Administrative information

To know what is a CNN and its main differences with Densely-connected NN
To know the main difference between Locally-Connected Layers to Convolutional ones
To know how to configure a CNN layer
Pooling and batch normalization layers
To know the most famous CNNs: LeNet, AlexNet, ResNet, VGG16, AllConvNet

Expected Preparation

Learning Events to be Completed Before

None

Obligatory for Students

Theory on Artificial Neural Networks

Optional for Students

None

References and background for students:

Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT press. - Chapter 9

Recommended for Teachers

Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT press. - Chapter 9

Lesson Materials

Instructions for Teachers

This lecture will introduce students to the Convolutional Neural Networks (CNNs), explaining the main differences between classical Fully-Connected layers and Convolutional ones. The advantages of the weight sharing given by the Convolutional Layer are introduced and discussed, together with a comparison with the Locally-Connected layers. The Convolution operator is introduced, and kernel size, stride and padding are discussed as the main hyperparameters of a Convolutional layer. Then, Pooling and Batch Normalization layers will be introduced as part of several CNN architectures. To better undestand what a Convolutional layer has learned, possible ways to visualize learned filters will be introduced. Finally, an introduction to the most famous CNN architectures such as NetworkInNetwork and LeNet will be presented.

Introduction to CNNs
- Main issues about fully-connected layers for high-dimensionality data
- Convolutional operator
- Description of a Convolutional layer in terms of neurons
Convolutional layers
- Main properties of a Convolutional layer
  - Local connectivity
  - Weight sharing
- Convolutional layer hyperparameters
  - Filter size
  - Stride
  - Padding
Pooling layers
Visualizing layers
Well-known architectures
- LeNet
- NiN

Time schedule

Duration (min)	Description	Concepts	Activity	Material
10	Introduction to CNNs
15	Convolutional layers
5	Pooling layers
15	Visualizing layers
15	Well-known architectures

Title	Convolutional Neural Networks
Duration	60 min
Module	B
Lesson Type	Lecture
Focus	Technical - Deep Learning
Topic	Deep learning