The Local Interaction Basis: Identifying Computationally-Relevant and Sparsely Interacting Features in Neural Networks

Documenti analoghi

• Using Degeneracy in the Loss Landscape for Mechanistic Interpretability
  di: Bushnaq, Lucius, et al.
  Pubblicazione: (2024)
• Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition
  di: Braun, Dan, et al.
  Pubblicazione: (2025)
• Detecting Strategic Deception Using Linear Probes
  di: Goldowsky-Dill, Nicholas, et al.
  Pubblicazione: (2025)
• Identifying Functionally Important Features with End-to-End Sparse Dictionary Learning
  di: Braun, Dan, et al.
  Pubblicazione: (2024)
• Mathematical Models of Computation in Superposition
  di: Hänni, Kaarel, et al.
  Pubblicazione: (2024)