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.

14 activities

Lesson

At a glance

Course Overview: Deep Residual Learning

Course Overview: Deep Residual Learning

Lesson

Understanding the Degradation Problem

The Degradation Problem and Optimization Difficulties

The Degradation Problem in Deep Networks

Lesson

Overfitting vs. Degradation

The Degradation Problem and Optimization Difficulties

In classical machine learning, a model with more parameters typically risks overfitting, which is characterized by extremely low training error but high test error. The degradation problem is distinct because the training error itself increases, indicating that the optimization algorithm is failing to converge to a suitable parameter configuration.

Lesson

Mathematical Formulation of Residual Learning

Mathematical Formulation of Residual Learning

Residual Learning Formulation

Lesson

Residual Building Block with Identity Mapping

Mathematical Formulation of Residual Learning

This equation defines the fundamental building block of Deep Residual Learning (ResNet). Instead of forcing stacked layers to directly fit a desired underlying mapping mathcal H ( mathbf x ) , we let these layers approximate a residual mapping mathcal F ( mathbf x , W i ) = mathcal H ( mathbf x ) - mathbf x . The original mapping is recast into mathcal F ( mathbf x ) + mathbf x via an identity shortcut connection.

Visual

Residual Building Block with Identity Mapping: step by step

Mathematical Formulation of Residual Learning

Follow the existing bounded walkthrough in its intended sequence.

Lesson

Zero-Padding vs. Projection Shortcuts

Dimension Matching and Projection Shortcuts

Dimension Matching in Shortcut Connections

Lesson

Residual Building Block with Projection Shortcut

Dimension Matching and Projection Shortcuts

This equation defines a residual building block with a projection shortcut. When the input dimension of mathbf x differs from the output dimension of the residual function mathcal F , a linear projection matrix W s is applied to the shortcut connection to match the dimensions.

Visual

Residual Building Block with Projection Shortcut: step by step

Dimension Matching and Projection Shortcuts

Follow the existing bounded walkthrough in its intended sequence.

Lesson

The Bottleneck Building Block

Deeper Bottleneck Architectures

Deeper Bottleneck Architectures

Lesson

Empirical Analysis of Layer Responses

Empirical Analysis of Layer Responses

An analysis of the standard deviations (std) of layer responses on CIFAR-10 reveals that ResNets generally have smaller responses than their plain counterparts. This supports the motivation that residual functions are closer to zero than non-residual functions. Furthermore, deeper ResNets (such as ResNet-56 and ResNet-110 compared to ResNet-20) exhibit even smaller response magnitudes, indicating that individual layers modify the signal less as depth increases.

Quiz

Test your understanding

Comprehensive Assessment

12 questions grounded in this paper section.

Resource

Further learning

Comprehensive Assessment

3 supplementary resources for this paper section.

Resource

Research and implementation context

Comprehensive Assessment

Explore notable related work and public implementation context.