Research Associate or Research Fellow in Federated Learning for Healthcare
Medicine is undergoing a data revolution, with AI being the engine of change. To unlock this potential, AI algorithms need to learn from very large datasets scattered across multiple hospitals and multiple countries, all in a privacy-preserving and transparent manner. New approaches to algorithmic learning based on recently developed concepts of Federated Learning and Differential Privacy provide the mathematical framework to enable this vision. This research programme will develop a new set of algorithms and associated software platforms to enable federated learning at scale, respectful of the hospital IT infrastructure, with privacy guarantees, and with full traceability of actions, thus ensuring the trust of our patients. Such a platform would enable unprecedented access to data at scale, necessary to solve many pressing healthcare problems.
The London Medical Imaging & AI Centre for Value-Based Healthcare is a consortium of academic, NHS and industry partners led by King's and based at St Thomas' Hospital. Our diverse research teams are training sophisticated artificial intelligence algorithms from a vast wealth of NHS medical images and patient pathway data to create new healthcare tools. For patients, these will provide faster diagnosis, personalised therapies and effective screening across a range of conditions and procedures. Through a focus on our experience in value-based healthcare we are examining how AI can be used to optimise triage and target resources to deliver significant financial savings for the NHS and healthcare systems overall. The centre has been established as part of the UK Government's Industrial Strategy Challenge Fund, delivered through UK Research and Innovation.
The purpose of this role:
This is an exciting opportunity for an enthusiastic privacy-minded deep learning researcher to push the boundaries of Federated Learning and Differential Privacy applied to healthcare data.
This role will build of the AI4VBHC centre to deliver on two major deep learning research challenges, namely Data Governance, and Patient Privacy. More specifically, this post will develop new algorithms and associated software stack to enable Federated Learning at scale, in a way that is respectful of the hospital IT infrastructure and constraints. Furthermore, the post holder will also develop AI algorithmic privacy guarantees under the framework of Differential Privacy to ensure that any user would be able to utilise the federated learning network and the distributed raw patient data without endangering patient privacy. These algorithms shall be integrated and deployed into the AI4VBHC centre infrastructure as a proof-of-concept, demonstrating that federated and privacy preserving learning can safely learn from patient data scattered across multiple hospitals, depicted in the figure below.
This is a full time fixed-term contract for two years.
The successful applicant will be responsible for developing federated learning models for imaging and non-imaging data across multiple hospitals, optimising the federated compute plan to the realities of hospital IT constraints, and develop differential privacy approaches that preserve subject privacy. Models and algorithms shall build on the MONAI (monai.io) and Pytorch for the machine learning stack, and on Substra (substra.ai) for the deep learning stack, continuing the AI centre's open source tradition. The applicant shall also engage and further develop research relationships with key AI-centre-related hospital partners (e.g. KCH, GSTT, and UCLH), and major industrial partners (e.g. NVIDIA, and OWKIN).
The applicant should ideally have some knowledge and experience of:
- Medical image analysis
- Deep Learning
- Data science
- Statistical modelling
The position would appeal to a candidate with strong software development skills, including:
- Classic machine learning and deep learning packages (e.g. pytorch,…)
- Data science and statistical modelling packages (Pandas, SciPy/Statsmodels)
- Container orchestration tools (Kubernetes, Docker)
An interest in federated learning and differential privacy applications is important, but previous experience is not essential. Previous experience with large-scale computing services would be useful but not essential. Strong communication skills are required to work with researchers from other disciplines, such as clinical end-users, and industry collaborators.
The candidate is also expected to:
- work in close collaboration with KCH, GSTT, and UCLH colleagues
- work in close collaboration with our key industrial partners, NVIDIA and OWKIN
- work towards a common software stack
Experience working with interdisciplinary teams of engineers and clinicians will be valued. A strongly independent applicant is required who will need to work well with inter-disciplinary teams.
The above list of responsibilities may not be exhaustive, and the post holder will be required to undertake such tasks and responsibilities as may reasonably be expected within the scope and grading of the post.