Decentralized Machine Learning on Edge Devices

University of Birmingham

About the Project

Smartphones, wearables, autonomous vehicles, and other IoT devices record much of our daily lives through their sensors like camera, microphone, accelerometer, etc. Such sensing data is at the core of many applications ranging from health to personal recommenders, smart houses, and smart cities. However, several challenges arise when analyzing such sensing data, for example, it naturally contains sensitive private information, the devices have limited compute, battery power, and communication resources, etc. Recently proposed methods in Federated Learning (and decentralized learning in general) have shown great promise to solve these challenges, by enabling training models without aggregating privacy sensitive data. In this project, we’ll study fundamental, important, and interesting problems around the above setting from both algorithmic and system aspects.

Possible topics include but are not limited to:

i) Adapting pre-trained models on domain shifted data in a collaborative environment,

ii) Building a communication and compute efficient on-device fully decentralized learning system,

iii) Continually learning robust and fair decentralized models,

iv) Building uncertainty-aware decentralized models.

Ideal applicant profile:

• An exceptional master’s degree in Computer Science, Mathematics, Machine Learning or a related technical field.

• Strong background in deep learning, machine learning, and data mining algorithms. Preferably be aware of distributed systems concepts.

• Ability to work independently and within a team.

• Excellent analytical, problem-solving, and software engineering skills with proficiency in Python programming, prior experience implementing machine learning algorithms using well-known frameworks like PyTorch, TensorFlow.

• Aspiration to achieve high-quality research contributions and publications in leading conferences and journals.

• Strong communication skills, including proficiency in written and spoken English and the ability to effectively present technical information to academic and non-expert audiences.

• Interest and ideally experience in design research methods to understand human needs, behaviours, and experiences.

 First or Upper Second Class Honours undergraduate degree and/or postgraduate degree with Distinction (or an international equivalent). We also consider applicants from diverse backgrounds that have provided them with equally rich relevant experience and knowledge. Full-time and part-time study modes are available. We want our PhD student cohorts to reflect our diverse society. UoB is therefore committed to widening the diversity of our PhD student cohorts.

To help us track our recruitment effort, please indicate in your email – cover/motivation letter where ( you saw this job posting.