LaRA: Layer-wise Representation Analysis for Detecting Data Contamination in RL Post-Training
cs.LG, cs.AI
Submitted: 2026-05-28
Updated: 2026-08-31
Comments: EMNLP 2026 Findings
License: http://creativecommons.org/licenses/by/4.0/
The gist: Reinforcement learning (RL) post-training has shown to improve reasoning in large language models (LLMs).
Terminology
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.
Sources
- Process Reinforcement through Implicit Rewards
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- Revisiting the Uniform Information Density Hypothesis in LLM Reasoning
- Tracing the Representation Geometry of Language Models from Pretraining to Post-training
- LIMR: Less is More for RL Scaling
- Olmo 3
- Embracing Anisotropy: Turning Massive Activations into Interpretable Control Knobs for Large Language Models
- Gap-K%: Measuring Top-1 Prediction Gap for Detecting Pretraining Data
- HybridFlow: A Flexible and Efficient RLHF Framework
- Training-free LLM Verification via Recycling Few-shot Examples
- Steering Language Models With Activation Engineering
- Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks