PhD in Computer Science

Introduction to Deep Learning

Introduction to Deep Learning

Academic year 2022/2023

---- This is tempoprary text, the course program is under development ----

Sommario del corso

• Course objectives
• Course delivery
• Learning assessment methods
• Program
• Suggested textbooks and readings
• More information
• Teaching tools

Course objectives

Deep learning based approaches outperform traditional approaches in a number of applications spanning from data analysis  to image processing and natural language processing. The availability of large sets of annotated data, powerful yet affordable parallel computers and improved optimization techniques are behind this revolution. This course will provide first an introduction to multilayer neural networks and its cornerstones such as learning with descent of the backpropagated error gradient. Next, convolutional and deep architectures will be introduced in the context of image processing, Finally, the course will look into recurrent architectures for time series and natural language processing.

Course delivery

Frontal lessons (with slides) with seminars and hands-on sessions with Keras on Google Colab.

Learning assessment methods

Students are required to discuss a project related to the topics addressed in the course.

Program

1) Introduction to Neural Networks (8h)

Introduction to machine learning and classical  approaches: SVMs, decition trees, etc. (2h)
Lab: machine learning with Python and Sklearn  (2h)

Multilayer neural networks and learning with descent of backpropagated error gradient (2h)
Lab: exercises on training a simple neural network with Keras (with examples where gradients are applied automatically and manually) (2h)

2) Convolutional and Deep Architectures (8h)

Convolutional neural networks (CNNs) and overview of deep convolutional architectures (2 h)
Lab: Image classification with CNNs using Keras (2 h)

Autoencoders and adversarial learning (2 h)
Lab: Image generation using a GAN approach implemented with  Keras (2h)

3) Natural Language Processing (8h)

RNN: Recurrent Neural Networks, Long Short-term Memory, Gated Recurrent Unit, Transformer (2 h)
Natural language: Text classification with Transformers (3 h)
Natural Language Generation (3 h)

.

More information

• Teaching materials
• Exams and finals
• Go to Moodle
• Visit the forums