Data Contamination

Ask about this topic

Overview

Data contamination covers both deliberately or accidentally corrupted examples and overlap between training data and evaluation sets, which can make models seem more accurate or reliable than they are. Researchers develop robust methods such as Huber-style estimation, trimming, and conformal prediction, while auditing tools look for memorization, suspicious token patterns, or unusually fast adaptation. Benchmarks increasingly use spatially, temporally, or source-disjoint splits to test whether results generalize beyond familiar examples and locations. A central challenge is that contamination can be hard to detect, and removing exact test-set overlap may still leave models familiar with the test domain; stronger audits can also be costly or computationally limited.

Latest papers

CardsList

33 papers3 papers in the last four weeks, down 40% on the four weeks before. 0.0% of all new papers.

  1. Adaptive Robust Confidence Sets in High Dimensions

    Oct 11, 2026Ilias Diakonikolas, Chao Gao, Daniel M. Kane +1Robust Mean EstimationLow-Degree Polynomial Method

    University of Wisconsin-Madison · University of Chicago · University of California, San Diego +1

  2. Split Conformal Interval Length under Huber Contamination

    Oct 11, 2026Xiao Zhang, Bowen Zhang, Lie HeConformal PredictionRegression Prediction Intervals

    Shanghai University of Finance and Economics · MoE Key Laboratory of Interdisciplinary Research of Computation and Economics, Shanghai University of Finance and Economics

  3. Data Leakage in Patch-Based Hyperspectral Image Classification: Quantifying the Impact of Spatial Overlap

    Oct 6, 2026Mohammed Q. AlkhatibData ContaminationHyperspectral Image Classification

    College of Engineering and IT, University of Dubai, Dubai, 14143, UAE

  4. Prevalence Determines Precision:Silent Contamination in Detector-Defined Datasets

    Sep 10, 2026Jia Huang, Yankai Wan, Yangjun OuBenchmark ValidityData Contamination

    Guanghua School of Management, Peking University · College of Artificial Intelligence, Jilin University · School of Mathematical Sciences, Peking University

  5. A Later Test Set Is Not a New Domain: Pretraining Familiarity Survives a Contamination-Free Hold-Out

    Sep 9, 2026Mahdi Naser Moghadasi, Faezeh GhaderiZero-Shot Time Series ForecastingDomain Generalization

    BrightMind AI · University of Texas at Arlington

  6. When Auditors Fabricate: Batch-Size Degradation and Confident Hallucination in LLM Detection of Planted Document Contamination

    Sep 9, 2026Karan Parekh, Sanjana Pendyala Ravinder, Sana Mhapsekar +1LLM AuditingData Contamination

    University of North Texas

  7. Forward-Facing Near-Infrared Adds Little to Colour for Farm-Machinery Traversability: A Site-Disjoint Evaluation of Sensor-Dependent Spatial Leakage

    Sep 7, 2026Sungwoo KangAgricultural Image AnalysisData Contamination

    Department of Electrical and Computer Engineering, Korea University, Seoul, Republic of Korea

  8. Ten Architectures, One Error: Shared Failure Modes in Hyperspectral Classification under Spatially Disjoint Evaluation

    Sep 1, 2026Ehsan Faghih, Fatemeh Ashrafi, Marguerite Moore +1Data ContaminationRemote Sensing Benchmarks

    North Carolina State University, Wilson College of Textiles Raleigh, NC, USA

  9. Robustness of Anomaly Detection Models for Industrial Control Systems under Training-Time Data Contamination

    Aug 24, 2026Mustafa Umut Ozbek, Taiwo Ojo, Pooria Madani +2Industrial Control System SecurityAnomaly Detection

    Ontario Tech University, Oshawa, Ontario, Canada

  10. What Memory Composition Does Not Tell Us About Anomaly Detection

    Aug 24, 2026Joongwon Chae, Runming Wang, Peiwu QinAnomaly DetectionData Contamination

    Institute of Biopharmaceutical and Health Engineering, Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen, China · RatelSoft · Guangdong Provincial Laboratory of Traditional Chinese Medicine, Hengqin, Guangdong, China

  11. Auditing Chinese Web-scale Corpora via Sampled BPE Token Statistics

    Aug 11, 2026Qingjie Zhang, Ziqi Tang, Jie Zhang +7LLM AuditingData Contamination

    Tsinghua University · SiliconProspect AI · Nanyang Technological University +1

  12. Conformal Changepoint Localization and Root Cause Analysis with Corrupted Observations

    Jul 29, 2026Seunghun Yu, Meiyi Zhu, Petar Popovski +2Conformal PredictionChange-Point Detection

    Department of Electrical Engineering, Korea Advanced Institute of Science and Technology, Daejeon 34141, South Korea · Department of Engineering, King’s College London, WC2R 2LS, London, U.K. · Department of Electronic Systems, Aalborg University, 9220 Aalborg, Denmark +2

  13. Adversarial Contamination Meets Hard Thresholding: An Iterative Algorithm with Signal Adaptivity and Minimax Optimality

