Concept Steerers: Leveraging K-Sparse Autoencoders for Test-Time Controllable Generations

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Kim, Dahye, Ghadiyaram, Deepti
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915547319369728
author Kim, Dahye
Ghadiyaram, Deepti
author_facet Kim, Dahye
Ghadiyaram, Deepti
contents Despite the remarkable progress in text-to-image generative models, they are prone to adversarial attacks and inadvertently generate unsafe, unethical content. Existing approaches often rely on fine-tuning models to remove specific concepts, which is computationally expensive, lacks scalability, and/or compromises generation quality. In this work, we propose a novel framework leveraging k-sparse autoencoders (k-SAEs) to enable efficient and interpretable concept manipulation in diffusion models. Specifically, we first identify interpretable monosemantic concepts in the latent space of text embeddings and leverage them to precisely steer the generation away or towards a given concept (e.g., nudity) or to introduce a new concept (e.g., photographic style) -- all during test time. Through extensive experiments, we demonstrate that our approach is very simple, requires no retraining of the base model nor LoRA adapters, does not compromise the generation quality, and is robust to adversarial prompt manipulations. Our method yields an improvement of $\mathbf{20.01\%}$ in unsafe concept removal, is effective in style manipulation, and is $\mathbf{\sim5}$x faster than the current state-of-the-art. Code is available at: https://github.com/kim-dahye/steerers
format Preprint
id arxiv_https___arxiv_org_abs_2501_19066
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Concept Steerers: Leveraging K-Sparse Autoencoders for Test-Time Controllable Generations
Kim, Dahye
Ghadiyaram, Deepti
Computer Vision and Pattern Recognition
Despite the remarkable progress in text-to-image generative models, they are prone to adversarial attacks and inadvertently generate unsafe, unethical content. Existing approaches often rely on fine-tuning models to remove specific concepts, which is computationally expensive, lacks scalability, and/or compromises generation quality. In this work, we propose a novel framework leveraging k-sparse autoencoders (k-SAEs) to enable efficient and interpretable concept manipulation in diffusion models. Specifically, we first identify interpretable monosemantic concepts in the latent space of text embeddings and leverage them to precisely steer the generation away or towards a given concept (e.g., nudity) or to introduce a new concept (e.g., photographic style) -- all during test time. Through extensive experiments, we demonstrate that our approach is very simple, requires no retraining of the base model nor LoRA adapters, does not compromise the generation quality, and is robust to adversarial prompt manipulations. Our method yields an improvement of $\mathbf{20.01\%}$ in unsafe concept removal, is effective in style manipulation, and is $\mathbf{\sim5}$x faster than the current state-of-the-art. Code is available at: https://github.com/kim-dahye/steerers
title Concept Steerers: Leveraging K-Sparse Autoencoders for Test-Time Controllable Generations
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2501.19066