The performance of all reasoning systems crucially depends on problem representation: the same problem may be easy or difficult, depending on the way we describe it. Researchers in psychology and artificial intelligence have accumulated much evidence on the importance of appropriate representations for both human and artificial intelligence systems. The book proposes techniques for automatic improvement of problem representation, which are based on integration of multiple learning and problem-solving algorithms. It gives theoretical foundations of the proposed techniques, describes their implementation, and discusses empirical evidence of their utility.