Learning Multi-Modal Electronic Health Records for Inter-Modal Correspondence and Phenotypes PROJECT TITLE : Learning Inter-Modal Correspondence and Phenotypes from Multi-Modal Electronic Health Records ABSTRACT: It has been demonstrated that non-negative tensor factorization is a practical solution that can be used to automatically discover phenotypes from electronic health records (EHR) with minimal supervision from humans. In general, these methods call for the establishment of an input tensor that describes the inter-modal interactions; however, in practice, the correspondence between various modalities (for example, the correspondence between medications and diagnoses) is frequently absent. Even though heuristic methods can be used to estimate them, doing so invariably results in the introduction of errors and causes the quality of the phenotype to be less than ideal. This is of utmost significance for patients who have multiple diagnoses and medications listed simultaneously in their medical records, such as those who are receiving critical care because they have a complex medical condition. We propose the collective hidden interaction tensor factorization (cHITF) to infer the correspondence between multiple modalities jointly with the discovery of phenotypes in order to alleviate this problem and discover phenotypes from EHR with unobserved inter-modal correspondence. This will allow us to discover phenotypes from EHR with unobserved inter-modal correspondence. We make the assumption that the observed matrix for each modality is a marginalization of the unobserved inter-modal correspondence. These unobserved inter-modal correspondences are then reconstructed by optimizing the likelihood of the observed matrices. Extensive experiments carried out on the real-world MIMIC-III dataset demonstrate that the cHITF effectively infers clinically meaningful inter-modal correspondence, discovers phenotypes that are more clinically relevant and diverse, and achieves better predictive performance when compared with a number of state-of-the-art computational phenotyping models. Did you like this research project? To get this research project Guidelines, Training and Code... Click Here facebook twitter google+ linkedin stumble pinterest A Quick Method to Accurately Find Important Items in Data Streams is LTC. Traffic Forecasting: Learning Dynamics and Heterogeneity of Spatial-Temporal Graph Data