Discovering optimization algorithms through automated learning
Eric Breimer, Mark Goldberg, David S. Hollinger, Darren Lim · DIMACS series in discrete mathematics and theoretical computer science · 2007
In this paper, we describe the supervised learning approach to optimization problems in the spirit of the PAC learning model. By this approach, we discover domain-specific algorithms by learning from an oracle, which is also an optimization algorithm for the problem in question. We describe examples of learning backtracking-based algorithms and algorithms that implement the dynamic programming paradigm.