Do Sparse Autoencoders Learn Meaningful Concept Hierarchies?
For researchers using SAEs for interpretability, this work provides a much-needed evaluation framework and highlights a fundamental challenge.
The paper identifies a lack of agreed-upon requirements for evaluating hierarchical structure in sparse autoencoder (SAE) feature spaces, proposes a set of requirements and an evaluation protocol, and finds that current SAE approaches struggle to produce meaningful hierarchies due to feature absorption.
Sparse autoencoders (SAEs) have become an important tool for unsupervised concept discovery in large models. To make the resulting feature spaces more interpretable and manageable, recent approaches have begun imposing hierarchical structure, either explicitly or as an implicit effect of training constraints, yet rigorous comparison remains difficult. There are no agreed-upon requirements for what a meaningful feature hierarchy should satisfy, and evaluation has largely relied on qualitative illustrations with fragmented quantitative protocols. To address this, we derive a set of key requirements for generalization/specialization hierarchies in unsupervised concept discovery, drawing on semantic net and taxonomy research alongside recent SAE work, and use them to derive a concrete evaluation protocol. Applying this protocol to current SAE approaches trained on visual data, we find that while feature spaces generally provide a basis for sensible hierarchies, establishing good hierarchical structure remains challenging. In particular, feature absorption, both in its well-known hard form and in a continuous, soft form, systematically compromises hierarchy quality, pointing to a fundamental tension that future approaches will need to navigate.