Unverified50% confidenceFactExact time
When the Brier score or log score of a sampled action's own probability is used as the REINFORCE reward, each incorrect action receives a reward at least comparable to that of the correct (gold) action.
1
Sources
50%
Confidence
Long-term
Relevance
11/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTaskProof's probability-weighted expected net income formula is "net prize × win probability", with a payoff of 0 when losing, and this number applies only to the long-run average of many repeated decisions, not to the actual amount received in a single instance.72% similarUnverifiedIn the synthetic experiment, "always output RISK_LEVEL: 1" achieves an ordinary accuracy as high as 0.922, but an average reward of −0.000, showing that class-balanced rewards can prevent laziness.70% similarUnverified期望价值的计算方式为各种可能结果的收益乘以其发生概率之和69% similarUnverifiedThompson Sampling models the posterior reward probability of each arm in the multi-armed bandit problem using a Beta distribution, samples from the posterior each round and selects the arm with the highest sampled value, and achieves a theoretically better regret bound than ε-greedy.68% similarUnverifiedOn a set of ordered-scale questions, using probability scores as the REINFORCE reward caused accuracy to drop sharply from 0.636 to 0.111.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1164992API
curl https://kongchang.com/api/v1/knowledge/claims/1164992MCP
get_claim(id=1164992)