Zhiyao Delivers Spotlight Poster Presentation at ICML 2024

On 25 July 2024, Zhiyao presented his highlighted poster at the 41st International Conference on Machine Learning (ICML 2024), sharing key insights from his recent work on the evaluation of reinforcement learning (RL)-based dynamic treatment regimes (DTRs). His position paper critically examined the current state of offline RL in healthcare, particularly within the context of DTRs.

Zhiyao’s work addressed significant challenges in the field, including inconsistencies in evaluation metrics, the lack of appropriate baselines, and varying RL formulations across existing research. Drawing on a large-scale case study involving over 17,000 experiments with a public Sepsis dataset, he demonstrated that RL algorithm performance can fluctuate considerably depending on evaluation methods and reward design. The presentation sparked thoughtful discussion among conference attendees, highlighting the need for more rigorous evaluation standards and development practices in future RL-based DTR research.

For more details, please refer to our group’s research direction: Treatment Effect Modelling and Recommendation.