Unverified50% confidenceFactExact time
深度学习中的随机种子控制权重初始化、数据加载批次打乱顺序、Dropout神经元随机失活以及优化器中的随机梯度估计等多个随机性来源
1
Sources
50%
Confidence
Long-term
Relevance
7/13/2026
First Seen
Sources
McNemar检验对比机器学习模型：随机种子下的正确用法
redditr/learnmachinelearning7/11/2026
Related Claims
Unverified即便固定所有随机种子，在不同硬件或并行度下，训练结果仍可能出现被大模型放大为显著能力差异的细微偏差76% similarUnverified过度域随机化会引入与真实分布相差过大的不可能样本，稀释有效训练信号，随机化范围的设定需要大量领域专家经验73% similarUnverified通过记录随机种子可精确复现导致失败的特定顺序，再通过二分法逐步缩小污染源的范围72% similarUnverified权重初始化的随机性对模型性能影响最深远，数据批次打乱顺序的随机性影响最小72% similarUnverifiedRandom forests reduce variance and improve generalization through bagging, making them a classic representative of ensemble learning.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/502762API
curl https://kongchang.com/api/v1/knowledge/claims/502762MCP
get_claim(id=502762)