# Amplifying Membership Signal Through Chained Regeneration Source: [Amplifying Membership Signal Through Chained Regeneration]() Collection index: [llms.txt]() Data updated: 2026-07-02T05:00:13.060Z Authors: Wojciech {\\L}apacz, Stanis{\\l}aw Pawlak Published: 2026-06-30T17:29:04.000Z ## Abstract The tendency of large generative models to memorize training data makes sample verification critical for privacy auditing and copyright enforcement. Current membership (MIA) and dataset inference (DI) attacks often rely on one-shot generations, which yield weak signals and limited sensitivity across modalities. Inspired by Model Autophagy Disorder (MAD), we introduce MADreMIA, a model-agnostic framework that enhances white-, gray-, and black-box MIA and DI. Rather than relying on shadow model training -- often infeasible for large generative models -- our framework facilitates scalable inference by leveraging inherent signals through iterative trajectories. This process utilizes chained generations across diverse modalities, where each output serves as the subsequent input, to improve membership evidence at low FPR. We demonstrate that memorized training samples exhibit significantly higher coherence and slower degradation during iterative regeneration than non-member generations. Our results show that MADreMIA provides richer signals across diverse model families and modalities; we present comprehensive evaluations for IARs, diffusion, and language models, alongside preliminary results demonstrating its potential for audio models. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Membership Inference Attack on Pythia 6.9B](): Membership Inference Attack; Pythia 6.9B; TPR@1%FPR: 13.1. - [Membership Inference Attack on OLMo-7B](): Membership Inference Attack; Olmo-7B; TPR@1%FPR: 55.3. - [Membership Inference Attack on Llama-13B](): Membership Inference Attack; LLaMA-13B; TPR@1%FPR: 21. - [Membership Inference Attack on VAR-d30](): Membership Inference Attack; VAR-d30; TPR@1%FPR: 9. - [Membership Inference Attack on RAR-XXL](): Membership Inference Attack; RAR-XXL; TPR@1%FPR: 8.4. - [Membership Inference Attack on OPT-6.7B](): Membership Inference Attack; OPT-6.7B; TPR @ 1% FPR: 12.1. ## Original sources and implementation [Original paper]() [PDF]()