Gradient Starvation in Binary-Reward GRPO: Why Group-Mean Centering Fails and Why the Simplest Fix Works

Documenti analoghi

• The Coupling Tax: How Shared Token Budgets Undermine Visible Chain-of-Thought Under Fixed Output Limits
  di: Nie, Wenhua, et al.
  Pubblicazione: (2026)
• Identified-Set Geometry of Distributional Model Extraction under Top-$K$ Censored API Access
  di: Nie, Wenhua, et al.
  Pubblicazione: (2026)
• Future Validity is the Missing Statistic: From Impossibility to $Φ$-Estimation for Grammar-Faithful Speculative Decoding
  di: Nie, Wenhua, et al.
  Pubblicazione: (2026)
• Why Transaction Cost Economics Failed and How to Fix It
  di: Mingqian, Li
  Pubblicazione: (2024)
• Equilibrium Residuals Expose Three Regimes of Matrix-Game Strategic Reasoning in Language Models
  di: Nie, Wenhua, et al.
  Pubblicazione: (2026)