Optimization Algorithms to Find Most Similar Deductive Consequences (MSDC)

Abstract

Finding most similar deductive consequences, MSDC, is a new approach which builds a unified framework to integrate similarity-based and deductive reasoning. In this paper we introduce a new formulation $\mathcal{OP}$-MSDC(q) of MSDC which is a mixed integer optimization problem. Although mixed integer optimization problems are exponentially solvable in general, our experimental results show that $\mathcal{OP}$-MSDC(q) is surprisingly solved faster than previous heuristic algorithms. Based on this observation we expand our approach and propose optimization algorithms to find the k most similar deductive consequences k-MSDC.