Student Casual - Research Associate - Computer Science - EPS - Grade 6 - 906594

Position Details

School or Department: Computer Science

Location: University of Birmingham, Edgbaston, Birmingham UK

Grade: 6

Positions available: 1

Hourly rate: £16.69 + holiday entitlement

Casual contract from 1/7/25 - 28/9/25

Closing date: 30/5/25

This vacancy is only open for University of Birmingham students. Please be aware that the vacancy may be taken down early dependent on the number of applications received.

Our offer to you

People are at the heart of what we are and do.

The University of Birmingham is proud to have been a part of the City of Birmingham and the wider region for over 100 years, and we are equally proud to be recognised as a leading global university.  We want to attract talented people from across the city and beyond, support them to succeed, and celebrate their success.

We believe there is no such thing as a typical member of staff and that diversity is a source of strength that underpins the exchange of ideas, innovation, and debate.  We warmly welcome people from all backgrounds and are committed to fostering an inclusive environment where diversity is at the heart of who and what we are, and how we work.

The University is situated in leafy Edgbaston and there are excellent transport links to our beautiful campus, including main bus routes and a train station on site.  On campus we have a state-of-the-art sports centre with pool, shops, places to eat and drink, our own art gallery, museum and botanical gardens.

Find out more about the benefits of working for the University of Birmingham 

Background

To assist in research on Federated Test Time Adaptation, when using a pre-trained model on domain-shifted data, the model often underperforms. The most popular way to mitigate this is to adapt or fine-tune the model parameters given a small amount of test data. Test Time Adaptation (TTA) considers model adaptation with unlabelled test data.

However, TTA methods do not work well when the test data is small and imbalanced [1]. In this project, we will implement a Federated Learning solution to this problem, where multiple devices or clients will collaborate to build a better-adapted model.

[1] Michal Danilowski, Soumyajit Chatterjee, and Abhirup Ghosh. Botta: Benchmarking on-device test time adaptation. arXiv preprint arXiv:2504.10149, 2025.

Role Summary

• Contribute to ideation towards defining a concrete problem, solutions, and experiments.
• Contribute to writing papers and preparing presentations as appropriate.

Main Duties

• Implement baselines using different Federated Learning solution combining with TTA methods.
• Experiment using commonly used vision (cifar-10c) and audio datasets.
• Theoretically analysis and experimenting on system metrics such as memory usage will be a plus.
• Present research outputs, including drafting academic publications or parts thereof, for example at seminars and as posters
• Promotes equality and values diversity acting as a role model and fostering an inclusive working culture

Person Specification 

• At least BSc Degree or equivalent in Computer Science and Artificial Intelligence. 
• Practical experience of training and using a deep neural network using Pytorch or Tensorflow.
• Experience with Federated Learning is preferred.
• Ability to analyse information and communicate effectively
• Ability to access and organise resources successfully
• Knowledge of the protected characteristics of the Equality Act 2010, and how to actively ensure in day to day activity in own area that those with protected characteristics are treated equally and fairly