The 1202-layer CIFAR-10 ResNet optimizes to below 0.1% training error but has 7.93% test error, worse than ResNet-110 despite similar training error. The authors attribute this generalization gap to overfitting by a 19.4M-parameter model on the small dataset and note that they did not use maxout or dropout.

The 1202-layer CIFAR-10 ResNet optimizes to below 0.1% training error but has 7.93% test error, worse than ResNet-110 despite similar training error. The authors attribute this generalization gap to overfitting by a 19.4M-parameter model on the small dataset and note that they did not use maxout or dropout. · CiteArk

测量结果

已评估 0/5 项

获得支持 0有争议或冲突 0尚未评估 5

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。
论文报告	最近可用实测	证据判断
0.1%	—	尚未评估
7.93%	—	尚未评估
19.4 million parameters	—	尚未评估
6.43%	—	尚未评估
6.61%	—	尚未评估

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果