Characterizing the dynamics of symmetry breaking in genetic programming

This paper introduces a metric that measures symmetry in tree graphs, which allows for a statistical characterization of GP solutions by their architectural "shapes." A case study is given that applies this metric to 80.4 million trees to identify trends in GP runs. Results provide a first quantitative look at the dynamics of symmetry breaking. Categories and Subject Descriptors I.2.2 [Artificial Intelligence]: Automatic Programming

• Type: Conference
• Year: 2006
• Where: GECCO
• Authors: Jason M. Daida
• Keywords: GECCO 2006, General Terms Algorithms, Languages Keywords Symmetry, Optimization, Symmetry Breaking