Which Sparse Autoencoder Features Are Real? Model-X Knockoffs for False Discovery Rate Control

2025/11/12 by Tsogt-Ochir Enkhbayar, Enkhbayar, Tsogt-Ochir · 1 citation
Computer Science · #Adversarial Robustness in Machine Learning #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Machine Learning (cs.LG)

paper · pdf · doi:10.48550/arxiv.2511.11711

openalex publication_date 2025/11/12 · openalex created_date 2025/11/19 · openalex updated_date 2026/07/28

Abstract

Although sparse autoencoders (SAEs) are crucial for identifying interpretable features in neural networks, it is still challenging to distinguish between real computational patterns and erroneous correlations. We introduce Model-X knockoffs to SAE feature selection, using knock-off+ to control the false discovery rate (FDR) with finite-sample guarantees under the standard Model-X assumptions (in our case, via a Gaussian surrogate for the latent distribution). We select 129 features at a target FDR q=0.1 after analyzing 512 high-activity SAE latents for sentiment classification using Pythia-70M. About 25% of the latents under examination carry task-relevant signal, whereas 75% do not, according to the chosen set, which displays a 5.40x separation in knockoff statistics compared to non-selected features. Our method offers a re-producible and principled framework for reliable feature discovery by combining SAEs with multiple-testing-aware inference, advancing the foundations of mechanistic interpretability.

Citations

• Scaling and evaluating sparse autoencoders
• Sparse Autoencoders Find Highly Interpretable Features in Language Models
• Towards Automated Circuit Discovery for Mechanistic Interpretability
• A Note on Normalized Emergence Timing (in Pythia Language Model Evaluations)
• Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small
• Governance Architecture for Neural Network Superposition: A Structural Solution to Hallucination via Routing and Interference Filtering
• Fast and Powerful Conditional Randomization Testing via Distillation
• DeepPINK: reproducible feature selection in deep neural networks
• The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks
• Panning for Gold: Model-X Knockoffs for High-dimensional Controlled Variable Selection

Cited by

• Learning from Negative Examples: Why Warning-Framed Training Data Teaches What It Warns Against

Related

• Evaluating Sparse Autoencoders on Targeted Concept Erasure Tasks
• Sparse Autoencoders Do Not Find Canonical Units of Analysis
• Analyzing (In)Abilities of SAEs via Formal Languages
• Evaluating Sparse Autoencoders: From Shallow Design to Matching Pursuit
• Sparse but Wrong: Incorrect L0 Leads to Incorrect Features in Sparse Autoencoders