Searcharxiv⌕ Search

arXiv subjects

Eric Aubinais

Publications and source records attributed to Eric Aubinais.

4 recordsLinked to original sources

Genesis Steering: Better Activation Steering Through Training Data

Activation steering controls large language models at inference time, but its effectiveness depends on the choice of reference data. We introduce genesis steering, which derives a genesis vector from contrastive examples in the model's own fine-tuning data. When these data are unavailable, we draw on the membership inference literature to identify likely training examples and construct a proxy dataset. We establish two theoretical results on optimal displacement to the decision boundary and reduced proxy-estimation bias. We evaluate genesis steering and membership-based proxy construction on Tülu-3-8B, whose fine-tuning data are publicly available, and Llama-3.1-8B-Instruct as a control. We find that safety data from Tülu-3-8B's training mixture yield stronger steering than external datasets on both models. On Tülu-3-8B, membership scores largely agree in identifying training examples, and vectors built from top-ranked examples achieve competitive performance against random subsets of the same size. In contrast, the results on Llama-3.1-8B-Instruct hint that a MIA method will only have a positive effect if the mean direction shift with MIA-filtered in-distribution data is much less severe than the direction shift with OoD data.

cs.LG↗

Concerns in Benchmarking Jailbreaks: An Empirical Study of Attack Success Rates

Jailbreak methods are commonly compared using Attack Success Rate (ASR), but these comparisons depend on how responses are generated and judged. In this paper, we examine eight attacks against three open-weight language models using three evaluation procedures. Low-resource language attacks achieve a mean ASR of 96.0% under keyword matching but 2.7% under a Qwen3 evaluator following StrongREJECT scoring instructions. Changing the number and arrangement of attempts reverses the ranking of DeepInception and rewriting. In a separate experiment, sampling 100 responses for each unchanged request at Ollama's default temperature of 0.8 yields 100% ASR under keyword matching but 50% under the Qwen3 evaluator. Beyond evaluator sensitivity, these findings raise a safety concern: repeated sampling can expose residual unsafe behavior despite an initial refusal, suggesting probabilistic suppression rather than robust exclusion. We recommend explicit success definitions, matched response budgets, validated evaluators, and reporting cumulative failure under repeated interaction rather than relying on single-generation tests.

cs.LG↗

Membership Inference Risks in Quantized Models: A Theoretical and Empirical Study

Quantizing machine learning models has demonstrated its effectiveness in lowering memory and inference costs while maintaining performance levels comparable to those of the original models. In this work, we investigate the impact of quantization procedures on privacy in data-driven models, focusing on their vulnerability to membership inference attacks. Membership Inference Security (MIS) has recently been proposed to characterize the privacy of machine learning models against the most powerful (and possibly unknown) attacks. However, quantifying MIS appears to be computationally very difficult. In this paper, we propose a new MIS indicator for post-training quantization procedures of machine learning models that minimizes an empirical loss. This new indicator is a byproduct of a theoretical asymptotic analysis of the MIS in this context. We also present a methodology for empirically estimating our MIS indicator. Using synthetic datasets and real-world data (in the context of drug discovery), we demonstrate the effectiveness of our approach in assessing and ranking the MIS of different quantizers.

stat.ML↗

Fundamental Limits of Membership Inference Attacks on Machine Learning Models

Membership inference attacks (MIA) can reveal whether a particular data point was part of the training dataset, potentially exposing sensitive information about individuals. This article provides theoretical guarantees by exploring the fundamental statistical limitations associated with MIAs on machine learning models at large. More precisely, we first derive the statistical quantity that governs the effectiveness and success of such attacks. We then theoretically prove that in a non-linear regression setting with overfitting learning procedures, attacks may have a high probability of success. Finally, we investigate several situations for which we provide bounds on this quantity of interest. Interestingly, our findings indicate that discretizing the data might enhance the learning procedure's security. Specifically, it is demonstrated to be limited by a constant, which quantifies the diversity of the underlying data distribution. We illustrate those results through simple simulations.

stat.ML↗