Research map / Architectures and efficiency / RNNs and normalization
Sparse Autoencoders Find Highly Interpretable Features in Language Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Trains sparse autoencoders on language model activations to recover interpretable monosemantic features from polysemantic neurons, evaluated by human and automated interpretability scores.
Problem
Polysemanticity in neural network neurons, hypothesized to arise from superposition, prevents concise human-understandable explanations of model internals.
Approach
Use sparse autoencoders trained on intermediate language model activations to decompose them into a larger set of sparse, more interpretable features, as a way to undo superposition.
Limitations
- The provided text is only the NeurIPS template instructions, not the paper body; details are inferred from the title and abstract fragment.
Experimental setup
Metrics: interpretability score human interpretability rating
Baselines: neuron basis
Builds on: superposition hypothesis sparse dictionary learning
Closest papers by meaning
- Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words (ICLR 2025 · similarity 0.90)
- SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability (ICML 2025 · similarity 0.90)
- Scaling and evaluating sparse autoencoders (ICLR 2025 · similarity 0.89)
- Automatically Interpreting Millions of Features in Large Language Models (ICML 2025 · similarity 0.89)
- Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models (ICLR 2025 · similarity 0.88)
- Projecting Assumptions: The Duality Between Sparse Autoencoders and Concept Geometry (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.