SERSEM: Selective Entropy-Weighted Scoring for Membership Inference in Code Language Models

TL;DR

SERSEM is a novel white-box attack framework that enhances membership inference in code language models by focusing on human-centric coding anomalies using static analysis and internal activations.

Contribution

It introduces SERSEM, a new method combining static AST analysis and transformer activation pooling to improve membership inference attacks on code LLMs.

Findings

SERSEM achieves an AUC-ROC of 0.7913 on StarCoder2-3B.

SERSEM outperforms probability-based baselines in membership inference.

Focusing on coding anomalies yields more robust memorization indicators.

Abstract

As Large Language Models (LLMs) for code increasingly utilize massive, often non-permissively licensed datasets, evaluating data contamination through Membership Inference Attacks (MIAs) has become critical. We propose SERSEM (Selective Entropy-Weighted Scoring for Membership Inference), a novel white-box attack framework that suppresses uninformative syntactical boilerplate to amplify specific memorization signals. SERSEM utilizes a dual-signal methodology: first, a continuous character-level weight mask is derived through static Abstract Syntax Tree (AST) analysis, spellchecking-based multilingual logic detection, and offline linting. Second, these heuristic weights are used to pool internal transformer activations and calibrate token-level Z-scores from the output logits. Evaluated on a 25,000-sample balanced dataset, SERSEM achieves a global AUC-ROC of 0.7913 on the StarCoder2-3B model and 0.7867 on the StarCoder2-7B model, consistently outperforming the implemented probability-based baselines Loss, Min-K% Prob, and PAC. Our findings demonstrate that focusing on human-centric coding anomalies provides a significantly more robust indicator of verbatim memorization than sequence-level probability averages.