    Jun 26, 2026Shixiang Liu, Hanming YangRobust RegressionSparse Recovery

    School of Statistics, Renmin University of China · Institute of Statistics and Big Data, Renmin University of China

  14. Computationally tractable robust differentially private mean estimation

    Jun 10, 2026Kelly RamsayRobust Mean EstimationDifferential Privacy

    Department of Mathematics and Statistics York University Toronto, ON, Canada

  15. Ethical and Technical Limits of Deepfake Speech Datasets

    Jun 9, 2026Vojtěch Staněk, Eva Trnovská, Kamil Malinka +1Data LeakageAudio Deepfake Detection

  16. Log-Ratio Propagation on the Simplex: A Theory of Cellwise Contamination for Compositional Data

    May 29, 2026Matthias TemplData Contamination

    School of Business, FHNW Fachhochschule Nordwestschweiz, Riggenbachstrasse 16, 4600 Olten, Switzerland

  17. No Safe Dose: How Training Data Drives Unsafe Image Generation

    May 27, 2026Felix Friedrich, Lukas Helff, Niharika Hegde +2Safe T2I GenerationData Contamination

    1Black Forest Labs · 2TU Darmstadt & hessian.AI · 3DFKI +1

  18. CITYREP: A Unified Benchmark for Urban Representations Across Cities, Tasks, and Modalities

    May 25, 2026Junyuan Liu, Xinglei Wang, Zichao Zeng +5Representation LearningGeospatial Representation Learning

    SpaceTimeLab, University College London, UK · 3DIMPact, University College London, UK · School of Resource and Environmental Sciences, Wuhan University, China +1

  19. TSFMAudit: Data Contamination Auditing in Forecasting Time Series Foundation Models

    May 24, 2026Hongkai Li, Shifeng Xie, Lefei Shen +7Data ContaminationTime-Series Foundation Models

    Zhejiang University · Télécom Paris · State Street Technology (Zhejiang) Ltd. +1

  20. Spiking the training data to correct for test set contamination

    May 24, 2026Johnny Tian-Zheng Wei, Jerry Li, Ameya Godbole +1Post-Hoc CalibrationBenchmark Contamination

    University of Southern California

  21. On the Sample Complexity of Robust Binary Hypothesis Testing

    May 23, 2026Shankar Vallinayagam, Ankit Pensia, Varun JogSample ComplexityData Contamination

    Department of Pure Mathematics and Mathematical Statistics, University of Cambridge · Department of Statistics, Carnegie Mellon University

  22. Physics-R1: An Audited Olympiad Corpus and Recipe for Visual Physics Reasoning

    May 13, 2026Shan YangVisual ReasoningRL for Language Model Reasoning

    Independent Researcher

  23. The frame-level leakage trap: rethinking evaluation protocols for intrinsic image decomposition, with source-separable uncertainty as a case study

    May 7, 2026Jihwan WooData LeakageIntrinsic Image Decomposition

    Amazon Web Services (AWS), Partner Field APJ, Sr. Specialist Solutions Architect, AI/ML

  24. When Does Trimming Help Conformal Prediction? A Retained-Law Diagnostic under Calibration Contamination

    May 7, 2026Congye WangConformal PredictionData Contamination

    Newcastle University

  25. SHIFT: Robust Double Machine Learning for Average Dose-Response Functions under Heavy-Tailed Contamination

    Apr 30, 2026Eichi UeharaDouble MLRobust Regression

    Independent researcher.

  26. High-Dimensional Gaussian Mean Estimation under Realizable Contamination

    Mar 17, 2026Ilias Diakonikolas, Daniel M. Kane, Thanasis PittasData ContaminationStatistical Learning Theory

    University of Wisconsin-Madison · University of California, San Diego

  27. AI-Enhanced Spatial Cellular Traffic Demand Prediction with Contextual Clustering and Error Correction for 5G/6G Planning

    Mar 11, 2026Mohamad Alkadamani, Colin Brown, Halim YanikomerogluTraffic ForecastingWireless Communications

    Innovation, Science and Economic Development Canada (ISED) · Carleton University · Communications Research Centre (CRC)

  28. The Conundrum of Trustworthy Research on Attacking Personally Identifiable Information Removal Techniques

    Mar 9, 2026Sebastian Ochs, Ivan HabernalData LeakagePrivacy Auditing

    Trustworthy Human Language Technologies www.trusthlt.org · Technical University of Darmstadt, Germany · Research Center for Trustworthy Data Science and Security, Ruhr University Bochum, Germany

  29. No Memorization, No Detection: Output Distribution-Based Contamination Detection in Small Language Models

    Mar 3, 2026Omer SelaData Contamination in Language ModelsData Contamination

    Tel Aviv University

  30. Sample Complexity Bounds for Robust Mean Estimation with Mean-Shift Contamination

    Feb 25, 2026Ilias Diakonikolas, Giannis Iakovidis, Daniel M. Kane +1Robust Mean EstimationRobust Estimation

    UW Madison · UC San Diego