English
Related papers

Related papers: Sparse but Wrong: Incorrect L0 Leads to Incorrect …

200 papers

Sparse autoencoders (SAEs) are one of the main methods to interpret the inner workings of deep neural networks (DNNs), decomposing activations into higher-dimensional features. However, they exhibit critical shortcomings where a large…

Machine Learning · Computer Science 2026-05-19 Michał Brzozowski , Neo Christopher Chung

Recent developments in Large Language Model (LLM) capabilities have brought great potential but also posed new risks. For example, LLMs with knowledge of bioweapons, advanced chemistry, or cyberattacks could cause violence if placed in the…

Machine Learning · Computer Science 2025-03-17 Matthew Khoriaty , Andrii Shportko , Gustavo Mercier , Zach Wood-Doughty

Sparse Autoencoders (SAEs) have been proposed as an unsupervised approach to learn a decomposition of a model's latent space. This enables useful applications such as steering - influencing the output of a model towards a desired concept -…

Machine Learning · Computer Science 2025-12-23 Dana Arad , Aaron Mueller , Yonatan Belinkov

We study how reliably sparse autoencoders (SAEs) support claims about reasoning-related internal features in large language models. We first give a stylized analysis showing that sparsity-regularized decoding can preferentially retain…

Machine Learning · Computer Science 2026-05-19 George Ma , Zhongyuan Liang , Irene Y. Chen , Somayeh Sojoudi

Large language models (LLMs) excel at handling human queries, but they can occasionally generate flawed or unexpected responses. Understanding their internal states is crucial for understanding their successes, diagnosing their failures,…

Computation and Language · Computer Science 2025-02-24 Xuansheng Wu , Jiayi Yuan , Wenlin Yao , Xiaoming Zhai , Ninghao Liu

Sparse autoencoders (SAEs) have recently emerged as a powerful tool for language model steering. Prior work has explored top-k SAE latents for steering, but we observe that many dimensions among the top-k latents capture non-semantic…

Computation and Language · Computer Science 2025-10-03 Jiaqing Xie

Sparse autoencoders (SAEs) decompose large language model (LLM) activations into latent features that reveal mechanistic structure. Conventional SAEs train on broad data distributions, forcing a fixed latent budget to capture only…

Machine Learning · Computer Science 2025-08-14 Charles O'Neill , Mudith Jayasekara , Max Kirkby

Sparse autoencoders (SAEs) have become an important tool for analyzing and interpreting the activation space of transformer-based language models (LMs). However, SAEs suffer several shortcomings that diminish their utility and internal…

Computation and Language · Computer Science 2025-06-27 Ryosuke Takahashi , Tatsuro Inaba , Kentaro Inui , Benjamin Heinzerling

Understanding the multilingual mechanisms of large language models (LLMs) provides insight into how they process different languages, yet this remains challenging. Existing studies often focus on individual neurons, but their polysemantic…

Sparse autoencoders (SAEs) are widely used in mechanistic interpretability to project LLM activations onto sparse latent spaces. However, sparsity alone is an imperfect proxy for interpretability, and current training objectives often…

Machine Learning · Computer Science 2026-04-09 Vivek Narayanaswamy , Kowshik Thopalli , Bhavya Kailkhura , Wesam Sakla

Sparse autoencoders (SAEs) are a promising unsupervised approach for identifying causally relevant and interpretable linear features in a language model's (LM) activations. To be useful for downstream tasks, SAEs need to decompose LM…

Machine Learning · Computer Science 2024-08-02 Senthooran Rajamanoharan , Tom Lieberum , Nicolas Sonnerat , Arthur Conmy , Vikrant Varma , János Kramár , Neel Nanda

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…

Computer Vision and Pattern Recognition · Computer Science 2025-12-01 Mateusz Pach , Shyamgopal Karthik , Quentin Bouniot , Serge Belongie , Zeynep Akata

Sparse Autoencoders (SAEs) have become an important tool in mechanistic interpretability, helping to analyze internal representations in both Large Language Models (LLMs) and Vision Transformers (ViTs). By decomposing polysemantic…

Machine Learning · Computer Science 2026-05-11 Jakub Stępień , Marcin Mazur , Jacek Tabor , Przemysław Spurek

Recent LLMs like DeepSeek-R1 have demonstrated state-of-the-art performance by integrating deep thinking and complex reasoning during generation. However, the internal mechanisms behind these reasoning processes remain unexplored. We…

Computation and Language · Computer Science 2025-08-07 Andrey Galichin , Alexey Dontsov , Polina Druzhinina , Anton Razzhigaev , Oleg Y. Rogov , Elena Tutubalina , Ivan Oseledets

Sparse autoencoders (SAEs) have emerged as powerful techniques for interpretability of large language models (LLMs), aiming to decompose hidden states into meaningful semantic features. While several SAE variants have been proposed, there…

Machine Learning · Computer Science 2025-10-03 Xudong Zhu , Mohammad Mahdi Khalili , Zhihui Zhu

Large Language Models (LLMs) have transformed natural language processing, yet their internal mechanisms remain largely opaque. Recently, mechanistic interpretability has attracted significant attention from the research community as a…

Machine Learning · Computer Science 2025-09-24 Dong Shu , Xuansheng Wu , Haiyan Zhao , Daking Rai , Ziyu Yao , Ninghao Liu , Mengnan Du

Sparse autoencoders (SAEs) are a technique for sparse decomposition of neural network activations into human-interpretable features. However, current SAEs suffer from feature absorption, where specialized features capture instances of…

Machine Learning · Computer Science 2025-09-29 Anton Korznikov , Andrey Galichin , Alexey Dontsov , Oleg Rogov , Elena Tutubalina , Ivan Oseledets

Sparse autoencoders (SAEs) have been applied to large language models and protein language models, but not systematically to electronic health record (EHR) foundation models. We train TopK SAEs on FlatASCEND, a 14.5-million-parameter…

Machine Learning · Computer Science 2026-05-07 Chris Sainsbury , Feng Dong , Andreas Karwath

A common goal of mechanistic interpretability is to decompose the activations of neural networks into features: interpretable properties of the input computed by the model. Sparse autoencoders (SAEs) are a popular method for finding these…

Machine Learning · Computer Science 2025-02-10 Patrick Leask , Bart Bussmann , Michael Pearce , Joseph Bloom , Curt Tigges , Noura Al Moubayed , Lee Sharkey , Neel Nanda

Sparse autoencoders (SAEs) are a recent technique for decomposing neural network activations into human-interpretable features. However, in order for SAEs to identify all features represented in frontier models, it will be necessary to…

Machine Learning · Computer Science 2025-06-04 Anish Mudide , Joshua Engels , Eric J. Michaud , Max Tegmark , Christian Schroeder de Witt