C9739 - GENERATIVE AI: METODI E APPLICAZIONI

Learning outcomes

At the end of the course, the student is able to: (i) describe fundamental principles in modern generative models (in particular, LLM for text and diffusion models for images); (ii) analyze and compare the main AI tools available on the market, evaluating their limits and possibilities; (iii) develop simple Python applications integrating generative models for the resolution of practical problems; (iv) critically interpret the rapid evolution of scientific and industrial literature; and (v) recognize the ethical, legal and responsible implications of GenAI technologies (e.g. privacy, copyright, hallucinations, bias)

Course contents

- Basic concepts of probability theory and information theory.

- Brief overview of recent Artificial Intelligence models: CNNs, U-Nets, and Transformers.

- Introduction to generative models for images, with particular emphasis on the four main paradigms: Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), Diffusion Models, and Flow Matching.

- Limitations and future perspectives of Artificial Intelligence models for image generation.

- Language models: tokenizers, embeddings, and next-token prediction.

- Pre-training and fine-tuning of Large Language Models.

- Chain-of-Thought (CoT), reasoning, and agents.

- Model specialization through Retrieval-Augmented Generation (RAG).

- Main limitations and future perspectives of language models.

- Integrating text and images: multimodal models and prompt-based image generation.

All the topics will be then integrated with practical coding sessions from the professor.

Readings/Bibliography

- S. J. D. Prince, Understanding Deep Learning, MIT Press, 2023.

- K. P. Murphy, Probabilistic Machine Learning: An Introduction, MIT Press, 2022.

- D. Foster, Generative Deep Learning: Teaching Machines to Paint, Write, Compose, and Play, 2nd edition, O’Reilly Media, 2023.

- J. Alammar and M. Grootendorst, Hands-On Large Language Models: Language Understanding and Generation, O’Reilly Media, 2024.

- Teaching material provided through Virtuale

Teaching methods

In-person lectures and guided classroom exercises.

Assessment methods

The assessment consists of a multiple-choice written test covering the topics addressed during the course, followed by a short oral examination aimed at evaluating the student’s level of understanding, as well as their critical thinking skills.

As an alternative to the oral examination, the student may complete an individual project assigned by the instructor. The project must be carried out according to the provided guidelines and presented by the student on the day scheduled for the oral examinations.

The final grade takes into account the result of the multiple-choice test and the oral examination or, when the project option is selected, the quality of the work completed and its presentation.

The maximum grade for the written test is of 20 points, while the maximum grade for the oral test or the project is of 12 points. The final evaluation will be the sum of the two grades. If the final grade is greater than 30, it will be assigned a 30L.

Office hours

See the website of Davide Evangelista