Superposition as Lossy Compression: Measure with Sparse Autoencoders and Connect to Adversarial Vulnerability

Documenti analoghi

• Mechanistic Interpretability for AI Safety -- A Review
  di: Bereska, Leonard, et al.
  Pubblicazione: (2024)
• From MLP to NeoMLP: Leveraging Self-Attention for Neural Fields
  di: Kofinas, Miltiadis, et al.
  Pubblicazione: (2024)
• CEAR: Certified Ensemble Adversarial Robustness in DNNs
  di: Sadig, Daniel, et al.
  Pubblicazione: (2026)
• GNN's Uncertainty Quantification using Self-Distillation
  di: Daneshvar, Hirad, et al.
  Pubblicazione: (2025)
• Mechanistic Neural Networks for Scientific Machine Learning
  di: Pervez, Adeel, et al.
  Pubblicazione: (2024)