Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Pach, Mateusz, Karthik, Shyamgopal, Bouniot, Quentin, Belongie, Serge, Akata, Zeynep
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911291098005504
author Pach, Mateusz
Karthik, Shyamgopal
Bouniot, Quentin
Belongie, Serge
Akata, Zeynep
author_facet Pach, Mateusz
Karthik, Shyamgopal
Bouniot, Quentin
Belongie, Serge
Akata, Zeynep
contents Sparse Autoencoders (SAEs) have recently gained attention as a means to improve the interpretability and steerability of Large Language Models (LLMs), both of which are essential for AI safety. In this work, we extend the application of SAEs to Vision-Language Models (VLMs), such as CLIP, and introduce a comprehensive framework for evaluating monosemanticity at the neuron-level in visual representations. To ensure that our evaluation aligns with human perception, we propose a benchmark derived from a large-scale user study. Our experimental results reveal that SAEs trained on VLMs significantly enhance the monosemanticity of individual neurons, with sparsity and wide latents being the most influential factors. Further, we demonstrate that applying SAE interventions on CLIP's vision encoder directly steers multimodal LLM outputs (e.g., LLaVA), without any modifications to the underlying language model. These findings emphasize the practicality and efficacy of SAEs as an unsupervised tool for enhancing both interpretability and control of VLMs. Code and benchmark data are available at https://github.com/ExplainableML/sae-for-vlm.
format Preprint
id arxiv_https___arxiv_org_abs_2504_02821
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models
Pach, Mateusz
Karthik, Shyamgopal
Bouniot, Quentin
Belongie, Serge
Akata, Zeynep
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Sparse Autoencoders (SAEs) have recently gained attention as a means to improve the interpretability and steerability of Large Language Models (LLMs), both of which are essential for AI safety. In this work, we extend the application of SAEs to Vision-Language Models (VLMs), such as CLIP, and introduce a comprehensive framework for evaluating monosemanticity at the neuron-level in visual representations. To ensure that our evaluation aligns with human perception, we propose a benchmark derived from a large-scale user study. Our experimental results reveal that SAEs trained on VLMs significantly enhance the monosemanticity of individual neurons, with sparsity and wide latents being the most influential factors. Further, we demonstrate that applying SAE interventions on CLIP's vision encoder directly steers multimodal LLM outputs (e.g., LLaVA), without any modifications to the underlying language model. These findings emphasize the practicality and efficacy of SAEs as an unsupervised tool for enhancing both interpretability and control of VLMs. Code and benchmark data are available at https://github.com/ExplainableML/sae-for-vlm.
title Sparse Autoencoders Learn Monosemantic Features in Vision-Language Models
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2504.02821