Research map / Architectures and efficiency / RNNs and normalization
Circuit Insights: Towards Interpretability Beyond Activations
Elena Golimblevskaia, Aakriti Jain, Bruno Puri, Ammar Ibrahim, Wojciech Samek, Sebastian Lapuschkin
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 WeightLens token descriptions instead of activation-maximization descriptions | Gemma-2-2B transcoder features · ~250 features per layer | Clarity and Responsiveness | Improved |
| use WeightLens instead of MaxAct* (ablation) | Gemma-2-2B transcoder features · ~250 features per layer | Purity | Hurt |
| add LLM refinement (gpt-4o-mini) to WeightLens+Out (ablation) | WeightLens · Gemma-2-2B transcoders | FADE scores | No clear effect |
| add output-token (unembedding) information to WeightLens descriptions (ablation) | WeightLens · Gemma-2-2B transcoders | FADE scores | No clear effect |
| add validation step keeping only tokens that activate the feature (ablation) | WeightLens · Gemma-2-2B transcoders | description quality | Improved |
| use circuit-based input masking and clustering instead of activation-only explainer inputs | CircuitLens · Gemma-2-2B transcoders | FADE scores | Improved |
| use the larger 2.3B-token dataset instead of 24M tokens (ablation) | CircuitLens · Gemma-2-2B transcoders | input-centric FADE metrics | Improved |
| combine WeightLens with CircuitLens-Full (ablation) | CircuitLens · Gemma-2-2B transcoders | robustness to dataset size | Improved |
| use circuit-based clustering | CircuitLens · Gemma-2-2B transcoders | number of very low-clarity features | Improved |
| use WeightLens or MaxAct* descriptions (ablation) | all methods · Gemma-2-2B, GPT-2, Llama-3.2-1B | Faithfulness | No clear effect |
| apply WeightLens to middle layers (ablation) | WeightLens · Gemma-2-2B and Llama-3.2-1B | fraction of token-based features | Hurt |
Closest papers by meaning
- Transcoders find interpretable LLM feature circuits (NeurIPS 2024 · similarity 0.90)
- The Computational Complexity of Circuit Discovery for Inner Interpretability (ICLR 2025 · similarity 0.90)
- Efficient Automated Circuit Discovery in Transformers using Contextual Decomposition (ICLR 2025 · similarity 0.89)
- Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable? (ICLR 2025 · similarity 0.88)
- Towards Understanding Fine-Tuning Mechanisms of LLMs via Circuit Analysis (ICML 2025 · similarity 0.86)
- Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models (ICLR 2025 · similarity 0.86)
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.
