EST-PRM: Stress-Testing Process Reward Models Before They Become Load-Bearing

Ähnliche Einträge

• Detecting Proxy Gaming in RL and LLM Alignment via Evaluator Stress Tests
  von: Shihab, Ibne Farabi, et al.
  Veröffentlicht: (2025)
• Dynamic Proxy-Mixing: Transferring Replay Controllers from Small to Large Models for Continual Instruction Tuning
  von: Shihab, Ibne Farabi, et al.
  Veröffentlicht: (2026)
• Continual Calibration: Coverage Can Collapse Before Accuracy in Lifelong LLM Fine-Tuning
  von: Shihab, Ibne Farabi, et al.
  Veröffentlicht: (2026)
• What Fundamental Structure in Reward Functions Enables Efficient Sparse-Reward Learning?
  von: Shihab, Ibne Farabi, et al.
  Veröffentlicht: (2025)
• Topology-Aware State Abstraction with Tangle Cores for Markov Decision Processes
  von: Shihab, Ibne Farabi, et al.
  Veröffentlicht: (2026)