analiza aiqbench
This is a research paper proposing a method for removing specific concepts from text-to-image diffusion models by training sparse autoencoder latents with contrastive learning so that related concepts stop sharing features. The reported gains are on the UnlearnCanvas benchmark, which is a controlled testbed, so readers should check how well the results transfer to real-world models and to adversarial prompts before drawing broader conclusions.
Disentangled Sparse Representations for Concept-Separated Diffusion Unlearning
arXiv:2605.12122v2 Announce Type: replace Abstract: Unlearning specific concepts in text-to-image diffusion models has become increasingly important for preventing undesirable content generation. Among prior approaches, sparse autoencoder (SAE)-based methods have attracted attention due to their ability to suppress target concepts through lightweight manipulation of latent features, without modifying model parameters. However, SAEs trained primarily for sparse reconstruction do not explicitly enforce concept-wise separation, allowing semantically related concepts to share latent features. To address this, we propose SAEParate, which explicitly separates semantically overlapping concepts in the SAE latent space through concept-aware contrastive learning, supported by cross-attention-guided spatial aggregation that emphasizes concept-relevant regions. In addition, we enhance the encoder with a nonlinear transformation to increase its expressive capacity under this separation objective, enabling more concept-discriminative latent representations. Experiments on UnlearnCanvas demonstrate state-of-the-art performance, with particularly strong gains in joint style-object unlearning, a challenging setting where existing methods suffer from severe interference between target and non-target concepts.
Abstract:Unlearning specific concepts in text-to-image diffusion models has become increasingly important for preventing undesirable content generation. Among prior approaches, sparse autoencoder (SAE)-based methods have attracted attention due to their ability to suppress target concepts through lightweight manipulation of latent features, without modifying model parameters. However, SAEs trained primarily for sparse reconstruction do not explicitly enforce concept-wise separation, allowing semantically related concepts to share latent features. To address this, we propose SAEParate, which explicitly separates semantically overlapping concepts in the SAE latent space through concept-aware contrastive learning, supported by cross-attention-guided spatial aggregation that emphasizes concept-relevant regions. In addition, we enhance the encoder with a nonlinear transformation to increase its expressive capacity under this separation objective, enabling more concept-discriminative latent representations. Experiments on UnlearnCanvas demonstrate state-of-the-art performance, with particularly strong gains in joint style-object unlearning, a challenging setting where existing methods suffer from severe interference between target and non-target concepts.
53 pages, 25 figures. Accepted at NeurIPS 2026
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV)
arXiv:2605.12122 [cs.LG]
(or arXiv:2605.12122v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.12122 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 257 cuvinte