An Iterative Similarity-Based Adaptation Technique for Cross-Domain Classification
Himanshu S. Bhatt and Raghu Krishnapuram
Xerox Research Centre India
Wednesday, September 16, 2015
Brickyard Engineering (BYENG) 210, Tempe campus [map]
In real life applications, training and test samples are not independently and identically distributed, and hence lead to poor generalization. Training models from scratch for every new domain hinders large-scale adoption of supervised statistical learning. Transfer learning techniques allow domains, tasks and distributions used in training and testing to be different, thus reducing the requirement for labelled data. However, brute force techniques suffer from the problem of negative transfer, and we need to judge when and how much to transfer. In this talk, Himanshu S. Bhatt and Raghu Krishnapuram present a technique to adapt models with minimum supervision, where the notion of similarity between source and target domains helps in moderating knowledge transfer. They demonstrate the effectiveness of the technique for micro-blog categorization, so that the models can be shared across different industries such as telecom, retail, and health care. And will also touch upon more realistic and practical scenarios for domain adaptation where target domain data is not available upfront but is available incrementally, and machine learning based solutions can adapt progressively (on the fly) as and when data is available.
Himanshu S. Bhatt is a research scientist at Xerox Research Centre India where he is a member of Text and Graph Analytics Group and leads a project on efficient scaling of machine learning based solutions/offerings across different domains and industries. He received his Ph.D. from IIIT-Delhi, India in 2014 where he worked on various machine learning paradigms such as online learning, co-training, transfer learning, clustering, re-ranking, as well as genetic and memetic algorithms. His Ph.D. dissertation received a Best Doctoral Thesis Award by the Indian National Academy of Engineering (INAE) and the Indian Unit of Pattern Recognition and Artificial Intelligence (IUPRAI) in 2014. Bhatt has over 20 publications in refereed journals, book chapters, and conferences. He is a recipient of IBM Ph.D. fellowship 2011-2013 and two best poster awards in IEEE conferences.
Raghu Krishnapuram joined XRCI (Xerox Research Center India) recently from IBM T J Watson Center, Yorktown Heights, New York, where he was a technical leader for cognitive computing research. Earlier, Raghu served as associate director at IBM India Research Lab, where he led projects in the areas of Knowledge, Information, and Smarter Planet Solutions, with a particular focus on emerging markets. He has also served as a relationship manager for IBM’s services divisions such as IBM Global Process Services and IBM Business Services. Raghu is an alumnus of IIT-Bombay and Carnegie Mellon University. He has been a faculty member at the University of Missouri and Colorado School of Mines. Raghu has published over 160 papers in journals and conferences, and he has been recognized as a Master Inventor by IBM. He is also a Fellow of IEEE and the Indian National Academy of Engineers, and has served on the Technology Council of the IBM Academy of Technology