Sparse Weight Decomposition for Efficient Circuit Extraction
Chuanhao Yan, Xuhan Huang, Yawen Duan, Zhenfei Yin, Hang Zhao, Bryan Dai, Jie Fu
cs.LG, cs.CL
Submitted: 2026-08-04
Code: https://github.com/veri-safe/SWD
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition
- Stochastic Parameter Decomposition
- Towards Automated Circuit Discovery for Mechanistic Interpretability
- Sparse Autoencoders Find Highly Interpretable Features in Language Models
- Transcoders Find Interpretable LLM Feature Circuits
- Sparse Attention Decomposition Applied to Circuit Tracing
- Quantized Sparse Weight Decomposition for Neural Network Compression
- Locating and Editing Factual Associations in GPT
- Qwen2.5 Technical Report
- Circuit Claims Depend on What Is Extracted and How It Is Compared
- Attribution Patching Outperforms Automated Circuit Discovery
- Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small
- ASVD: Activation-aware Singular Value Decomposition for Compressing Large Language Models
- Identifiability in Two-Layer Sparse Matrix Factorization
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks