thesis
The thesis projects we propose are based on researching and testing new algorithms or methods to tackle issues in existing machine learning and deep learning scenarios. The world of computer vision is the core of our research.
In order to obtain satisfying results when working on a thesis, the following requirements must be met before applying:
- the student must have high programming skills and possess basic knowledge of machine learning theory and its practical applications
- it is mandatory to have attended one of the courses we provide (visit Teaching for more info).
The thesis will not begin until the student has a maximum of two exams to complete his studies.
If you meet the previous requirements and decide to ask us for a thesis, you will first have to pass a test in which we will evaluate:
- your programming skills in Python, especially when using the PyTorch library
- your mastery of machine learning and deep learning theory
- any knowledge on a specific topic, agreed in advance.
When you send us your application, be sure to include your updated CV, containing the following information:
- list of exams taken during your Master’s degree program with marks, together with your weighted average
- list of exams taken during the Bachelor’s degree program with marks and your graduation grade
- any past experience in the field of machine learning and/or computer vision.
If you are interested in our research and have the required skills, we will be glad to here from you!
- Scene graph generation from egocentric videos. Gabriele Goletto, Giuseppe Averta.
- A graph-based unified architecture for Egocentric Video Understanding. Simone Alberto Peirone, Giuseppe Averta.
- Pass me the thing: learning to understand tools functionalities in the wild. Francesca Pistilli, Giuseppe Averta.
- 3 Theses on Transferability, Robustness, and Generalizability of foundational segmentation models for road scenes. Shyam Nandan Rai, Carlo Masone.
- 2 Theses on Semantic Segmentation on Tiny Devices (In collaboration with ST Microelectronics). Shyam Nandan Rai, Carlo Masone
- Tiny anomaly segmentation in urban road scenes. Shyam Nandan Rai, Carlo Masone.
- Generalizing Deep Reinforcement Learning for multi–DoF Robotic Grasping Across Objects. Raffaello Camoriano
- Machine Learning for Safe Robotic Planning. Raffaello Camoriano
- Re-Calibrating Out-of-distribution losses for dense prediction tasks. Shyam Nandan Rai
- Approaching Visual Geolocalization as a Regression Problem. Gabriele Berton
- Self-Supervised Learning for Visual Geolocalization. Gabriele Berton, Gabriele Trivigno
- Learning to generalize across end-effectors for Robotics Dexterous Grasping. Antonio Alliegro
- Domain Robustness against Adversarial Attacks. Tatiana Tommasi
- Person Reidentification Across Domains. Tatiana Tommasi
- Exploring pretrained representations for Out Of Distribution detection. Francesco Cappio Borlino
- Closing the simulation-to-reality gap in reinforcement learning for soft robots. Gabriele Tiboni
- Domain Adaptation for Visual Geolocalization. Gabriele Berton, Gabriele Trivigno
- World-wide Image Geolocalization. Gabriele Berton, Gabriele Trivigno
- Source Free Video Action Recognition. Chiara Plizzari, Mirco Planamente
- EGO-T^3:Test-Time Training for Egocentric videos. Chiara Plizzari, Mirco Planamente
- Pixel-by-pixel domain alignment across multiple sources in semantic segmentation (journal). Antonio Tavera
- Fairness in visual recognition through domain-agnostic learning. Silvia Bucci
- Speeding up Heterogeneous Federated Learning with Sequentially Trained Superclients (journal). Debora Caldarola
- Relative geometric transformations for domain adaptation. Andrea Bottino, Mirco Planamente
- Relative Norm Alignment for Audio-Visual Action Recognition (journal). Chiara Plizzari, Mirco Planamente
- Open World Semantic segmentation. Dario Fontanel, Fabio Cermelli
- Incremental and open world object detection (instance semantic segmentation). Dario Fontanel, Fabio Cermelli
- Learning to Grasp. Antonio Alliegro
- Contrastive learning for open-set domain generalization (journal). Francesco Cappio Borlin, Silvia Bucci
- Federated Visual Geo-localization. Debora Caldarola, Gabriele Berton
- On the use of Domain Adaptation for Machine Learning Fairness. Silvia Bucci
- Gabriele Moreno Berton. CNN-based method with self-supervision for visual place recognition. (2020)
- Valerio Paolicelli. Deep learning for visual place recognition: Large scale software and self-supervised approach to geo-localize a given photo. (2020)
- Luca Robbiano. Deep Domain Adaptation through Inter-modal Self-supervision. (2020)
- Francesco Cappio Borlino. Visual object detection across different domains by solving self supervised tasks. (2019)
- Fabio Cermelli. The RGB-D Triathlon Challenge: Towards Agile Visual Toolboxes for Robots. (2018)
For all the theses published in this group, take a look here [1][2].
PhD
Research topics of PhD programs are often extensions of the thesis work, previously started inside the laboratory.
Otherwise, if you wish to submit your application for joining us, feel free to send us an email with the following information:
- your updated CV
- your Master’s thesis
- references from two tutors or supervisors.
Please refer to the following PhD program pages for the specific requirements and deadlines:
- ScuDO – Politecnico di Torino: requirements and call for applications
- National PhD program in A.I.
- Ellis PhD
We hope to hear from you soon!