CHEER: Rich Model Helps Poor Model via Knowledge Infusion.

Abstract

There is a growing interest in applying deep learning (DL) to healthcare, driven by the availability of data with multiple feature channels in <i>rich-data</i> environments (e.g., intensive care units). However, in many other practical situations, we can only access data with much fewer feature channels in a <i>poor-data</i> environments (e.g., at home), which often results in predictive models with poor performance. How can we boost the performance of models learned from such <i>poor-data</i> environment by leveraging knowledge extracted from existing models trained using <i>rich data</i> in a related environment? To address this question, we develop a knowledge infusion framework named CHEER that can succinctly summarize such <i>rich model</i> into transferable representations, which can be incorporated into the <i>poor model</i> to improve its performance. The infused model is analyzed theoretically and evaluated empirically on several datasets. Our empirical results showed that CHEER outperformed baselines by 5.60% to 46.80% in terms of the macro-F1 score on multiple physiological datasets.