Improving Dictionary Learning with Gated Sparse Autoencoders

This is experimental HTML to improve accessibility. We invite you to report rendering errors. Use Alt+Y to toggle on accessible reporting links and Alt+Shift+Y to toggle off. Learn more about this project and help improve conversions. srajamanoharan@google.com and neelnanda@google.com Recent work has found that sparse autoencoders (SAEs) are an effective technique for unsupervised discovery of interpretable features in language models’ (LMs) activations, by finding sparse, linear reconstructions of LM activations. We introduce the Gated Sparse Autoencoder (Gated SAE), which achieves a Pareto improvement over training with prevailing methods. In SAEs, the L1 penalty used to encourage sparsity introduces many undesirable biases, such as shrinkage – systematic underestimation of feature activations. The key insight of Gated SAEs is to separate the functionality of (a) determining which directions to use and (b) estimating the magnitudes of those directions: this enables us to apply the L

\correspondingauthor srajamanoharan@google.com and neelnanda@google.com Improving Dictionary Learning with Gated Sparse Autoencoders Senthooran Rajamanoharan : Joint contribution. † : Core infrastructure contributor. Arthur Conmy : Joint contribution. † : Core infrastructure contributor. Lewis Smith Tom Lieberum † Vikrant Varma † János Kramár Rohin Shah Neel Nanda Abstract Recent work has found that sparse autoencoders (SAEs) are an effective technique for unsupervised discovery of interpretable features in language models’ (LMs) activations, by finding sparse, linear reconstructions of LM act

related reading

• An Intuitive Explanation of Sparse Autoencoders for LLM Interpretability | Adam Karvonenadamkarvonen.github.io
• Towards Monosemanticity: Decomposing Language Models With Dictionary Learningtransformer-circuits.pub
• OpenAI SAE Training Paperarxiv.org
• Identifying Functionally Important Features with End-to-End Sparse Dictionary Learning — LessWronglesswrong.com
• Sparse Autoencoders Find Highly Interpretable Features in Language Modelsarxiv.org
• A gentle introduction to sparse autoencoders — LessWronglesswrong.com
• sparse-autoencoders.pdfcdn.openai.com
• Matryoshka Sparse Autoencoders — LessWronglesswrong.com
• Efficient Dictionary Learning with Switch Sparse Autoencoders — LessWronglesswrong.com
• A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Modelsarxiv.org
• Towards Monosemanticity: Decomposing Language Models With Dictionary Learningtransformer-circuits.pub
• SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretabilityarxiv.org