Integrating context, domain knowledge, and data-driven modeling of complex phenomena based on observations that are highly structured and interdependent is a very challenging task. In this project we address this problem by using the structured predictive modeling framework based on a probabilistic exponential graphical model instead of relying on traditional statistical approaches that assume independent and identically distributed random variables. Current state-of-the-art predictive modeling techniques usually cannot deal with such large and complex networks; thus we provide knowledge-based compression techniques for complexity reduction. We use multiple kinds of domain knowledge and context to capture additional information that might be missing from the observed data. We also directly constrain the model optimization based on domain constraints and embed other problem-specific qualitative knowledge directly into the framework. The main innovation of this project is extending our structured learning models to explore in detail the hypothesis that a unified approach of integrating big data with sources of high-level knowledge (ontologies, domain-based constraints etc.) is beneficial for predictive modeling of complex phenomena.
Role I have developed predictive models for BIG temporal graphs, which resulted in publications in IEEE Big Data conference.
Temple University. ONR (N00014-15-1-2729). Jun 2015 - May 2016 .

The analysis of social networks often assumes a time invariant scenario, while in practice actor attributes and links in such networks evolve over time and are inextricably dependent on each other. In addition, the temporal graph is just partially observed, multiple kinds of links exist among actors, various actors have different temporal dynamics and environmental influence can be both positive and negative. This project is closely examining the hypothesis that a unified approach of jointly modeling these and related problems is beneficial for prospective analysis of large-scale partially observed temporal hypergraphs. Novel methods for analyzing large and evolving graphs developed on the project are evaluated on high impact applications related to predictive modeling of information networks, climate and human health.
Role I have mentored students and participated in developing predictive models for temporal graphs, which resulted in publications in top-tier AI conferences such as AAAI.
Temple University. DARPA (AFOSR award number FA 9550-12-1-0406). Aug 2012 - Jul 2016 .

This project develops and validates effective predictive modeling technology to achieve the following sepsis treatment related aims on high dimensional and noisy data at a clinically relevant scale
AIM 1 - Personalized sepsis therapy optimization for an individual patient’s state improvement.
AIM 2 - Early diagnosis of sepsis and accurate detection of change in the state of sepsis., and
AIM 3 - Gene expression analysis for sepsis biomarkers identification.
Role I was leading the first and second aims where I developed novel ideas for early detection of sepsis, which resulted in publication in high prestigious conferences such as KDD and SDM.
Temple University. DARPA (DARPA-N66001-11-1-4183). Aug 2011 - Aug 2015 .