Learning in Fuzzy Boolean Networks - Rule Distinguishing Power
Rua Alves Redol · 2002
Fuzzy Boolean Networks are Boolean networks with nature like characteristics, such as organization of neurons on cards or areas, random individual connections, structured meshes of links between cards. They also share with natural systems some interesting properties: relative noise immunity, capability of approximate reasoning and learning from sets of experiments. An overview of the processes in- volved in reasoning supported on an hardware architecture are pre- sented, as well as how Hebbian-Grossberg learning can be achieved. An interesting problem related with these nets is the number of different rules that they are able to capture from experiments without cross inter- ferences, that is, their rule capacity. This work establishes a lower bound for this number, proving that it depends on the number of inputs per consequent neurons and its relation to consequent granularity. An application for traffic problems is also provided.