I'm interested in scalable Optimization, Machine Learning and Artificial Intelligence with applications in Health Informatics,
Educational Data Mining and Domain Adaptation. I am a part of Sriraam Natarajan
Statistical Relational AI (StarAI
This section is currently being updated. Please check back soon!
Scalable, generalizable, kernelizable metric learning can dramatically improve the performance of several semi- and un-supervised learning algorithms. We are interested in developing general, sparsity-inducing techniques, with applications in NMF, collaborative filtering, clustering, and kernel and manifold learning. Optimization approaches on matrix manifolds are also considered, which can significantly reduce the complexity of several learning problems and lead to large-scale efficiency improvements.
Inspired by, knowlege-based SVMs, we are interested in incorporating (human) domain and expert advice into the learning setting.Advice helps speed up and bias learning so that better generalization can be obtained with less data. Current approaches to this include the adviceptron, a passive-aggressive online algorithm, and arkSVMs, advice-refining KBSVMs.
Inverse Reinforcement Learning.
IRL has emerged as a hot topic of research, with diverse approaches designed to learn the reward function of a MDP from observation of behavior/trajectories of an expert. We are interested in generalizing this to various settings: multi-agent learning, continuous state-advice spaces, advice-taking IRL agents, to name a few. The goal of this research is to extend the IRL technique to in both batch and online settings. Specifically, the potential of advice-giving techniques and the ability for non-AI experts to provide natural advice to learners are prime motivations for this study.