Structural Diversity Diagnostic
The diagnostic confirmed that GP and MQ-RBF are diverse while NN and random forest models are not across five seeds. Pre-drawn random restarts did not restore diversity for neural networks or random forests. The Structural Diversity Diagno…
1 sources - 4 claims
The diagnostic confirmed that GP and MQ-RBF are diverse while NN and random forest models are not across five seeds. Pre-drawn random restarts did not restore diversity for neural networks or random forests. The Structural Diversity Diagnostic fixes deterministic optimizer starting points to separate surrogate structure from optimizer randomness. Under the diagnostic, observed diversity must come from the surrogate response to pseudo-observations.