ANC Workshop - Massimiliano Patacchiola

Speaker: Massimiliano Patacchiola

Title: Learning more by learning less: machine learning with less data and less supervision.

Abstract:

One major bottleneck of modern deep learning systems is the need of large labelled datasets which are expensive to gather and annotate. Recent trends have approached this problem by defining the meta-learning framework for the few-shot setting (less data) and by exploiting a self-supervisory signal which can be used in absence of labels (less supervision). In this talk I will present our latest work in this direction. In particular, I will provide an overview of our recent NeurIPS submissions focusing on Bayesian meta-learning via deep kernels and Gaussian Processes, self-supervised learning with relational networks, and continual few-shot learning.

Bio:  Massimiliano Patacchiola is a postdoctoral researcher in the Bayesian and Neural Systems group at the University of Edinburgh under the supervision of prof. Amos Storkey. His research project is funded by Huawei and it aims at making machine learning more efficient by exploiting Bayesian methods and more broadly deep learning paradigms. 

Before joining the University of Edinburgh, Massimiliano was an inter at Snapchat where he worked on deep generative models, and a PhD student in robotics and machine learning at the University of Plymouth.