Elisa Ricci

[intermediate] Continual and Adaptive Learning in Computer Vision

Summary

Deep networks have revolutionized the computer vision field. Despite their power, challenges still exist in creating deep models for real-world applications that can continuously learn new concepts and adapt to diverse conditions. Domain adaptation tackles these challenges by addressing data distribution shifts between domains, while continual learning addresses the issue of retaining past knowledge during sequential learning tasks. In my course, I will explore the evolution of methods and trends in domain adaptation and continual learning for visual recognition. Emphasizing recent advancements in adapting large vision and language models, the course will provide insights into the latest techniques developed in this research field.

Syllabus

• Introduction to domain adaptation: basic concepts and challenges
• Techniques for addressing domain shift
• Introduction to continual learning
• Recent advances in domain adaptation and continual learning for visual recognition, with a focus on large vision and language models

References

L. Wang, X. Zhang, H.Su, J. Zhu, A comprehensive survey of continual learning: Theory, method and application. IEEE Trans. on PAMI, 2024.

G. Csurka, Domain adaptation for visual applications: A comprehensive survey. arXiv:1702.05374, 2017.

S. Bose, A. Jha, E. Fini, M. Singha, E. Ricci, B. Banerjee, StyLIP: Multi-Scale Style-Conditioned Prompt Learning for CLIP-based Domain Generalization. WACV 2024.

G. Zara, S. Roy, P. Rota, E. Ricci, AutoLabel: CLIP-based framework for Open-Set Video Domain Adaptation. CVPR 2023.

C. Saltori, A. Osep, E. Ricci, L. Leal-Taixé, Walking Your LiDOG: A Journey through Multiple Domains for LiDAR Semantic Segmentation. ICCV 2023.

G. Zara, A. Conti, S. Roy, S. Lathuilière, P. Rota, E. Ricci, The Unreasonable Effectiveness of Large Language-Vision Models for Source-free Video Domain Adaptation. ICCV 2023.

Z. Kang, E. Fini, M. Nabi, E. Ricci, K. Alahari: A soft nearest-neighbor framework for continual semi-supervised learning. ICCV 2023.

T. De Min, M. Mancini, K. Alahari, X. Alameda-Pineda, E. Ricci: On the Effectiveness of LayerNorm Tuning for Continual Learning in Vision Transformers. ICCV Workshops 2023.

Pre-requisites

Basic machine and deep learning and computer vision knowledge.

Short bio

Elisa Ricci is an Associate Professor at the Department of Information Engineering and Computer Science at University of Trento and the Head of the Deep Visual Learning Unit at Fondazione Bruno Kessler (FBK). Elisa received her MSc (2004) and PhD degree (2008) from the University of Perugia. Previously, she was an Assistant Professor at the University of Perugia and a researcher at the Idiap Research Institute and FBK. She has been a visiting researcher at the Swiss Federal Institute of Technology and the University of Bristol. Elisa has co-authored more than 150 scientific publications and she regularly publishes in top-tier journals and conferences in computer vision. She participated to several national and international projects. Currently, she is the local coordinator of the EU H2020 project SPRING, where she leads research activities in multi-modal human behavior analysis for human robot interactions, and the technical coordinator of the EU ISFP project PRECRISIS, focusing on developing AI solutions for enhancing the security of public spaces. Her research lies at the intersection of computer vision, deep learning and robotics perception.