We look forward to seeing you online at 11am today.
Abstract: Despite the extensive literature on out-of-distribution detection, some relevant settings for real-world applications have been overlooked. The talk will focus on them and on the strategies to use foundational models for semantic novelty detection on 2D (images) and 3D (point-clouds) data.
Bio: Tatiana Tommasi is associate professor in the Control and Computer Engineering department at the Polytechnic of Turin and directs the ELLIS Unit in Turin. She received her PhD from EPFL Lausanne in 2013 and subsequently undertook post-doctoral roles in both Belgium and the USA. Prior to her current position, she was assistant professor at Sapienza University in Rome and senior researcher at the Italian Institute of Technology. Her expertise lies in the development of theoretically grounded algorithms for automatic learning from images and 3D data, with medical and robotics applications. She pioneered the field of transfer learning in computer vision and possesses extensive experience in areas such as domain adaptation, generalization, multimodal learning, and open-set learning. Prof. Tommasi is Associate Editor of the IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE TPAMI) and IEEE Transactions on Emerging Topics in Computing (IEEE TETC).