BACKGROUND
Prominent NLP, CV applications can be modeled as big graphs and optimization problems. However, as the size of the graph increases, the problem is harder to solve. Thus, there is a need to develop scalable algorithms based on optimization techniques and applied mathematical methods to provide accurate and explainable solutions to these problems.
My research interests are driven by these three themes:
Identifying and modeling real world applications as big graphs and network problems
Designing explainable and scalable algorithms to solve them.
Developing parallel and GPU based implementations to handle large scale data.
Application areas:
NLP : Coreference resolution, entity deduplication, text summarization
CV: Multi-object tracking, Shape matching
Network optimization: Facility location, multi-source data association, record deduplication
Methodological areas:
Large scale optimization, integer programming, linear programming, combinatorial optimization, graph theory.
Deep learning, machine learning, reinforcement learning.
Specifically, my research interests are focused on designing and developing GPU accelerated (high performance computing) algorithms to solve NP-Hard integer programming problems. I apply optimization techniques and methods to design the algorithms and develop parallel implementations (CUDA and MPI), to handle large scale data.
Topics of Interest:
Intersection of Deep learning and Operations Research.
Network science and graph theory to model social issues.
Building explainable models using graph algorithms and optimization for NLP applications.
Building scalable and GPU-accelerated algorithms to solve integer programs that have been notoriously hard to solve.
PROJECTS:
Framework to solve Integer Programs (IPs)
Explainability in large scale applications and data
I. Multi-dimensional Assignment Problem
Application: Multi-Target Tracking (or Multi-Object Tracking)
II. Weighted Clique Partitioning
Application: Entity Resolution (or coreference resolution)