Data Study Group - December 2023

Introduction

Stage 1: Precursor Stage (part-time, online)

• The precursor stage will last one week in the run up to the 'event stage' (Monday 27 November – Friday 1 December 2023).
• The maximum time commitment is 2.5 hours a day.
• This includes online workshops, presentations and team building which will prepare participants for the 'event stage'.

Stage 2: Event Stage (full-time, in person)

• The 'event stage' will run at The Alan Turing Institute over one week (Monday 04 – Friday 08 December 2023).
• Group work begins and continues throughout.

Applicants should be able to commit to the duration of the event. The Alan Turing Institute is committed to supporting individual circumstances, please do not hesitate to email [email protected] to discuss any reasonable adjustments. 

Challenges 

Johnson Matthey

Meeting the Challenges of Sustainable Chemical Plant Operations: A Machine Learning Approach for Optimising Renewable Energy Use and Transient Dynamics.

Chemical processes are inherently complex, with multiple process states and paths possible for a given setting. As the industry moves towards for sustainable operations, the use of renewable energy and feedstocks is a key priority. As a consequence of this, chemical plant operations will depart from conventional steady-state operation (slow dynamics) to a more transient basis (fast dynamics) to maximise the use of the resource (e.g. wind power). Such a change introduces process control challenges and the use of large amounts of time-series data to ensure safe and reliable operation.

This challenge is to develop a machine learning-based modelling framework that can accurately predict and optimise plant response in subsequent time steps based on these time-series data.

Our early work in this area, as shown through a feasibility study on a Methanol simulator, has demonstrated that flexible Bayesian regression techniques can effectively optimise nonlinear processes at steady-state.

Ignota Labs

Using machine learning methods to best utilise in-silico toxicity prediction for drug discovery efficacy in new medicines.

Ignota Labs is a leading TechBio company focused on discovery toxicology. We're striving to employ machine learning to predict small molecule drug toxicity early in the drug discovery process, which is currently mired in expensive, slow late-stage wet lab tests. Despite AI's promise, its deployment is still limited in drug discovery due to issues like the vastness of chemical space, lack of standardised methods, limited data availability, challenges in data representation, and evaluation complexities.

We aim to surmount these challenges, venturing into novel algorithms, architectures, and methods to optimise predictive accuracy in drug toxicity. While we encourage exploration of deep learning methods like GNNs, multi-task learning, transfer learning, and self-supervised learning, we also acknowledge the successful use of tabular representations and ensemble approaches. The challenge offers an opportunity to contribute to the field of in silico toxicity prediction within drug discovery, which aligns with the grand challenge of Health. We invite you to join us in accelerating scientific understanding and improving human health through data-driven innovation in AI and statistical science.

UK Centre for Ecology & Hydrology

Utilising machine-learning for image classification to produce biodiversity metrics for insect populations across the globe.

Evidence is mounting of declines in wildlife across the globe but data on insects are very limited due to a lack of baseline knowledge (e.g. an estimated 80% of species are yet to be described). Automated sensors, deep learning and computer vision can be part of the solution to deliver more standardised monitoring of insects and systems are beginning to be deployed in more places across the world. This DSG challenge will help reach the potential of this nascent technology to deliver data where it is most needed (e.g. biodiversity hotspots in tropical ecosystems).  
  
This challenge will build upon data from the AMI system and existing image detection and classification models to:  

• Produce biodiversity metrics and data visualisations that engage and enthuse specialists (e.g. entomologists, researchers) and non-specialists (e.g. the general public, businesses).
• Tackle machine learning problems such as identification of insects on complex backgrounds, and extend algorithms to classify more species.