Speculative Expansion
Cross-source consensus on Speculative Expansion from 1 sources and 4 claims.
1 sources · 4 claims
How it works
Benefits
Evidence quality
Highlighted claims
- Speculative Expansion reduces sequential tree-search overhead by allowing workers to score from bounded-stale snapshots. — Maximizing Rollout Informativeness under a Fixed Budget: A Submodular View of Tree Search for Tool-Use Agentic Reinforcement Learning
- Speculative expansions are accepted only if rescoring with the current Q-table keeps them within current top-K UUCB ranks. — Maximizing Rollout Informativeness under a Fixed Budget: A Submodular View of Tree Search for Tool-Use Agentic Reinforcement Learning
- Speculative Expansion reduced L=2 InfoTree overhead while keeping Mean@32 nearly unchanged. — Maximizing Rollout Informativeness under a Fixed Budget: A Submodular View of Tree Search for Tool-Use Agentic Reinforcement Learning
- Speculative acceptance averaged 92.3%, and policy-KL change explained most acceptance variance. — Maximizing Rollout Informativeness under a Fixed Budget: A Submodular View of Tree Search for Tool-Use Agentic Reinforcement Learning