Turing Theory and Methodology Challenges Week
We are excited to invite the Turing community to participate in the challenges which are going to take place during the second week (5-9 September 2022) as part of the Theory and Methodology Challenges Fortnight (TMCF) event on Accelerating generative models and nonconvex optimisation. The event will be in person at Turing HQ based in the British Library, London.
This event will consist of methodology and coding challenges associated with the problems described below. The event is open to any level of expertise, but please note that there are limited places available. The organisers have a small fund to help with travel and accommodation if required, so please add a note in your application if you require this funding.
To submit an application please send a short description of your research and a brief CV to me. The deadline for applications is 12 August 2022. By sending the application you agree to attend the full week (5-9 September 2022).
There will be two challenges and two teams will be formed to work on the problems described below. Please indicate which project you are interested in when showing an interest.
Challenge 1: Score-based generative models: Implementation, optimisation, generalisation
This challenge will focus on the new emerging class of score-based generative models. The team will work on their implementation on a toy example, then investigate various issues related to optimisation and generalisation. The particular focus will be on (i) testing new and various optimisers, (ii) exploring why these models generalise.
Challenge 2: Physics-informed generative models
While generative models display excellent data generation capabilities for datasets like images, audio, and video, it is less clear how to use them for generating data that have physical constraints, e.g., simulating a fluid flow or generating data informed by a PDE. This challenge will be focusing on implementing a class of generative models to solve PDEs. A crash course on Tensorflow will be provided.