Supported by Dash Optimization
Period:
Xpress-MP Optimization Suite has been used to tackle complex problems that combine information from varied and noisy sources. Specifically, Xpress-MP has been used as a general inference engine for a suite of natural language tasks. In the past year, we have used integer linear programming with boolean constraints to incorporate various types of structural and linguistic knowledge into various tasks where one must combine a set of dependent information collected independently. For example, when predicting the semantic roles of sentence fragments, it is easy to learn functions that describe small parts of the decision task such as the part-of-speech of each word. However it is much more difficult to learn the entire role structure at once. We have developed a system that infers the final semantic role label given the labels of a collection of sub-problems. Xpress-MP is used to combine small pieces of knowledge into a coherent and accurate prediction.
