In my recent experiments on applying PSO to a high dimensional problem, I have observed that many well-known general initialization mechanisms have shown no roles in improving the performance of PSO, whereas when a domain-specific initialization mechanism is applied, a remarkable improvement in the performance of PSO is obtained! Any insight that you may be able to provide would be much appreciated.