No Caption, No Problem: Caption-Free Membership Inference via Model-Fitted Embeddings

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Jeon, Joonsung, Kim, Woo Jae, Ha, Suhyeon, Son, Sooel, Yoon, Sung-Eui
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914352546709504
author Jeon, Joonsung
Kim, Woo Jae
Ha, Suhyeon
Son, Sooel
Yoon, Sung-Eui
author_facet Jeon, Joonsung
Kim, Woo Jae
Ha, Suhyeon
Son, Sooel
Yoon, Sung-Eui
contents Latent diffusion models have achieved remarkable success in high-fidelity text-to-image generation, but their tendency to memorize training data raises critical privacy and intellectual property concerns. Membership inference attacks (MIAs) provide a principled way to audit such memorization by determining whether a given sample was included in training. However, existing approaches assume access to ground-truth captions. This assumption fails in realistic scenarios where only images are available and their textual annotations remain undisclosed, rendering prior methods ineffective when substituted with vision-language model (VLM) captions. In this work, we propose MoFit, a caption-free MIA framework that constructs synthetic conditioning inputs that are explicitly overfitted to the target model's generative manifold. Given a query image, MoFit proceeds in two stages: (i) model-fitted surrogate optimization, where a perturbation applied to the image is optimized to construct a surrogate in regions of the model's unconditional prior learned from member samples, and (ii) surrogate-driven embedding extraction, where a model-fitted embedding is derived from the surrogate and then used as a mismatched condition for the query image. This embedding amplifies conditional loss responses for member samples while leaving hold-outs relatively less affected, thereby enhancing separability in the absence of ground-truth captions. Our comprehensive experiments across multiple datasets and diffusion models demonstrate that MoFit consistently outperforms prior VLM-conditioned baselines and achieves performance competitive with caption-dependent methods.
format Preprint
id arxiv_https___arxiv_org_abs_2602_22689
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle No Caption, No Problem: Caption-Free Membership Inference via Model-Fitted Embeddings
Jeon, Joonsung
Kim, Woo Jae
Ha, Suhyeon
Son, Sooel
Yoon, Sung-Eui
Computer Vision and Pattern Recognition
Cryptography and Security
Latent diffusion models have achieved remarkable success in high-fidelity text-to-image generation, but their tendency to memorize training data raises critical privacy and intellectual property concerns. Membership inference attacks (MIAs) provide a principled way to audit such memorization by determining whether a given sample was included in training. However, existing approaches assume access to ground-truth captions. This assumption fails in realistic scenarios where only images are available and their textual annotations remain undisclosed, rendering prior methods ineffective when substituted with vision-language model (VLM) captions. In this work, we propose MoFit, a caption-free MIA framework that constructs synthetic conditioning inputs that are explicitly overfitted to the target model's generative manifold. Given a query image, MoFit proceeds in two stages: (i) model-fitted surrogate optimization, where a perturbation applied to the image is optimized to construct a surrogate in regions of the model's unconditional prior learned from member samples, and (ii) surrogate-driven embedding extraction, where a model-fitted embedding is derived from the surrogate and then used as a mismatched condition for the query image. This embedding amplifies conditional loss responses for member samples while leaving hold-outs relatively less affected, thereby enhancing separability in the absence of ground-truth captions. Our comprehensive experiments across multiple datasets and diffusion models demonstrate that MoFit consistently outperforms prior VLM-conditioned baselines and achieves performance competitive with caption-dependent methods.
title No Caption, No Problem: Caption-Free Membership Inference via Model-Fitted Embeddings
topic Computer Vision and Pattern Recognition
Cryptography and Security
url https://arxiv.org/abs/2602.22689