中文
相关论文

相关论文: Interpretability as Compression: Reconsidering SAE…

200 篇论文

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…

机器学习 · 计算机科学 2026-05-19 Michał Brzozowski , Neo Christopher Chung

Sparse autoencoders (SAEs) enable feature-level mechanistic interpretability and activation steering in large language models (LLMs), but SAE-based language control remains unreliable in multilingual settings: most SAEs are trained on…

Radiological services are experiencing unprecedented demand, leading to increased interest in automating radiology report generation. Existing Vision-Language Models (VLMs) suffer from hallucinations, lack interpretability, and require…

计算机视觉与模式识别 · 计算机科学 2024-10-07 Ahmed Abdulaal , Hugo Fry , Nina Montaña-Brown , Ayodeji Ijishakin , Jack Gao , Stephanie Hyland , Daniel C. Alexander , Daniel C. Castro

Sparse autoencoders (SAEs) have emerged as a powerful technique for decomposing language model representations into interpretable features. Current interpretation methods infer feature semantics from activation patterns, but overlook that…

机器学习 · 计算机科学 2026-02-02 Yiting Liu , Zhi-Hong Deng

Sparse autoencoders (SAEs) interpret neural network representations by decomposing activations into sparse combinations of dictionary atoms. However, SAEs assume features combine additively through linear reconstruction, an assumption that…

机器学习 · 计算机科学 2026-05-26 Panagiotis Koromilas , Andreas D. Demou , James Oldfield , Yannis Panagakis , Mihalis Nicolaou

A key challenge in interpretability is to decompose model activations into meaningful features. Sparse autoencoders (SAEs) have emerged as a promising tool for this task. However, a central problem in evaluating the quality of SAEs is the…

机器学习 · 计算机科学 2024-10-11 Constantin Venhoff , Anisoara Calinescu , Philip Torr , Christian Schroeder de Witt

Sparse autoencoders (SAEs) are a popular technique for interpreting language model activations, and there is extensive recent work on improving SAE effectiveness. However, most prior work evaluates progress using unsupervised proxy metrics…

The power of sparse signal modeling with learned over-complete dictionaries has been demonstrated in a variety of applications and fields, from signal processing to statistical inference and machine learning. However, the statistical…

信息论 · 计算机科学 2017-04-26 Ignacio Ramírez , Guillermo Sapiro

The power of sparse signal coding with learned dictionaries has been demonstrated in a variety of applications and fields, from signal processing to statistical inference and machine learning. However, the statistical properties of these…

信息论 · 计算机科学 2010-10-25 Ignacio Ramírez , Guillermo Sapiro

Sparse auto-encoders (SAEs) have become a prevalent tool for interpreting language models' inner workings. However, it is unknown how tightly SAE features correspond to computationally important directions in the model. This work…

机器学习 · 计算机科学 2025-02-25 Thomas Dooms , Daniel Wilhelm

Sparse autoencoders (SAEs) trained on large language model activations output thousands of features that enable mapping to human-interpretable concepts. The current practice for analyzing these features primarily relies on inspecting…

计算与语言 · 计算机科学 2026-03-26 Wilson E. Marcílio-Jr , Danilo M. Eler

Sparse Autoencoders have emerged as powerful tools for interpreting the internal representations of Large Language Models, yet they often fail to capture domain-specific features not prevalent in their training corpora. This paper…

机器学习 · 计算机科学 2025-07-18 Nikita Koriagin , Yaroslav Aksenov , Daniil Laptev , Gleb Gerasimov , Nikita Balagansky , Daniil Gavrilov

Sparse autoencoders (SAEs) have emerged as a promising approach for learning interpretable features from neural network activations. However, the optimization landscape for SAE training can be challenging due to correlations in the input…

机器学习 · 计算机科学 2025-11-19 Ashwin Saraswatula , David Klindt

Sparse autoencoders (SAEs) have emerged as a powerful technique for extracting human-interpretable features from neural networks activations. Previous works compared different models based on SAE-derived features but those comparisons have…

计算机视觉与模式识别 · 计算机科学 2025-07-25 Clément Cornet , Romaric Besançon , Hervé Le Borgne

Large Language Models (LLMs) are traditionally viewed as black-box algorithms, therefore reducing trustworthiness and obscuring potential approaches to increasing performance on downstream tasks. In this work, we apply an effective LLM…

计算与语言 · 计算机科学 2025-07-10 Shun Wang , Tyler Loakman , Youbo Lei , Yi Liu , Bohao Yang , Yuting Zhao , Dong Yang , Chenghua Lin

Sparse autoencoders (SAEs) have emerged as a promising approach in language model interpretability, offering unsupervised extraction of sparse features. For interpretability methods to succeed, they must identify abstract features across…

机器学习 · 计算机科学 2025-09-08 Lovis Heindrich , Philip Torr , Fazl Barez , Veronika Thost

Understanding how features evolve across layers in deep neural networks is a fundamental challenge in mechanistic interpretability, particularly due to polysemanticity and feature superposition. While Sparse Autoencoders (SAEs) have been…

机器学习 · 计算机科学 2025-03-04 Nikita Balagansky , Ian Maksimov , Daniil Gavrilov

Sparse Autoencoders (SAEs) are powerful tools for interpreting neural representations, yet their use in audio remains underexplored. We train SAEs across all encoder layers of Whisper and HuBERT, provide an extensive evaluation of their…

The ability of large language models (LLMs) to follow instructions is crucial for their practical applications, yet the underlying mechanisms remain poorly understood. This paper presents a novel framework that leverages sparse autoencoders…

机器学习 · 计算机科学 2025-02-18 Zirui He , Haiyan Zhao , Yiran Qiao , Fan Yang , Ali Payani , Jing Ma , Mengnan Du

Sparse Autoencoders (SAEs) are an interpretability technique aimed at decomposing neural network activations into interpretable units. However, a major bottleneck for SAE development has been the lack of high-quality performance metrics,…

机器学习 · 计算机科学 2024-12-02 Adam Karvonen , Can Rager , Samuel Marks , Neel Nanda