Feb 2020 – Jun 2022 · 2 yrs 4 mos · Greater New York City Area

Nov 2017 – Feb 2020 · 2 yrs 3 mos · Dallas/Fort Worth Area

Oct 2013 – Nov 2017 · 4 yrs 1 mo · Dallas/Fort Worth Area

• I led R&D efforts on using machine-learning for:
• tracking individual and team skill mastery to deliver personalized pedagogical interventions and gamification in intelligent tutoring systems;
• large-scale text-mining and social media analytics and social networking for first-responder situational awareness, disaster response, planning and co-ordination;
• spatio-temporal and player behavior mining from popular multi-player online games such as Defense of the Ancients 2 (DOTA2) and Counter Strike: Global Offensive;
• medical informatics problems in diverse medical domains including renal cancer, traumatic brain injury and burn injuries.

Feb 2008 – Feb 2013 · 5 yrs · Madison, WI

• Metric Learning: Developed MDML, a parallelizable, highly-scalable, online metric-learning algorithm based on mirror descent that simultaneously performs feature selection while learning a metric, with applications in semi-supervised and unsupervised learning approaches
• Knowledge-Based Learning: Investigated knowledge-based support vector machines (KBSVMs) that can incorporate prior knowledge and expert advice for better generalization with less data
• Developed the Adviceptron, an online, passive-aggressive algorithm for efficiently learning KBSVMs with applications in tuberculosis strain identification, breast cancer diagnosis and handwriting recognition
• Developed the arkSVM (advice-refining KBSVM), a non-convex advice-refinement approach that learns using data and expert advice; can provide expert with interpretable refinement of provided advice
• Developed knowledge-based inverse reinforcement learning (KBIRL), where an agent learns from demonstrations, and from a domain expert that provides natural advice as preferences over states/actions
• Fast Inference in MLNs: Led the development of datalog-driven, Java-based extension of Tuffy, a highly scalable, SQL-based Markov logic network (MLN) inference engine
• Bootstrap Learning: Co-led development of an intelligent student – an autonomous computer agent that can learn from natural-language teacher instruction – in this large, collaborative DARPA project
• Co-developed Inductive-Logic-Programming-based discriminative algorithms to learn from teacher instruction given as examples and teacher advice
• Integrated, debugged, and tested Java learning modules for applications such as unmanned aerial vehicles, and intelligent surveillance and recon
• Co-developed and tested a graphical, human-computer interface for non-AI experts to easily provide advice and domain knowledge to learning algorithms
• First point of collaborator contact in a challenging, fast-paced environments, meeting multiple deadlines

Aug 2003 – Jan 2008 · 4 yrs 5 mos · Troy, NY

• Bilevel Machine Learning: Developed unifying framework for continuous cross-validation for support vector machines (SVMs) using bilevel optimization; the formulation solves cross validation as a continuous non-convex optimization problem and includes other tasks such as feature selection, and learning from incomplete data
• Fast, Locally-Optimal MPEC Optimization: Developed SLAMS (Successive Linearization Algorithm for Model Selection), for fast computation of local solutions to linear mathematical programs with equilibrium constraints (MPECs) arising in bilevel machine learning problems

Aug 2001 – May 2005 · 3 yrs 9 mos · Troy, NY

• Teaching assistant for graduate-level machine learning courses, and undergraduate math courses