ANC Workshop - Adam Jelley, Matthias Hennig

Tuesday, 12th December 2023

Contrastive Meta-Learning for Partially Observable Few-Shot Learning - Adam Jelley

Abstract: Many contrastive and meta-learning approaches learn representations by identifying common features in multiple views. However, the formalism for these approaches generally assumes features to be shared across views to be captured coherently. Our work (published at ICLR earlier this year, https://arxiv.org/abs/2301.13136) considers the problem of learning a unified representation from partial observations, where useful features may be present in only some of the views. We approach this problem with a probabilistic formalism that enables the mapping of views to representations with different levels of uncertainty in different components; these views can then be integrated with one another through marginalisation over that uncertainty. In this talk I will provide an overview of our approach, named Partial Observation Experts Modelling (POEM), that enables us to meta-learn consistent representations from partial observations. 

824 issues, or how to open source research software - Matthias Hennig

Abstract: In 2019 we created the SpikeInterface project as we needed better software and algorithms to analyse the data from new high-density electrophysiological data acquisition systems. These devices allow recording from many neurons in the brain, but produce large and complex data sets - interpreting these first requires extensive analysis of the raw data. The project started with a small grant from the Wellcome trust and a team of five, and has since grown significantly. We now have hundreds of users and over 50 contributors to the code base. I will talk about the development process and the challenges of managing such an open-source project. 

Event type: Workshop

Date: Tuesday, 12th December

Time: 11:00

Location: G.07

Speaker(s): Adam Jelley, Matthias Hennig

Chair/Host: Amos Storkey