Go to CMU 2025 LTI-SRS homepageDecoding Dark Matter: Specialized Sparse Autoencoders for Interpreting Rare Concepts in Foundation Models
Track: Main Track
Keywords: free-text/natural language explanations, hierarchical & concept explanations, feature attribution, probing
TL;DR: Specialized Sparse Autoencoders enhance interpretability of LLMs by capturing rare features in specific subdomains.
Abstract: Understanding and mitigating the potential risks associated with foundation models (FMs) hinges on developing effective interpretability methods. Sparse Autoencoders (SAEs) have emerged as a promising tool for disentangling FM representations, but they struggle to capture rare, yet crucial concepts in the data. We introduce Specialized Sparse Autoencoders (SSAEs), designed to illuminate these elusive dark matter features by focusing on specific subdomains. We present a practical recipe for training SSAEs, demonstrating the efficacy of dense retrieval for data selection and the benefits of Tilted Empirical Risk Minimization as a training objective to improve concept recall. Our evaluation of SSAEs on standard metrics, such as downstream perplexity and
sparsity, show that they effectively capture subdomain tail concepts, exceeding the capabilities of general-purpose SAEs. We showcase the practical utility of SSAEs in a case study on the Bias in Bios dataset, where SSAEs achieve a 12.5% increase in worst-group classification accuracy over the pretrained general-purpose SAE when applied to remove spurious gender information. SSAEs provide a powerful new lens for peering into the inner workings of FMs in subdomains.
Submission Number: 2
Loading