A multidisciplinary approach to safer reinforcement learning

Here Yan tells us about her work in healthcare, with a focus on her research investigating how to improve the safety of reinforcement learning when used to treat sepsis.

What does safety assurance mean in healthcare?

Safety in the healthcare context is patient safety. Whilst we are developing AI/ML systems for diagnosing or treating patients and they have very promising performance, we still need to make sure that they don’t compromise patient safety as well. For example, does the system introduce new hazards, and is the risk associated with the new hazard acceptable, can we mitigate it further? 

Safer sepsis treatment

Two interesting use cases I have worked on are recommending doses for sepsis treatment and predicting weaning readiness for patients on mechanical ventilation in ICU.

Currently, there is no agreed optimal treatment for sepsis patients, especially regarding the doses of vasopressor and fluid. So it is possible to use reinforcement learning - a type of machine learning - to learn the optimal treatment based on the episodes where doctors treated sepsis patients in the hospital.

A unique community

The University of York stands apart from other universities. I think one of the strengths of the University is its expertise in safety assurance. If you are doing such research, this is the best place to come.

Find out more

• Find out more about our demonstrator project on sepsis treatment
• Read the AAIP impact report