Currently, working as a Data Scientist/Quant(AI/ML) where I perform highly complex activities related to the design, development, validation, implementation, documentation, and ongoing maintenance of quantitative models (AI/ML) that offer insight into a wide range of product, business or risk mitigation initiatives by utilizing advanced mathematical skills and programming to create and validate analytic models. Before, I have worked as a AI Research Scientist at Fusemachines in the field of time-series analysis for sales and demand forecasting and as a Postdoctoral Research Associate at NYU Tandon School of Engineering in the field of fairness aware AI for healthcare. I have a Ph.D. in Computer Science from Tennessee Tech University, and Master’s in Computer Science from Delft University of Technology (TU Delft), Netherlands.

Research Interest

• Machine Learning/Deep Learning/Data Mining
• Time-Series Analysis
• Causal Effect in Time Series
• Fairness-aware AI for Health
• Temporal Pattern Mining
• Social Network Analysis

News

• Paper on “Fairness Violations and Mitigation under Covariate Shift” accepted to ACM Conference on Fairness, Accountability, and Transparency (ACM FAccT 2021)
• Paper on “Evaluating Scientific Workflow Engines for Data and Compute Intensive Discoveries” accepted at Workshop on Big Data Tools and Use Cases for Innovative Scientific Discovery (BTSD) 2019 @IEEE BigData 2019 LA, USA.
• Paper title “Fair Predictors under Distribution Shift” accepted at Fair Machine Learning for Health Workshop at NeurIPS 2019 (spotlight presentation).

Current Research Projects

• Time-Series Analysis for Sales and Demand forecasting
  • Competitor marketing strategies (Share of Voice)
  • Effects of social media posts and content drop
  • Effects of seasonality on revenue optimization
  • Causal Effect in time series analysis

Past Research Projects

As part of my postdoctoral research involved the study of the problem of learning fair models under mismatch in train-test distributions when either limited or no data is available from the test distribution. We consider the setup of causal domain adaptation where possible shifts are expressed using causal graphs with the goal of learning models with stable performance under the specified shifts. We find that methods to address distribution shift, while controlling for decay in accuracy, can result in fairness violations. As a counter measure, we show that it is possible to obtain accurate and fair predictors for widely studied fairness definitions and under a large class of shifts particularly prevalent in healthcare tasks. Read more related to this research

As part of my Ph.D. research involved the design and development of algorithms to more efficiently and effectively analyze of sequences associated with specific events. For example, the identification of patterns of events that precede or follow a diagnosis of sepsis could be important for understanding and improving healthcare care decisions in a hospital, and while the primary motivation for my research was to improve our ability to analyze clinical data, the research described in the dissertation is applicable in many other domains.