Graph Grammar Encoding and Evolution of Automata Networks

The global dynamics of automata networks (such as neural networks) are a function of their topology and the choice of automata used. Evolutionary methods can be applied to the optimisation of these parameters, but their computational cost is prohibitive unless they operate on a compact representation. Graph grammars provide such a representation by allowing network regularities to be efficiently captured and reused. We present a system for encoding and evolving automata networks as collective hypergraph grammars, and demonstrate its efficacy on the classical problems of symbolic regression and the design of neural network architectures.

• Type: Conference
• Year: 2005
• Where: ACSC
• Authors: Martin H. Luerssen
• Keywords: ACSC 2005, Automata Networks, Collective Hypergraph Grammars, Computer Science, Neural Network