Administrative information

Implementing and training a CNN for an image classification problem from scratch
Fine-tuning of an already trained network
Transfer learning using architectures trained on ImageNet

Expected Preparation

Learning Events to be Completed Before

Lecture: Convolutional Neural Networks

Obligatory for Students

Theory on CNN

Optional for Students

None

References and background for students:

None

Recommended for Teachers

None.

Lesson Materials

Instructions for Teachers

This Tutorial covers fundamental CNN development, training and testing. Three different tutorial implemented in form of Jupyter Notebook will be shown and discussed. In particular:

the implementation of a simple CNN will be shown. The training will be made with a simple freely available dataset (e.g. MNIST). Evaluation in terms of accuracy of a test set after the training stage will be shown.
the fine tuning of an already trained network will be made on a new dataset (e.g., Fashion-MNIST). Evaluation and a comparison with a network trained from scratch will be shown and discussed.
how to load and save custom models will be shown.

Time schedule

Duration (min)	Description	Concepts	Activity	Material
20	Implementing and training a simple CNN
20	Fine tuning of an already trained network
20	load and save architectures

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