professional bio |
Heuristic optimization methods are adaptive when they sample problem solutions based on knowledge of the search space gathered from past sampling. Recently, competent evolutionary optimization methods have been developed that adapt via probabilistic modeling of the search space. However, their effectiveness requires the existence of a compact problem decomposition in terms of prespecified solution parameters.
How can we use these techniques to effectively and reliably solve program learning problems, given that program spaces will rarely have compact decompositions? One method is to manually build a problem-specific representation that is more tractable than the general space. But can this process be automated? My thesis is that the properties of programs and program spaces can be leveraged as inductive bias to reduce the burden of manual representation-building, leading to competent program evolution.
The central contributions of this dissertation are a synthesis of the requirements for competent program evolution, and the design of a procedure, meta-optimizing semantic evolutionary search (MOSES), that meets these requirements. In support of my thesis, experimental results are provided to analyze and verify the effectiveness of MOSES, demonstrating scalability and real-world applicability.
committeeDr. Ron Loui (chair), Dr. Guy Genin, Dr. Ben Goertzel, Dr. David E. Goldberg, Dr. John Lockwood, Dr. Martin Pelikan, Dr. Robert Pless, Dr. William Smart