Loading page…
Explore ArkGraph and select the steps to run.
This paper introduces residual learning for training substantially deeper neural networks. Instead of directly fitting a desired mapping, stacked layers learn a residual that is added to an identity shortcut. The authors compare plain and residual networks on ImageNet and CIFAR-10, examine shortcut variants and residual-response magnitudes, and scale residual networks to 152 layers on ImageNet and 1202 layers on CIFAR-10. Reported results show reduced optimization degradation and improved classification accuracy with depth. The learned representations also improve Faster R-CNN detection on PASCAL VOC and COCO and support competitive ImageNet detection and localization systems.
Residual shortcuts made very deep networks much easier to optimize and allowed depth to improve accuracy instead of causing the training degradation seen in matched plain networks. The evidence spans classification, detection, and localization, but the paper also reports limits: a 1202-layer CIFAR model generalizes worse than a 110-layer model, very deep training can require learning-rate warm-up, and several conclusions from plots are qualitative rather than exact numeric targets.
The paper’s claims are available in Research claims.
Past research reports
Saved reports remain available. New report generation is paused.