Adaptive Data Analysis Using Information Theory

  2015cited by this paper
• Understanding Machine Learning - From Theory to Algorithms

  2014cited by this paper
• Dynamical Models and tracking regret in online convex programming

  2013cited by this paper
• Non-Stationary Stochastic Optimization

  2013cited by this paper
• Thermodynamics as a theory of decision-making with information-processing costs

  2012cited by this paper
• Information geometry and its applications

  2012cited by this paper
• New Analysis and Algorithm for Learning with Drifting Distributions

  2012cited by this paper
• The thermodynamics of prediction

  2012cited by this paper
• Introduction to Nonparametric Estimation

  2008influential reference
• Learning Theory

  2004influential reference
• Learning Rates for Q-learning

  2004cited by this paper
• On-line learning in changing environments with applications in supervised and unsupervised learning

  2002cited by this paper
• Reinforcement Learning: An Introduction

  1998cited by this paper
• Riemannian Manifolds: An Introduction to Curvature

  1997cited by this paper
• Information and the Accuracy Attainable in the Estimation of Statistical Parameters

  1992cited by this paper
• Statistical decision rules and optimal inference

  1982influential reference
• Learning under distributional shift

  year unknowncited by this paper

• No citing papers are available for this paper.