Towards Artificial Intelligence by Combining Qualitative Reasoning and Machine Learning

XiaoYu (Gary) Ge (ANU)

ARTIFICIAL INTELLIGENCE SEMINAR PhD Monitoring

DATE: 2013-10-30
TIME: 12:30:00 - 13:00:00
LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
Qualitative reasoning (QR) and machine learning (ML) have both proved powerful tools for artificial intelligence. Qualitative reasoning is crucial since it captures conceptual knowledge in ways that can be processed by machine. A well designed qualitative representation significantly reduces problemsa search space by leaving out details that are not essential for conceptual understanding. With the ever increasing amounts of data becoming available, the use of machine learning has spread rapidly throughout computer science and beyond. Machine learning systems can automatically learn knowledge from data. There are many different types of machine learning exist and the most mature and widely used is classification. QR usually relies on domain specific knowledge which is usually manually constructed and can hardly be generalized while MLas success often relies on having a good feature representation of the data, and having poor representations can severely limit the performance of learning algorithms. Our goal is to integrate the QR and ML techniques to solve difficult AI problems which cannot be solved by any of them alone due to their own limitaions. For example, automatically learning useful representation from continuous environment requires a ML component to discretise the continuous features and a QR component to aunderstanda those features, by which meaningful representation can be constructed. The first part of the talk reviews the related works on combining qualitative and quantitative methods in artificial intelligence. The talk highlights several methods that have the potential to achieve the goal and explores the areas where those methods can be applied. We will then discuss one possible strategy by demonstrating our previous work in qualitative stability verification. The second part describes a recently developed representation of general solid rectangles (X.Ge & J.Renz, 2013). The representation can be efficiently obtained from the minimum bounding rectangles of the rectangular objects and is expressive enough to capture the force-dynamic relations between those objects. I will show how human can convert their knowledge to machine-understandable representation by talking about the approach we used to develop the representation; and whether there is an automatic way to develop such representation by replacing the human module with ML.