On Learning Sparse Structured Input-Output Models Eric Xing, Carnegie Mellon University
In many modern problems across areas such as natural language processing, computer vision, and social media inference, one is often interested in learning a Sparse Structured Input-Output Model (SIOM), in which the input variables of the model such as lexicons in a document bear rich structures due to the syntactic and semantic dependences between them in the text; and the output variables such as the elements in a multi-way classification, a parse, or a topic representation are also structured because of their interrelatedness. A SIOM can nicely capture rich structural properties in the data and in the problem, but it also raises severe computational and theoretical challenge on sparse, consistent, and tractable model identification and inference.
In this talk, I will present models, algorithms, and theories that learn Sparse SIOMs of various kinds in very high dimensional input/output space, with fast and highly scalable optimization procedures, and strong statistical guarantees. I will demonstrate application of our approach to problems in large-scale text classification, topic modeling, and dependency parsing.
Dr. Eric Xing is an associate professor in the School of Computer Science at Carnegie Mellon University. His principal research interests lie in the development of machine learning and statistical methodology; especially for solving problems involving automated learning, reasoning, and decision-making in high-dimensional, multimodal, and dynamic possible worlds in social and biological systems. Professor Xing received a Ph.D. in Molecular Biology from Rutgers University, and another Ph.D. in Computer Science from UC Berkeley. His current work involves, 1) foundations of statistical learning, including theory and algorithms for estimating time/space varying-coefficient models, sparse structured input/output models, and nonparametric Bayesian models; 2) computational and statistical analysis of gene regulation, genetic variation, and disease associations; and 3) large-scale information & intelligent system in social networks, computer vision, and natural language processing. Professor Xing has published over 150 peer-reviewed papers, and is an associate editor of the Annals of Applied Statistics, the Journal of American Statistical Association (JASA), the IEEE Transaction of Pattern Analysis and Machine Intelligence (PAMI), the PLoS Journal of Computational Biology, an Action Editor of the Machine Learning journal, and a member of the DARPA Information Science and Technology (ISAT) Advisory Group. He is a recipient of the NSF Career Award, the Alfred P. Sloan Fellowship, the United States Air Force Young Investigator Award, the IBM Open Collaborative Research Award, and best paper awards in a number of premier conferences including UAI, ACL, SDM, and ISMB.
The Appification of the Web and the Renaissance of Conversational User Interfaces Patrick Pantel, Microsoft Research
The appification of the Web is triggering a fundamental shift in how users access information. We are moving from centralized access points, such as search engines, towards highly specialized, and yet fragmented, functionalities in disconnected apps. This talk explores an entity-centric conversational interface as a mechanism to overcome this fragmentation, highlighting the numerous associated NLP challenges and opportunities that lie ahead.
Consider mobile scenarios, where the traditional search engine paradigm is being cannibalized by search and browse functionalities built directly into specialized apps. For example, while users can search for restaurants and products using their mobile browser, they are increasingly turning directly to applications such as Yelp, Urbanspoon and Amazon. However, interoperability between applications and lacking generalized interfaces to their functionalities pose serious scalability challenges. In this talk, we argue for an entity-centric conversational interface in which natural user interactions with entities are paired with actions that can be performed on the entities, thus enabling the brokering of web pages and applications that can satisfy the intended action. In this vision, the broker is aware of all entities and actions of interest to its users, understands the intent of the user, and provides direct actionable results through APIs with external providers satisfying the intent. The user saves clicks and time to accomplish her intended action and can discover related actions. New revenue streams open up from paid action placement and lead generation opportunities. At the forefront of this direction are a number of NLP challenges in the areas of entity recognition, entity linking, knowledge extraction, intent recognition, and dialog modeling, to name a few.
We end by proposing one particular technique for learning and mapping user intents in a search interface. In an annotation study conducted over a traffic sample of web usage logs, we found that a large proportion of user queries involve actions on entities, calling for an automatic approach to identifying relevant actions for entity-bearing queries. We pose the problem of finding actions that can be performed on entities as the problem of doing probabilistic inference in a graphical model that captures how entity-bearing information requests are generated. Given a large collection of real-world queries and clicks from a commercial search engine, the models are learned efficiently through maximum likelihood estimation using an EM algorithm. Given a new query, inference enables the recommendation of a set of pertinent actions and providers. We propose an evaluation methodology for measuring the relevance of our recommended actions, and show empirical evidence of the quality and the diversity of the discovered actions.
Patrick Pantel is a Senior Researcher at Microsoft Research, conducting research in large-scale natural language processing, text mining, web search, and knowledge acquisition. Prior he served as a Senior Research Manager at Yahoo! Labs, and as a Research Assistant Professor at the USC Information Sciences Institute. In 2003, he received a Ph.D. in Computing Science from the University of Alberta in Edmonton, Canada.