You are here
DH Seminar "Evolutionary Optimization for NLP"
Speaker: Asif Ekbal (Indian Institute of Technology, Patna)
Abstract: In decision science, optimization is quite an obvious and important tool. Depending on the number of objectives, the optimization technique can be single or multiobjective. We encounter numerous real life scenarios where multiple objectives need to be satisfied in the course of optimization. Finding a single solution in such cases is very difficult, if not impossible. In such problems, referred to as multiobjective optimization problems (MOOPs), it may also happen that optimizing one objective leads to some unacceptably low value of the other objective(s).
Evolutionary algorithms and simulated annealing, from the family of meta-heuristic search and optimization techniques, have shown promise in solving complex single as well as multiobjective optimization problems in a wide variety of domains.
In most of NLP tasks, we often optimize various metrics, parameters etc. to obtain reasonable performance. For example, in Information retrieval, it is often necessary to optimize the recall and precision parameters. In automatic summarization, it is desired to optimize different objective functions like similarity to user query, ROUGE metric, important sentence
score, difference in length between the scored sentence and the desired sentence and many others. Other examples of optimization in NLP include parsing, machine translation, and computational models of language acquisition.
In my talk I will first introduce the basic concepts of evolutionary optimization, and then present very briefly some of our works related to named entity extraction in biomedical texts and coreference resolution.
Biography: Asif Ekbal is an Assistant Professor in the department of Computer Science and Engineering, Indian Institute of Technology Patna, India. Before joining to IIT Patna he worked as a post-doctoral research fellow in University of Trento, Italy and Heidelberg University, Germany.
His research interests include named entity extraction, anaphora resolution, machine transliteration, machine learning to NLP and biotext mining.