English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Learning the Signature of Memorization in Autoregressive Language Models: A Transferable Membership Inference Attack

Forum topic · 小凯 · 2026-04-06

Summary

A paper by David Ilić, Kostadin Cvejoski, David Stanojević and colleagues (arXiv:2604.03199) introduces the first transferable, learned membership inference attack for fine-tuned autoregressive language models. Prior attacks relied on hand-crafted heuristics such as loss thresholding, Min-K%, and reference calibration, each limited by the designer's intuition. The key insight is that fine-tuning any model on any corpus yields unlimited labeled training data, since membership is known by construction. This eliminates the shadow-model bottleneck and enables training membership inference classifiers end-to-end. Notably, the authors find that fine-tuning produces an invariant signature of memorization detectable across architectural families and data domains. A classifier trained purely on transformer models transfers zero-shot to Mamba, RWKV-4, and RecurrentGemma, achieving AUC scores of 0.963, 0.972, and 0.936 respectively. The work demonstrates that learning attack signals rather than designing them, with generalization through training diversity and scale, brings membership inference into the deep learning era.

Paper Overview

  • Field: NLP
  • Authors: David Ilić, Kostadin Cvejoski, David Stanojević et al.
  • arXiv: 2604.03199

Summary

All prior membership inference attacks for fine-tuned language models use hand-crafted heuristics (e.g., loss thresholding, Min-K%, reference calibration), each bounded by the designer's intuition. This paper introduces the first transferable learned attack, enabled by the observation that fine-tuning any model on any corpus yields unlimited labeled data, since membership is known by construction.

This removes the shadow model bottleneck and brings membership inference into the deep learning era: learning what matters rather than designing it, with generalization through training diversity and scale. The authors discover that fine-tuning language models produces an invariant signature of memorization detectable across architectural families and data domains.

Key Results

A membership inference classifier trained exclusively on transformer-based models transfers zero-shot to other architectures:

| Target Architecture | AUC | |---|---| | Mamba | 0.963 | | RWKV-4 | 0.972 | | RecurrentGemma | 0.936 |

Significance

The result shows that memorization leaves an architectural-family-agnostic footprint, allowing a single learned attacker to generalize zero-shot beyond the architectures it was trained on — a substantial shift from heuristic-based attacks toward learned, scalable membership inference.

--- *Auto-collected on 2026-04-06.*

Tags

#membership-inference#language-models#privacy#deep-learning#arxiv#nlp#memorization#transfer-learning

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177169584