- AgriFoRwArdS CDT Annual Conference (2022): Improved Control of Invasive Plant Disease Epidemics Using Partially Observable Markov Decision Processes.
Other Activities and Outputs
- Panel member for the October 2021 AgriFoRwArdS Seminar Series with Charles Nicklin, watch here.
- Panel member for the November 2021 AgriFoRwArdS Seminar Series with Prof Dionysis Bochtis, watch here.
- Chaired the February 2022 AgriFoRwArdS Seminar Series with Dr Mark Ryan, watch here.
- Panel member for the March 2022 AgriFoRwArdS Seminar Series with Ayse Kucukyilmaz, watch here.
Before joining the CDT I was an electronics engineer working for Arm for over a decade. I led some big teams building Neural network Processing Units and I’ve come back to academia to explore how ML technology can be applied in agriculture. There’s a crunch coming up in the next couple of decades between population, climate change and biodiversity and I believe improving agricultural efficiency is a big part of how we can get through that. I’ve also had various placements and consultancy roles including keeping BBC Alba on air in Glasgow, working for Hitachi in Japan, designing control electronics for a telecoms start up and developing a drug delivery device for arthritis patients. I’ll be studying my PhD in the plant sciences department in Cambridge in cooperation with Defra. I also enjoy running, climbing and cooking.
Improved Control of Invasive Plant Disease Epidemics Using Partially Observable Markov Decision Processes
Invasive plant diseases cause damage to agricultural crops and ecosystems. However, growers and policy makers have limited resources for control of any given epidemic and collecting information about epidemic status can be expensive. Partially Observable MDPs (POMDPs) are used to model sequential decision making problems where the agent has only partial information about the situation. This project will explore application of POMDP techniques to plant disease epidemics by building a compartment based epidemic model and measuring the effectiveness of epidemic control against a simple, heuristic-based agent.
Using reinforcement learning to optimise adaptive control of invading plant disease epidemics
Invasive plant diseases threaten agricultural production and natural ecosystems. For example, Xylella Fastidiosa has killed large numbers of olive trees in Italy with future economic impact estimated in the billions of euros. At the same time, the resources to manage these pathogens are limited. Fast responses to outbreaks can minimise the damage. However, this means decisions on how resources are deployed must be made at the start of an outbreak when information about the disease progress and epidemic parameters in a new environment may be limited.
This project aims to explore and evaluate the use of reinforcement learning approaches to optimise control of invasive plant diseases. This will mean building plant disease models and reinforcement learning agents and evaluating the performance of the agents. The work will investigate the simulated cost of control, epidemic outcomes and robustness to uncertainty in the epidemic model. The techniques will aim to be generally applicable but will also be tested with real case studies.