Mitigating Shortcut Reasoning in Language Models: A Gradient-Aware Training Approach

Documenti analoghi

• Efficient Post-Training Refinement of Latent Reasoning in Large Language Models
  di: Wang, Xinyuan, et al.
  Pubblicazione: (2025)
• Break the Chain: Large Language Models Can be Shortcut Reasoners
  di: Ding, Mengru, et al.
  Pubblicazione: (2024)
• Evolving Demonstration Optimization for Chain-of-Thought Feature Transformation
  di: Wang, Xinyuan, et al.
  Pubblicazione: (2026)
• CURE: Controlled Unlearning for Robust Embeddings -- Mitigating Conceptual Shortcuts in Pre-Trained Language Models
  di: Kocak, Aysenur, et al.
  Pubblicazione: (2025)
• Deceptive Semantic Shortcuts on Reasoning Chains: How Far Can Models Go without Hallucination?
  di: Li, Bangzheng, et al.
  Pubblicazione: (2023)