Related papers: One Language, Two Scripts: Probing Script-Invarian…
Reinforcement Learning (RL) agents often struggle to generalize knowledge to new tasks, even those structurally similar to ones they have mastered. Although recent approaches have attempted to mitigate this issue via zero-shot transfer,…
Audio pretrained models are widely employed to solve various tasks in speech processing, sound event detection, or music information retrieval. However, the representations learned by these models are unclear, and their analysis mainly…
Sparse autoencoders (SAEs) aim to disentangle model activations into monosemantic, human-interpretable features. In practice, learned features are often redundant and vary across training runs and sparsity levels, which makes…
In-context learning, the ability to adapt based on a few examples in the input prompt, is a ubiquitous feature of large language models (LLMs). However, as LLMs' in-context learning abilities continue to improve, understanding this…
Sparse Autoencoders (SAEs) extract features from LLM internal activations, meant to correspond to interpretable concepts. A core SAE training hyperparameter is L0: how many SAE features should fire per token on average. Existing work…
The high-level concepts that a neural network uses to perform computation need not be aligned to individual neurons (Smolensky, 1986). Language model interpretability research has thus turned to techniques such as \textit{sparse…
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…
Sparse autoencoders (SAEs) are a core interpretability tool for large language models, and progress on SAE architectures depends on benchmarks that reliably distinguish better SAEs from worse ones. We audit the SAE quality metrics in…
Traditionally, character-level transduction problems have been solved with finite-state models designed to encode structural and linguistic knowledge of the underlying process, whereas recent approaches rely on the power and flexibility of…
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…
Dense embeddings deliver strong retrieval performance but often lack interpretability and controllability. This paper introduces a novel approach using sparse autoencoders (SAE) to interpret and control dense embeddings via the learned…
Interpretability is critical in high-stakes domains such as medical imaging, where understanding model decisions is essential for clinical adoption. In this work, we introduce Sparse Autoencoder (SAE)-based interpretability to breast…
Unlearning methods for vision-language models (VLMs) have primarily adapted techniques from large language models (LLMs), relying on weight updates that demand extensive annotated forget sets. Moreover, these methods perform unlearning at a…
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…
While sparse autoencoders (SAEs) have generated significant excitement, a series of negative results have added to skepticism about their usefulness. Here, we establish a conceptual distinction that reconciles competing narratives…
Recent work on sparse autoencoders (SAEs) has shown promise in extracting interpretable features from neural networks and addressing challenges with polysemantic neurons caused by superposition. In this paper, we apply SAEs to the early…
Two traditions of interpretability have evolved side by side but seldom spoken to each other: Concept Bottleneck Models (CBMs), which prescribe what a concept should be, and Sparse Autoencoders (SAEs), which discover what concepts emerge.…
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…
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…
We present a new flavor of Variational Autoencoder (VAE) that interpolates seamlessly between unsupervised, semi-supervised and fully supervised learning domains. We show that unlabeled datapoints not only boost unsupervised tasks, but also…