Unverified50% confidenceFactExact time
In the reproduction experiments, the variance across different random seeds was almost as large as any trend across training set sizes N.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027
Sources
Related Entities
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77% similarUnverifiedTraining samples for the N2 category tend to be scarcer, and data imbalance exacerbates the model's tendency to favor the majority class N1, which is one of the important contributing factors to the asymmetric recall.72% similarUnverifiedEven with identical inputs, different random seeds or sampling parameters can lead to entirely different outputs from large models.70% similarUnverifiedWhen individually tuning dedicated learning rates for each control group and repeating experiments with three random seeds, the advantage of repairing only the encoder disappears, reaching parity with the other approaches.69% similarUnverifiedZero-shot prediction results are cached and reused across random seeds, so this part does not constitute truly independent repeated experiments.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1012974API
curl https://kongchang.com/api/v1/knowledge/claims/1012974MCP
get_claim(id=1012974)