University of Bologna · Prof. Matteo Ferrara · a.y. 2025/2026

Deep Learning

12 chapters54 interactive widgets63 plates~9 hours of study
Course roadmap: foundations feed neural networks and training, then architectures for structured data, generative models, and reinforcement learning
Plate 00 — The road of the course. Foundations (math and automatic differentiation), then neural networks and how to train them, then the architectures that made deep learning scale (CNNs, RNNs, autoencoders, transformers), and finally the two frontiers: generating data and learning from interaction.

How to study

I

Foundations

  1. 01Introduction: machine learning and deep learning25 min · 4 widgets
  2. 02Linear algebra, calculus and automatic differentiation35 min · 5 widgets
Tavola: una funzione composta con gradiente che scorre all'indietro
Tav. I — Forward path, backward gradient
II

Neural networks and training

  1. 03Artificial neural networks45 min · 6 widgets
  2. 04Backpropagation45 min · 5 widgets
  3. 05Optimization algorithms45 min · 5 widgets
Tavola: una curva di loss con discesa del gradiente a passi verso il minimo
Tav. II — Descent into the minimum
III

Architectures for structured data

  1. 06Convolutional neural networks40 min · 4 widgets
  2. 07CNN architectures: from LeNet-5 to ResNet40 min · 5 widgets
  3. 08Recurrent neural networks40 min · 4 widgets
  4. 09Autoencoders45 min · 4 widgets
  5. 10Transformers50 min · 4 widgets · 1 film
Tavola: tre architetture: una griglia convolutiva, una catena ricorrente, una coppia encoder-decoder con attenzione
Tav. III — Grids, sequences, attention
IV

Generative models

  1. 11Generative models55 min · 4 widgets
Tavola: generatore e discriminatore in un gioco avversario, e una catena di denoising a diffusione
Tav. IV — The adversarial game and the denoising chain
V

Reinforcement learning

  1. 12Reinforcement learning55 min · 4 widgets
Tavola: il ciclo agente-ambiente con azione, ricompensa e stato
Tav. V — The agent–environment loop