Course 2025-2026 a.y.

• Theoretical foundations of artificial neural networks and backpropagation

• Regularisation techniques (e.g., weight decay, dropout, early stopping)

• Initialisation strategies and their impact on training dynamics

• Overparameterization, bias-variance trade-off, and the double descent phenomenon

• Convolutional Neural Networks (CNNs) and their application to biomedical image and video data

• Recurrent Neural Networks (RNNs) and their applications to temporal and sequential biomedical data

• Transformer architectures and attention mechanisms in life sciences

• Graph Neural Networks (GNNs) for structured biomedical data

• Unsupervised learning with deep models: Autoencoders, Variational Autoencoders (VAEs), and Generative Adversarial Networks (GANs)

• Hands-on labs with real-world biomedical datasets

• Current challenges and future perspectives in AI for biomedicine

• Describe the theoretical foundations of artificial neural networks and their learning dynamics

• Explain the role of regularisation techniques and weight initialisation in deep model training

• Illustrate the effects of overparameterisation, including the bias-variance trade-off and the double descent phenomenon

• Distinguish between major deep learning architectures, such as CNNs, RNNs, transformers, and graph neural networks

• Summarise the principles and use cases of generative models

• Identify appropriate architectures for different types of biomedical data (e.g., images, sequences, graphs) and apply them to analyse and interpret the data

• Implement neural networks tailored to specific biomedical problems using Python and PyTorch, and evaluate their performance

• Analyse the training behaviour of neural networks and interpret performance metrics to guide model refinement

• Compare different deep learning approaches (e.g., CNNs, RNNs, Transformers, GNNs, VAEs, GANs) in terms of suitability for a given biomedical task

• Recognise current limitations and emerging research challenges in AI applied to biomedicine

• Collaborate effectively in small teams to develop and present solutions to real-world biomedical challenges

• Communicate methodological choices and results clearly and effectively through oral presentations and written technical reports

• Lectures
• Guest speaker's talks (in class or in distance)
• Practical Exercises
• Collaborative Works / Assignments

• Face-to-face lectures will cover the theoretical, methodological, and computational foundations of deep learning and generative models, with a particular focus on applications in biomedical domains. Key concepts will be illustrated through case examples and visual explanations.
• Guest speakers will present real-world applications of AI in biomedicine, offering students insight into current trends and challenges in the field.
• Practical exercises will be carried out through hands-on sessions, where students will implement deep learning models and training strategies using Python and commonly adopted machine learning libraries (e.g., PyTorch). Students will work directly on biomedical datasets to apply and consolidate the concepts introduced in lectures.
• Collaborative assignments will involve team-based development of a project, where students will identify a biomedical problem, design and implement deep learning solutions, and produce a final presentation. This activity aims to foster teamwork, project planning, critical thinking, and scientific communication skills.

Students who attend at least 75% of the course activities are considered attending students. They may pass the written exam either through two partial exams or through a single comprehensive final exam. In addition to the written exam, attending students will participate in a group project related to the course content. The project involves designing and implementing deep learning models for biomedical data, culminating in a final group presentation. The project aims to evaluate the "applying knowledge and understanding' learning objectives. The final grade will be determined as the average of the written exam (50%) and the group project (50%). A minimum passing grade of 16 out of 31 must be achieved in both parts. The final grade, rounded up, must be at least 18 to pass the course.

Students who do not meet the minimum attendance requirement (75%) must take the general final exam. In place of the group project, non-attending students will do an individual project, focused on a topic proposed or approved by the instructors. The project involves the independent development and implementation of deep learning models for the chosen topic, accompanied by a written final report.

• Bishop, Chris, and Bishop, Hugh. Deep Learning: Foundations and Concepts, Springer, 2024
• Prince, Simon J.D. Understanding Deep Learning, MIT Press, 2023
• Scardapane, Simone. Alice's Adventures in a Differentiable Wonderland (Volume I, A Tour of the Land), 2024