Latent Reward Steering: An Adaptive Inference-Time Framework that Implicitly Promotes Cognitive Behaviors in Reasoning LLMs

Documenti analoghi

• Weak Critics Make Strong Learners: On-Policy Critique Distillation for Scalable Oversight
  di: Jin, Can, et al.
  Pubblicazione: (2026)
• Two Heads are Better Than One: Test-time Scaling of Multi-agent Collaborative Reasoning
  di: Jin, Can, et al.
  Pubblicazione: (2025)
• Test-Time Spectrum-Aware Latent Steering for Zero-Shot Generalization in Vision-Language Models
  di: Dafnis, Konstantinos M., et al.
  Pubblicazione: (2025)
• Learning from Teaching Regularization: Generalizable Correlations Should be Easy to Imitate
  di: Jin, Can, et al.
  Pubblicazione: (2024)
• GSS: Gated Subspace Steering for Selective Memorization Mitigation in LLMs
  di: Zhang, Xuanqi, et al.
  Pubblicazione: (2026)