E-Values

Ask about this topic

Overview

E-values quantify evidence against a null hypothesis and can be multiplied or accumulated over time to support tests that remain valid even when analysts stop early or adapt their decisions. Common approaches include betting strategies and test supermartingales for sequential testing, conformal e-values and e-BH for selection and risk control, and likelihood-ratio methods for monitoring shifts or model performance. Applications include prediction, language-model evaluation, private testing, and adaptive search, where benchmarks assess power, data use, coverage, and false discoveries. A central tension is balancing strong evidence and efficient stopping against reliable type-II error guarantees, especially under adaptive choices, distribution shifts, or limited assumptions about the data.

Latest papers

CardsList

15 papers3 papers in the last four weeks, against 1 the four weeks before. 0.0% of all new papers.

  1. Valid Stopping in Adaptive Generator-Verifier Loops

    Oct 5, 2026Mahmoud Hegazy, Michael I. Jordan, Aymeric DieuleveutFDR ControlE-Values

    CMAP, École polytechnique, IP Paris, France · Inria Paris, France · EECS, UC Berkeley, USA

  2. Evidence-Gated Research: Statistically Controlled Model Adoption in Adaptive Search

    Oct 1, 2026Yifan GuoFDR ControlAnytime-Valid Inference

    Department of Applied Mathematics and Statistics Stony Brook University

  3. Prequential E-Values for Selected-GP Near-Optimality Certificates

    Sep 30, 2026Ami Tavory, Noa CohenBlack-Box OptimizationAnytime-Valid Inference

    Meta Platforms

  4. Discovering Translation-Worthy Languages with E-Values

    Sep 6, 2026Wajdi Ben Saad, Safa MadiouniMultilingual Text ClassificationText Classification

    Carthago Labs Paris, France · IÉSEG Paris, France

  5. The Type-II Error of Test Supermartingales: e-Power versus the Chernoff-Stein Exponent

    Aug 19, 2026Patrick ForréStatistical Hypothesis TestingSequential Hypothesis Testing

    AI4Science Lab, Korteweg-de Vries Institute for Mathematics, University of Amsterdam

  6. Type-II Error Bounds for Test Supermartingales from Lower-Tail Hypotheses

    Aug 19, 2026Patrick ForréSequential Hypothesis TestingE-Values

    AI4Science Lab Korteweg-de Vries Institute for Mathematics University of Amsterdam

  7. NxN E-valuation: Hypothesis Certification via a Conformal CRT Null

    Aug 6, 2026Bin Wang, Yan Zhong, Zhengyu Zhang +3LLM Hallucination MitigationE-Values

    Meta

  8. Anytime-Valid Confirmation of Label-Shift Corrections

    Jun 12, 2026Seungjin ChoiLabel ShiftAnytime-Valid Inference

    CROID Research and aSSIST University, Seoul, Korea.

  9. Set-Preserving Calibration from Conformal P-Values to E-Values

    Jun 2, 2026Nabil Alami, Jad Zakharia, Souhaib Ben TaiebE-ValuesConformal Prediction

    Department of Statistics and Data Science, MBZUAI · Ecole Normale Superieure de Cachan · Department of Computer Science, University of Mons

  10. Optimal Rates for Differentially Private Hypothesis Testing with E-values

    May 27, 2026Ben Jacobsen, Tomas Gonzalez, Gavin Brown +2E-ValuesDifferential Privacy

    University of Wisconsin-Madison · Carnegie Mellon University · University of Wisconsin–Madison

  11. Semi-Supervised Hypothesis Testing by Betting on Predictions

    May 27, 2026Yaniv Tenzer, Elad Tolochinsky, Yaniv RomanoE-ValuesSemi-Supervised Learning

    Department of Computer Science, Technion – Israel Institute of Technology · Department of Electrical and Computer Engineering, Technion – Israel Institute of Technology

  12. Towards E-Value Based Stopping Rules for Bayesian Deep Ensembles

    Apr 20, 2026Emanuel Sommer, Rickmer Schulte, Sarah Deubner +2Markov Chain Monte CarloEarly Stopping

    LMU Munich, MCML

  13. Beyond Fixed False Discovery Rates: Post-Hoc Conformal Selection with E-Variables

    Apr 13, 2026Meiyi Zhu, Osvaldo SimeoneFDR ControlE-Values

    Department of Engineering, King’s College London, WC2R 2LS, London, U.K. · Institute for Intelligent Networked Systems, Northeastern University London, E1 8PH London, U.K.

  14. Conformal Selective Prediction with General Risk Control

    Mar 25, 2026Tian Bai, Ying JinSelective PredictionConformal Risk Control

    Department of Statistics, Stanford University · Department of Statistics and Data Science, University of Pennsylvania

  15. Towards Anytime-Valid Statistical Watermarking

    Feb 19, 2026Baihe Huang, Eric Xu, Kannan Ramchandran +2LLM WatermarkingAnytime-Valid Inference

    University of California, Berkeley.