Beyond Input Activations: Identifying Influential Latents by Gradient Sparse Autoencoders

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Shu, Dong, Wu, Xuansheng, Zhao, Haiyan, Du, Mengnan, Liu, Ninghao
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909801160638464
author Shu, Dong
Wu, Xuansheng
Zhao, Haiyan
Du, Mengnan
Liu, Ninghao
author_facet Shu, Dong
Wu, Xuansheng
Zhao, Haiyan
Du, Mengnan
Liu, Ninghao
contents Sparse Autoencoders (SAEs) have recently emerged as powerful tools for interpreting and steering the internal representations of large language models (LLMs). However, conventional approaches to analyzing SAEs typically rely solely on input-side activations, without considering the causal influence between each latent feature and the model's output. This work is built on two key hypotheses: (1) activated latents do not contribute equally to the construction of the model's output, and (2) only latents with high causal influence are effective for model steering. To validate these hypotheses, we propose Gradient Sparse Autoencoder (GradSAE), a simple yet effective method that identifies the most influential latents by incorporating output-side gradient information.
format Preprint
id arxiv_https___arxiv_org_abs_2505_08080
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Beyond Input Activations: Identifying Influential Latents by Gradient Sparse Autoencoders
Shu, Dong
Wu, Xuansheng
Zhao, Haiyan
Du, Mengnan
Liu, Ninghao
Machine Learning
Artificial Intelligence
Computation and Language
Sparse Autoencoders (SAEs) have recently emerged as powerful tools for interpreting and steering the internal representations of large language models (LLMs). However, conventional approaches to analyzing SAEs typically rely solely on input-side activations, without considering the causal influence between each latent feature and the model's output. This work is built on two key hypotheses: (1) activated latents do not contribute equally to the construction of the model's output, and (2) only latents with high causal influence are effective for model steering. To validate these hypotheses, we propose Gradient Sparse Autoencoder (GradSAE), a simple yet effective method that identifies the most influential latents by incorporating output-side gradient information.
title Beyond Input Activations: Identifying Influential Latents by Gradient Sparse Autoencoders
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2505.08080