PhD studentship in "Machine Learning Systems: Towards methods for community-integrated autonomous federated AI agents"

Applications to open imminently

Consider helping people in complex environments. In a city, we have events (concerts, matches, festivals), work patterns, travel and traffic effects, roadworks, social needs, criminality, tourism, weather, building, regulation, elections, neighbourhoods etc. All of these interact.

When we build AI tools to e.g. plan for the traffic for an event, it is affected by all the other elements - weather may affect the demand for the event, time of day may affect the interaction with work commute patterns etc. A separate AI tool to predict e.g. the effects of different ticket pricing mechanisms for the event will depend on much the same information - it would be good for the tools to integrate. But how can they?

This is the problem of building integrated autonomous federated AI - where different tools interact to transact both their own data and their own processing to aid other tools in their goals. There are three challenges: first, how to integrate each problem so all are targeted; second, how to build methods that know when, what and how to use information from other sources; third, how this can be effectively and efficiently implemented in interacting computer systems - there are issues of payoffs between compute and communication, real time implementation, and efficient compilation.

Currently, federated learning splits a single machine learning problem over fully compliant communicating sub-processors. It is primarily a systems problem. Multi-agent learning focused on the reinforcement learning of interacting agents. Integrating objectives is a problem of multi-task machine learning. This project integrates these into a single target. It is a machine learning systems problem - how do we build systems that interact robustly and autonomously to solve related but independent problems? Depending on their individual strengths, students will likely end up focusing on different parts of this problem: there is a mathematical aspect, a computer systems aspect, an economic aspect, a social aspect etc. It is likely that any student would not target doing all the above!

Candidate’s profile

An ideal candidate would typically have:

  • a strong degree or higher qualification in a relevant field (e.g. computer science, mathematics, engineering,  physical sciences, economics or any other field where evidence is provided of sufficient computing and mathematical background)
  • solid experience of programming, machine learning methods and ideally deep learning environments (e.g. pytorch) or a computer systems background
  • preferably, good mathematical skills and an understanding of either computer systems architecture or economic systems

Studentship and eligibility

This post is suitable for a home student (e.g. students ordinarily resident in Scotland or the rest of the UK – England, Wales or Northern Ireland, Republic of Ireland, and EU-EEA nationals with Pre/Settled status).

Application information

Applications to open imminently.

Please fill in this form so that we can let you know when you can apply.

We advise eligible and potentially interested students to contact Amos Storkey with a CV and statement of research interest for more information, and an informal discussion of the PhD position.


The School of Informatics is one of the largest in Europe and currently the top Informatics institute in the UK for research power, with 40% of its research outputs considered world-leading (top grade), and almost 50% considered top grade for societal impact. The University of Edinburgh is constantly ranked among the world’s top universities and is a highly international environment with several centres of excellence.

Registration of Interest Form 

Email Prof Amos Storkey 

Professor Amos Storkey Profile     

Centres for Doctoral Training