Add to Google Calendar
Date:  Thu, March 07, 2019
Time:  10:00am - 11:00am
Location:  Holmes 389
Speaker:  Dr. Sathya Narayanan Ravi

Abstract:

It is exciting to witness, first hand, the enormous impact of Artificial Intelligence - machine learning, computer vision and more generally, data analysis methods, on fields spanning biochemistry, medicine, physics, astronomy and finance. Meanwhile, it is also becoming increasingly clear that to fully realize the potential promise of machine learning for specific application domains, the technical development cannot be undertaken agnostic of domain knowledge - which may be variously described in the form of prior knowledge, constraints or general purpose thumb-rules. When we can successfully find ways to endow our machine learning models with such knowledge, we often obtain (i) far superior performance for the application at hand, (ii)solutions that are more interpretable by practitioners, (iii) computationally efficient search schemes and occasionally, (iv) new strategies to improve a much broader class of machine learning (ML) models. The overarching theme of my recent and ongoing research is to develop efficient ML models - frequently with provable performance guarantees - that either explicitly or implicitly leverage such information. The technical thrust of my work is to identify ways in which such knowledge can be efficiently encoded mathematically and if so can the corresponding models be optimized and shown to perform well. Extending these ideas to the large scale setting and providing tools that can be used in a plug and play manner or within an industrial setting turns out to be challenging but also rewarding, and remains a core motivation behind research which seeks to balance theory and applications. I will present three broad areas from which constraints arise, solutions that have come out of my doctoral research (for one of them in detail), and promising directions fro each line of work.

Bio:

Sathya is a Ph.D. candidate at the University of Wisconsin-Madison. Sathya's research interests lies in the intersection of Computer Vision, Machine Learning, and their applications in neuroscience. Sathya is interested in developing novel optimization algorithms that exploit domain knowledge for superior performance while subject to various real life constraints.