Beyond Corroboration: Strengthening Model Validation by Looking for Unexpected Patterns
Fig 2
General scheme of the Pattern Space Exploration method.
M1 and M2 represent the dimensions of the pattern space. The pi represent individual parameter variables, and the mi represent pattern variables. The pattern space is discretised into cells and rare (novel) parents are selected based on the hit count of their corresponding cell. New individuals are generated by crossover and mutation and evaluated to get their pattern. Newly evaluated individuals increment the hit count of their respective cell. All individuals (parents and offsprings) are filtered at the elitism stage so only one per cell remains, randomly picked for each cell.