On MNIST with a three-layer ReLU MLP and AdamW, every EGD variant reaches 85% validation accuracy in fewer steps than Grokfast. RSVD variants also substantially reduce wall-clock time; rank 4 is fastest in time, ranks 8 and 4 tie for the fewest steps, and exact-SVD EGD has the highest printed final validation accuracy.

On MNIST with a three-layer ReLU MLP and AdamW, every EGD variant reaches 85% validation accuracy in fewer steps than Grokfast. RSVD variants also substantially reduce wall-clock time; rank 4 is fastest in time, ranks 8 and 4 tie for the fewest steps, and exact-SVD EGD has the highest printed final validation accuracy. · CiteArk

测量结果

已评估 0/34 项

获得支持 0有争议或冲突 0尚未评估 34

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
method: Vanilla AdamW	43,100 optimization steps	—	尚未评估
method: Vanilla AdamW	660 seconds	—	尚未评估
method: Vanilla AdamW	15.3 milliseconds	—	尚未评估
method: Vanilla AdamW	89%	—	尚未评估
method: Grokfast	1,800 optimization steps	—	尚未评估
method: Grokfast	23.9 fold	—	尚未评估
method: Grokfast	28.7 seconds	—	尚未评估
method: Grokfast	23 fold	—	尚未评估
method: Grokfast	16 milliseconds	—	尚未评估
method: Grokfast	91%	—	尚未评估
method: EGD (SVD)	700 optimization steps	—	尚未评估
method: EGD (SVD)	61.6 fold	—	尚未评估

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果