Curriculum
Computer Science for Societal Challenges and Innovation, XXXIX series
Grant sponsor
MUR PNRR DM117 and Cosmo Artificial Intelligence Ai LTD
Supervisor

Lamberto Ballan
Co-supervisor
s
TBD
Contact
luca.parolari@studenti.unipd.it

Project description
Thanks to diffusion of deep learning techniques and the availability of high-quality annotated datasets, in the last years the scientific community made outstanding advancements in the field of object detection and tracking.
However, traditional approaches often depend heavily on extensive annotations for training, resulting in expensive and time-consuming data collection processes that restrict the utilization of large-scale datasets. In a dynamic video environment, this problem significantly arises and is further stressed in the medical domain, where data are private, limited, difficult to collect, noisy and require high-level expertise to be properly categorized.
For this reason, exploring incremental and transfer learning methods becomes valuable in retaining previously acquired knowledge and transferring it across diverse domains and settings, such as noisy scenarios. Additionally, limited supervision learning can effectively reduce costs and accelerate data collection while still being relevant for human comprehension. In fact, limited supervision learning is much more common than supervised.