Abstract:

In many fields one needs to build predictive models for a set of related machine learning tasks. Traditionally these tasks are treated independently and the inference is done separately for each task, which ignores inherent connections among the tasks. Multi-task learning aims to improve the generalization performance by building models for all tasks simultaneously, leveraging inherent relatedness of these tasks. In this talk, we show how multi-task learning can be applied to improve the predictive modeling from electronic medical records (EMR). We consider a novel data-driven framework for densifying EMR to address the challenges from the data sparsity when EMR are used for predictive modeling. By treating the densification of each patient as a learning task, the proposed multi-task learning algorithm simultaneously densifies all patients. As such, the densification of one patient leverages useful information from other patients. Experiments on real clinical data show that the densification can significantly improve the predictive performance.

Biography:

Host:

Dr. Pang-Ning Tan