Munib and Ghadder Present at ICLR Practical ML for Developing Countries Workshop

On 5 May 2023, the Practical Machine Learning for Developing Countries (PML4DC) workshop was held, marking its fourth consecutive year at the International Conference on Learning Representations (ICLR). The event brought together researchers, practitioners, and policymakers from academia, industry, and government to discuss the challenges and opportunities in deploying machine learning (ML) solutions tailored for resource-limited settings. 

Munib presented his innovative work on predicting mortality and recurrent heart attacks in intensive care units (ICUs), particularly addressing the noisy, sparse, and irregular time-series data characteristic of low-resource hospital settings. Utilising retrospective cohorts from eICU and MIMIC-IV databases, Munib developed a pseudo-dynamic, interpretable machine learning framework that reliably predicts clinical events up to 24 hours in advance. 

Ghadeer showcased her research on leveraging deep generative models to overcome data limitations typical in low and middle-income countries. Using a small dataset from Ho Chi Minh City Hospital for Tropical Diseases in Vietnam, Ghadeer generated realistic synthetic electronic health records (EHRs) and demonstrated their effectiveness in developing predictive models for hospital-acquired infections (HAIs). Her work illustrated the potential of synthetic data in enhancing diagnostic accuracy, significantly outperforming traditional oversampling methods, thus enabling healthcare data owners in resource-constrained environments to build robust clinical decision support systems.

Our lab is committed to developing innovative ML tools for developing countries, aiming to facilitate equality in healthcare access and quality through technology-driven solutions.