English
Related papers

Related papers: Interpretability as Compression: Reconsidering SAE…

200 papers

Sparse autoencoders (SAEs) extract millions of interpretable features from a language model, but flat feature inventories aren't very useful on their own. Domain concepts get mixed with generic and weakly grounded features, while related…

Artificial Intelligence · Computer Science 2026-04-29 John Winnicki , Abeynaya Gnanasekaran , Eric Darve

Sparse autoencoders (SAEs) improve interpretability in multimodal models, but it remains unclear whether SAE features form modular, composable units for reasoning-an assumption underlying many intervention-based steering methods. We test…

Artificial Intelligence · Computer Science 2026-03-27 Yunpeng Zhou

Sparse autoencoders (SAEs) are widely used to extract human-interpretable features from neural network activations, but their learned features can vary substantially across random seeds and training choices. To improve stability, we studied…

Machine Learning · Statistics 2026-03-05 Piotr Jedryszek , Oliver M. Crook

Recent interpretability work on large language models (LLMs) has been increasingly dominated by a feature-discovery approach with the help of proxy modules. Then, the quality of features learned by, e.g., sparse auto-encoders (SAEs), is…

Machine Learning · Computer Science 2025-10-28 Mengyu Ye , Jun Suzuki , Tatsuro Inaba , Tatsuki Kuribayashi

Sparse autoencoders (SAEs) are increasingly used for safety-relevant applications including alignment detection and model steering. These use cases require SAE latents to be as atomic as possible. Each latent should represent a single…

Machine Learning · Computer Science 2026-04-07 Matthew Levinson

Adapting foundation models for specific purposes has become a standard approach to build machine learning systems for downstream applications. Yet, it is an open question which mechanisms take place during adaptation. Here we develop a new…

Computer Vision and Pattern Recognition · Computer Science 2025-03-24 Hyesu Lim , Jinho Choi , Jaegul Choo , Steffen Schneider

The Universality Hypothesis in large language models (LLMs) claims that different models converge towards similar concept representations in their latent spaces. Providing evidence for this hypothesis would enable researchers to exploit…

Machine Learning · Computer Science 2025-05-22 Michael Lan , Philip Torr , Austin Meek , Ashkan Khakzar , David Krueger , Fazl Barez

The rapid advancements in transformer-based language models have revolutionized natural language processing, yet understanding the internal mechanisms of these models remains a significant challenge. This paper explores the application of…

Machine Learning · Computer Science 2025-02-14 Edith Natalia Villegas Garcia , Alessio Ansuini

Previous studies primarily utilize MLP neurons as units of analysis for understanding the mechanisms of factual knowledge in Language Models (LMs); however, neurons suffer from polysemanticity, leading to limited knowledge expression and…

Computation and Language · Computer Science 2025-02-28 Yuheng Chen , Pengfei Cao , Kang Liu , Jun Zhao

Interpretability can improve the safety, transparency and trust of AI models, which is especially important in healthcare applications where decisions often carry significant consequences. Mechanistic interpretability, particularly through…

Sparse autoencoders (SAEs) are a popular method for decomposing Large Langage Models (LLM) activations into interpretable latents. However, due to their substantial training cost, most academic research uses open-source SAEs which are only…

Machine Learning · Computer Science 2025-06-13 Patrick Leask , Neel Nanda , Noura Al Moubayed

Mechanistic interpretability aims to explain neural model behaviour by reverse-engineering learned computational structure into human-understandable components. Without a formal framework, however, mechanistic explanations cannot be…

Machine Learning · Computer Science 2026-05-12 Ward Gauderis , Thomas Dooms , Steven T. Holmer , Kola Ayonrinde , Geraint A. Wiggins

Variational Autoencoders (VAEs) are powerful generative models for learning latent representations. Standard VAEs generate dispersed and unstructured latent spaces by utilizing all dimensions, which limits their interpretability, especially…

Computer Vision and Pattern Recognition · Computer Science 2025-05-21 Farshad Sangari Abiz , Reshad Hosseini , Babak N. Araabi

Current sparse autoencoder (SAE) approaches to neural network interpretability assume that activations can be decomposed through linear superposition into sparse, interpretable features. Despite high reconstruction fidelity, SAEs…

Neurons and Cognition · Quantitative Biology 2025-12-10 Omar Claflin

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…

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

Understanding how information is represented in neural networks is a fundamental challenge in both neuroscience and artificial intelligence. Despite their nonlinear architectures, recent evidence suggests that neural networks encode…

Machine Learning · Computer Science 2025-03-04 David Klindt , Charles O'Neill , Patrik Reizinger , Harald Maurer , Nina Miolane

Sparse autoencoders (SAEs) are used to disentangle the dense, polysemantic internal representations of large language models (LLMs) into interpretable, monosemantic concepts. However, standard $\ell_1$-regularized SAEs suffer from feature…

Machine Learning · Computer Science 2026-05-08 Faris Chaudhry , Keisuke Yano , Anthea Monod

Neurons in large language models often exhibit \emph{polysemanticity}, simultaneously encoding multiple unrelated concepts and obscuring interpretability. Instead of relying on post-hoc methods, we present \textbf{MoE-X}, a…

Vision-language models encode images and text in a joint space, minimizing the distance between corresponding image and text pairs. How are language and images organized in this joint space, and how do the models encode meaning and…

Computer Vision and Pattern Recognition · Computer Science 2025-08-25 Isabel Papadimitriou , Huangyuan Su , Thomas Fel , Sham Kakade , Stephanie Gil

EEG foundation models achieve state-of-the-art clinical performance, yet the internal computations driving their predictions remain opaque: a barrier to clinical trust. We apply TopK Sparse Autoencoders (SAEs) across three architecturally…