Francesco Pelosin

I'm a Research Scientist in Computer Vision and Machine Learning focusing on continual learning, machine unlearning, and transfer learning. My work aims to develop algorithms that can learn continuously from streams of data while maintaining good performance on previously learned tasks - a key challenge known as catastrophic forgetting. I previously conducted research at the Computer Vision Center (CVC) in Barcelona working with Joost van de Weijer's team on continual learning for Vision Transformers. My doctoral research at Ca' Foscari University of Venice with Professor Alessandro Torsello focused on rehearsal-based continual learning methods and developing more efficient ways to store and reuse past knowledge. My current work explores novel approaches like Dynamic Label Injection for handling imbalanced data, parameter isolation methods for incremental learning (MIND), and distance-based machine unlearning (DUCK). I'm particularly interested in understanding the fundamental trade-offs in continual learning systems, such as the relationship between memory capacity and performance, and developing simple yet effective baseline approaches that can serve as strong foundations for the field.

I have published papers in top venues including CVPR, AAAI, and Pattern Recognition. My research aims to push the boundaries of how AI systems can learn and adapt over time while being computationally efficient.

[ this stuff was generated automatically but I agree with almost everything ]

Publications

Dynamic Label Injection for Imbalanced Industrial Defect Segmentation

Dynamic Label Injection for Imbalanced Industrial Defect Segmentation

Emanuele Caruso, Francesco Pelosin, Alessandro Simoni, Marco Boschetti

Workshop on Vision-based Industrial Inspection (ECCV 2024)

MIND: Multi-Task Incremental Network Distillation

MIND: Multi-Task Incremental Network Distillation

Jacopo Bonato, Francesco Pelosin, Luigi Sabetta, Alessandro Nicolosi

AAAI 2023

DUCK: Distance-based Unlearning via Centroid Kinematics

DUCK: Distance-based Unlearning via Centroid Kinematics

Marco Cotogni, Jacopo Bonato, Luigi Sabetta, Francesco Pelosin, Alessandro Nicolosi

arXiv.org 2023

Dissecting Continual Learning a Structural and Data Analysis

Dissecting Continual Learning a Structural and Data Analysis

Francesco Pelosin

arXiv.org 2023

Simpler is Better: off-the-shelf Continual Learning Through Pretrained Backbones

Simpler is Better: off-the-shelf Continual Learning Through Pretrained Backbones

Francesco Pelosin

Workshop T4V (CVPR 2022)

Towards Exemplar-Free Continual Learning in Vision Transformers: an Account of Attention, Functional and Weight Regularization

Towards Exemplar-Free Continual Learning in Vision Transformers: an Account of Attention, Functional and Weight Regularization

Francesco Pelosin, Saurav Jha, A. Torsello, B. Raducanu, Joost van de Weijer

Workshop CLVISION (CVPR 2022)

Smaller is Better: An Analysis of Instance Quantity/Quality Trade-off in Rehearsal-based Continual Learning

Smaller is Better: An Analysis of Instance Quantity/Quality Trade-off in Rehearsal-based Continual Learning

Francesco Pelosin, A. Torsello

IEEE International Joint Conference on Neural Network 2021

Unsupervised semantic discovery through visual patterns detection

Francesco Pelosin, A. Gasparetto, A. Albarelli, A. Torsello

International Workshop on Structural and Syntactic Pattern Recognition 2021

Separating Structure from Noise in Large Graphs Using the Regularity Lemma

Separating Structure from Noise in Large Graphs Using the Regularity Lemma

M. Fiorucci, Francesco Pelosin, M. Pelillo

Pattern Recognition 2019

Graph Compression Using The Regularity Method

Graph Compression Using The Regularity Method

Francesco Pelosin

arXiv.org 2018

Incremental Automatic Image Annotation

Incremental Automatic Image Annotation

L. Sabetta, Francesco Pelosin, Giulia Denevi, Alessandro Nicolosi

Ital-IA 2023

8th International Conference on Network Analysis NET 2018

V. Boginski, Ernesto Estrada, Lecturer Mikhail Isaev, M. Pelillo, M. Ravetti, Angelo Sifaleras, Theodore Trafalis, Mikhail Batsyn, Mikhail Chernoskutov, Francesco Pelosin, M. Fiorucci, G. Gradoselskaya, Ilia Karpov, T. Shcheglova