Seed Grant Awards: May 2018 – December 2018

Robust Predictive Models of Fertility

PI: Ioannis Ch. PaschalidisProfessor (ECE, BME, SE), Boston University School of Engineering, Director, Center for Information & Systems Engineering
Co-PI: Shruthi Mahalingaiah, Assistant Professor (Obstetrics & Gynecology), Boston University School of Medicine

The goal of this project is to develop data-driven, accurate, and personalized fertility prediction models. These models can be used to help women and couples make timely and cost effective family planning decisions for early detection of reduced fertility, and for early detection of specific pathologies that lead to reduced fertility.  With respect to the latter application, the investigators are particularly interested in early detection of Polycystic Ovary Syndrome (PCOS)  – a common endocrine disorder associated with infrequent ovulation. The algorithms that the investigators develop will be based on quantitative reasoning leveraging the PI’s expertise in optimization, machine learning, data mining, applied probability, and decision theory, and benefiting from close collaboration with the Co PI, Dr. Mahalingaiah. This project seeds a new collaboration between the project Co-PIs. With access to new, rich datasets obtained through this project, the investigators seek to develop enough preliminary results from applying robust prediction methods to fertility-related problems so that successful federal grants to support this work can be written in the long run.

Seed Grant Report: Predictive Models of Fertility

Overlapping Graph Partitioning for Distributed Graph Mining

PI: Lorenzo Orecchia, Assistant Professor (CS) Boston University School of Engineering, CISE Affiliated Faculty 
Co-PI: Charalampos E. Tsourakakis, Assistant Professor (CS) Boston University School of Engineering

The investigators plan to design novel, practical approximation algorithm for the problem of overlapping graph partitioning, a variant of the classical graph partitioning problems in which clusters are allowed to overlap, i.e., some vertices may belong to more than one cluster. This problem is of interest in the context of mining large networks, such as social networks, in which vertices may naturally belong to more than one cluster, i.e. community. The technique underlying our approach is a novel semidefinite programming relaxation, which is efficiently solvable and provably captures various objectives for overlapping graph partitioning.  PI Orecchia and PI Tsourakakis are using this project to start a broader collaboration on graph partitioning. The PIs envision future grant submissions to the “Information Integration and Informatics” and the “Algorithmic Foundations” core program within CISE at NSF.

Seed Grant Report: Overlapping Graph Partitioning for Distributed Graph Mining