LaRA: Layer-wise Representation Analysis for Detecting Data Contamination in RL Post-Training

2026-05-28

AI Summary

This work addresses the problem of data contamination in post-training reinforcement learning, which often degrades model generalization and distorts evaluation. To tackle this issue, the authors propose LaRA, a novel framework that, for the first time, leverages the geometric properties of layer-wise representations in language models. By analyzing model responses to controlled input perturbations, LaRA constructs three complementary metrics—perturbation sensitivity, directional collapse, and local rigidity—and introduces a cross-layer bias aggregation mechanism to detect contamination signals. Experimental results demonstrate that LaRA significantly outperforms existing methods based on output likelihood or entropy across multiple RL-finetuned models, establishing a new paradigm for assessing data quality in post-training stages.

Abstract

Reinforcement learning (RL) post-training has shown to improve reasoning in large language models (LLMs). However, there has been little exploration on the problem of data contamination in RL post-training, potentially undermining generalization and evaluation reliability of the training process itself. Existing detection methods primarily rely on output-level signals such as likelihood or entropy, which become unreliable for RL-trained models since RL shapes behavior through trajectory-level rewards rather than token likelihoods. We propose LaRA, a layer-wise representation analysis framework for detecting contamination in RL post-trained LLMs. LaRA introduces three complementary metrics, measuring perturbation sensitivity, directional collapse, and local representation rigidity under controlled perturbations. We find that contamination produces progressive geometric deviations across layers, including amplified perturbation sensitivity, stronger directional collapse, and enhanced local rigidity. Based on our findings, we also develop a contamination detection protocol that aggregates representation-level deviations across layers and metrics. Experiments on RL-trained reasoning models show that our protocol outperforms existing output-level baselines for contamination detection.

Problem

• data contamination
• reinforcement learning post-training
• large language models
• generalization
• evaluation reliability

Innovation

• layer-wise representation analysis
• data contamination detection
• reinforcement learning post-training
• representation rigidity
• directional collapse