Continual learning: the State of the Art
While machine learning systems reached impressive, often super-human performance in a broad variety of tasks, they typically rely on the following recipe: training a big deep learning model on a large amount of data, for a sufficient amount of time. On the other hand, humans have the innate ability to learn new concepts on the fly, and throughout their lifespan. In this regard, machine learning systems are still severely under-performing: learning new concepts quickly is a difficult process, and it comes at the expense of forgetting previously learned information.
The overall goal of continual learning is to overcome such difficulties. In this seminar, we will cover the state of the art of continual learning research, from the initial efforts to the recent trends that aim at enabling neural networks with the ability to learn new concepts, tasks, domains, etc. effectively.
About the speaker
Riccardo Volpi is a research scientist and project lead at Naver Labs Europe, the biggest industrial research lab in artificial intelligence in France.
He's broadly passionate about machine learning, and in particular about its applications to computer vision problems. His main research interests at the moment are continual learning, domain adaptation/generalization, model robustness and uncertainty.
Before joining NLE, he was a Ph.D. student (2015-2018) and postdoc (2018-2019) at Istituto Italiano di Tecnologia. During his studies, he also spent some time at Stanford Vision and Learning Lab and University College Cork.
The participation is free but it is mandatory to register online due to Covid-19 restrictions that limit the seats in the room.
In compliance with Covid-19 restrictions, the access to the room is allowed only with the Digital Covid Certificate (Green Pass) and face masks must be worn at all times.