Research map / Architectures and efficiency / RNNs and normalization
Ensembling Sparse Autoencoders
Soham Gadgil, Chris Lin, Su-In Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use boosting ensemble of 8 SAEs instead of an expanded SAE | SAE (Gemma 2-2B) · Intrinsic evaluation | MSE: 472.330 -> 46.538 | Improved |
| use boosting ensemble of 8 SAEs instead of an expanded SAE | SAE (Pythia-160M) · Intrinsic evaluation | stability: 0.204 -> 0.680 | Improved |
| use naive bagging of 8 SAEs instead of an expanded SAE | SAE (GELU-1L) · Intrinsic evaluation | explained variance: 0.946 -> 0.895 | Hurt |
| use naive bagging of 8 SAEs instead of an expanded SAE | SAE (Gemma 2-2B) · Intrinsic evaluation | stability: 0.268 -> 0.633 | Improved |
| use boosting instead of naive bagging | SAE ensemble (Gemma 2-2B) · Intrinsic evaluation | MSE: 234.128 -> 46.538 | Improved |
| use boosting instead of naive bagging | SAE ensemble (Gemma 2-2B) · Intrinsic evaluation | stability: 0.633 -> 0.583 | Hurt |
| ensemble 8 SAEs with boosting instead of single base SAE | SAE (Gemma 2-2B) · Intrinsic evaluation | MSE: 716.659 -> 46.538 | Improved |
| ensemble 8 SAEs with boosting instead of expanded SAE | SAE (GELU-1L) · AutoInterp | AutoInterp score: 0.714 -> 0.863 | Improved |
| use naive bagging instead of boosting for top-1 feature | SAE ensemble (Pythia-70M) · Concept detection GitHub Code | accuracy: 0.682 -> 0.715 | Improved |
| use top-2 concept features instead of top-1 with boosting | SAE boosting (Pythia-70M) · Concept detection AG News | accuracy: 0.759 -> 0.803 | Improved |
| use boosting ensemble instead of expanded SAE | SAE (Pythia-70M) · SCR Professor vs. Nurse | S_SHIFT: 0.047 -> 0.066 | Improved |
| use naive bagging instead of expanded SAE | SAE (Pythia-70M) · SCR Attorney vs. Teacher | S_SHIFT: 0.021 -> 0.003 | Hurt |
Closest papers by meaning
- Improving Sparse Decomposition of Language Model Activations with Gated Sparse Autoencoders (NeurIPS 2024 · similarity 0.92)
- Are Sparse Autoencoders Useful? A Case Study in Sparse Probing (ICML 2025 · similarity 0.91)
- Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines? (NeurIPS 2026 · similarity 0.90)
- SAEBench: A Comprehensive Benchmark for Sparse Autoencoders in Language Model Interpretability (ICML 2025 · similarity 0.90)
- Low-Rank Adapting Models for Sparse Autoencoders (ICML 2025 · similarity 0.90)
- Compute Optimal Inference and Provable Amortisation Gap in Sparse Autoencoders (ICML 2025 · similarity 0.89)
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.
