Combining Deep Learning and Structural Kernels for Question Answering

13 Oct
Friday, 10/13/2017 2:30pm to 3:30pm
Computer Science Building, Room 151

Abstract: Learning high-level semantic tasks such as Question Answering (QA) requires a remarkable effort in terms of engineering features, thus methods for automatizing this process are extremely valuable. Kernel methods can map feature vectors or directly objects in richer feature spaces while neural networks have been shown to be very effective at generalizing words, constructing embedded representations.

In this talk, I will show some successful solutions based on kernels, deep learning and their combinations for: question type classification and answer sentence passage re-ranking.  In particular, my talk will highlight the important role of relational structural information defined between questions and answers.

Bio. Alessandro Moschitti is a Principal Research Scientist at the Qatar Computing Research Institute (QCRI) and a professor at the Computer Science (CS) Department of the University of Trento, Italy. He obtained his PhD in CS from the University of Rome in 2003. He has worked as (i) an research fellow for the University of Texas at Dallas, (ii) as a visiting professor for the University of Columbia (NY), Colorado and John Hopkins and (iii) as visiting researcher at the IBM Watson Research center (participating at the Jeopardy! Challenge) and at MIT-CSAIL. His expertise concerns theoretical and applied machine learning (ML) in the areas of Natural Language Processing (NLP), Information Retrieval (IR) and Data Mining. He has devised innovative structural kernels and neural networks for advanced syntactic/semantic processing, documented by more than 260 scientific articles published in NLP, IR and ML communities.

He has been the General Chair of EMNLP 2014 and a PC co-chair of CoNLL 2015. He has received four IBM Faculty awards, one Google Faculty award, five best paper awards and the best researcher award from Trento University. He has led many projects; currently is the PI (QCRI side) of a $10 million collaboration project between MIT CSAIL and QCRI.