Unmasking Backdoors: An Explainable Defense via Gradient-Attention Anomaly Scoring for Pre-trained Language Models

Documenti analoghi

• Adaptive Deviation Learning for Visual Anomaly Detection with Data Contamination
  di: Das, Anindya Sundar, et al.
  Pubblicazione: (2024)
• Multi-Cue Anomaly Detection and Localization under Data Contamination
  di: Das, Anindya Sundar, et al.
  Pubblicazione: (2026)
• Pruning Strategies for Backdoor Defense in LLMs
  di: Chapagain, Santosh, et al.
  Pubblicazione: (2025)
• Mechanistic Exploration of Backdoored Large Language Model Attention Patterns
  di: Baker, Mohammed Abu, et al.
  Pubblicazione: (2025)
• Unmasking Hallucinations: A Causal Graph-Attention Perspective on Factual Reliability in Large Language Models
  di: kurra, Sailesh kiran, et al.
  Pubblicazione: (2026)