Unverified50% confidenceFactExact time
在真实映射为均匀随机的假设下，论文证明了错误率下界为 E ≥ (M/N)δ*(B/M) + (1-M/N)(1-1/K)
1
Sources
50%
Confidence
Long-term
Relevance
9/14/2026
First Seen
Sources
Related Entities
Related Claims
Unverified蒙特卡洛估计的标准误差与模拟次数N的关系为标准误差正比于1/√N72% similarUnverifiedLLM每次调用可建模为伯努利试验，N次调用通过次数服从二项分布B(N, p)，通过率95%置信区间约为p̂ ± 1.96×√(p̂(1-p̂)/N)70% similarUnverified均匀量化引入的误差近似服从均匀分布，其方差为Δ²/12（Δ为量化步长）70% similarUnverified误差范围的计算源于中心极限定理，公式为 MOE = z × √(p(1-p)/n)，95%置信水平时z约为1.9670% similarUnverified论文中的"with probability at least 1-δ"是PAC（Probably Approximately Correct）学习理论的标准语言，意味着误差上界以至少1-δ的概率成立69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/919747API
curl https://kongchang.com/api/v1/knowledge/claims/919747MCP
get_claim(id=919747)