ResearchInteractive previewWrite with Jenni Residual connections make very deep neural networks easier to optimize. Instead of learning an entire transformation, layers learn a change relative to their input.
The original PDF. Select a passage to explore it with Jenni.
Residual learning enables effective training of substantially deeper networks.
The authors evaluate networks with up to 152 layers on ImageNet.
The learned representations also improve object detection results.
Residual networks are compared with plain networks on image classification, with additional evaluation of transferred representations for object detection.
Greater depth still has a computational cost. The reported vision benchmarks do not establish improvements for every architecture or application.
Editorial overview · Always refer to the original study.