Patients in intensive care are monitored very closely through a variety of bedside devices. The overwhelming number of clinically irrelevant alarms from these devices, however, has posed significant challenges to critical care, leading to serious problems ranging from desensitisation of clinical staff to the potential neglect of truly life-threatening conditions. This project seeks to achieve false alarm reduction by incorporating physiological information from multiple signal streams, rather than focusing on a single-signal or single-channel analysis, to accept or reject an issued alarm. This project also seeks to optimize the proposed algorithm for complexity in order to make it readily applicable for real-time operation in clinical settings.

I interned with Quantopian on developing machine learning algorithms for trading, and later decided to apply this knowledge in biomedical applications. I was involved in benchmarking clustering algorithms for protein-protein interaction networks at the Broad Institute, and designed unsupervised techniques to correlate genetic profiles with phenotype for colorectal cancer patients at Stem Cell Institute of the University of Cambridge.