You are reading immutable version 4. The current guide may be newer.

Study the paper

Deep Residual Learning for Image Recognition

Lessons, visuals, quizzes, flashcards, and resources—organized in teaching order.

All activities

Course Overview: Deep Residual Learning

At a glance

At a glance

Course Overview: Deep Residual Learning

Sources

section

1 Introduction

In deep learning, increasing network depth has been a primary driver of performance. However, as networks become deeper, a degradation problem emerges: beyond a certain depth, accuracy saturates and then degrades rapidly. This degradation is not caused by overfitting, as it manifests as higher training error alongside higher test error.

Sources

S1.p3.1

When deeper networks are able to start converging, a degradation problem has been exposed: with the network depth increasing, accuracy gets saturated (which might be unsurprising) and then degrades rapidly. Unexpectedly, such degradation is not caused by overfitting, and adding more layers to a suitably deep model leads to higher training error, as reported in [11, 42] and thoroughly verified by our experiments. Fig. 1 shows a typical example.