Welcome! This is the home page of the "Advanced Machine Learning" course, taking place in the winter semester of 2024/2025.
The course is part of the Master's Degree in Data Science -- Sapienza University of Rome -- organized jointly by the departments of Computer Science (DI), Information and Automation Engineering (DIAG) and Information, Electronics and Telecommunication Engineering (DIET), and Statistics (DSS).
Important: subscribe to the course mailing list to receive slides, course material, assignments and news.
Click on this link.
Note: join the mailing list with your institutional email.
If you did not receive your institutional address yet due to specific situations, please request access to the group but also send an email detailing i. the circumstance; ii. the proof of your acceptance; and iii. the proof of your identity. Requests from non-institutional addresses without those cannot be accepted.
Classes starts on September, 23rd (in agreement with the Data Science class schedule -- link).
The course takes place @:
Room A2 (Via Ariosto, 25) on Wednesdays 8:00-10:00
Room A2 (Via Ariosto, 25) on Fridays 16:00-19:00
Please also refer to the Data Science first-semester lecture times for the class schedule.
Classes are in presence, however we make efforts to support remote participation (details distributed via the mailing lists).
Please consider Sapienza's recommendations on the attendance of lectures.
Prof. Fabio Galasso, web-page, email: galasso remove_this @di DOT uniroma1 DOT it
The course will present advanced concepts of machine learning and their application in computer vision via deep neural network (DNN) models. It will include theory and practical coding, as well as a final hands-on project.
In a first part of the course, I will introduce state-of-the-art DNN models for classification, showing how to estimate which objects are within an image. I will then showcase regression, as applied to detection (where the objects are in the image), pose estimation (whether people stand, sit or crunch) and re-identification (estimating a unique vector representation for each person). I will further discuss DNNs for multi-task objectives (joint detection, pose estimation, re-identification, segmentation, depth estimation etc). This first part will include DNNs which apply to video sequences, by leveraging memory (e.g. LSTMs) or attention (Transformers), extending to long-term modelling with State-Space-Models (SSMs) and to modelling sequences of structured data with Graph Convolutional Networks.
In a second part of the course, I will discuss generalization and the effective use of labelled and unlabelled data for learning. Further to transfer learning (how pre-trained models may be deployed for other tasks), I will discuss multi-modal (with different sensor modalities such as depth or thermal cameras, with text and sound) and self-supervision (e.g. training the DNN model by leveraging time and causality in time) to auto-annotate large amounts of data. Also, I will present domain adaptation (e.g. apply daytime-detectors for night vision) and meta-learning, a most recent framework to learn how to learn a task, e.g. online or from little available data, and continual learning. I will introduce novel machine learning trends such as hyperbolic neural networks, showcasing their use for estimating uncertainty and enabling the model to request the intervention of humans and active learning. Finally, I will discuss generative AI techniques such as VAE, GANs and diffusion models, and their applications to tasks such as anomaly detection while estimating the model uncertainty.