Learning to satisfy

This paper investigates a class of learning problems called learning satisfiability (LSAT) problems, where the goal is to learn a set in the input (feature) space that satisfies a number of desired output (label/response) properties. LSAT problems are motivated, in part, by applications in computational finance, and an experimental investigation of LSAT in the context of portfolio selection is reported. A distinctive aspect of LSAT problems is that the output behavior is assessed only on the solution set, whereas in most statistical learning problems output behavior is evaluated over the entire input space. Consequently, certain learning criteria arising naturally in LSAT problems require a novel large deviation bounding technique. 1 Learning and Satisfaction In most statistical learning problems, one is interested in minimizing a risk function such as expected squared error or probability of error. However, in many applications, one is interested in a solution to the learning prob...

• Type: Conference
• Year: 2008
• Where: ICASSP
• Authors: Frederic Thouin, Mark Coates, Brian Eriksson, Robert D. Nowak, Clayton Scott
• Keywords: ICASSP 2008, Learning Problems, LSAT Problems, Signal Processing, Statistical Learning Problems