The parallelized evolution strategy that combines all of these features is summarized in the following pseudocode:
---------------------------------------------------------------------------------
Parallelized Evolution Strategy
---------------------------------------------------------------------------------
Initialize parameters on each worker
Initialize random seed on each worker
for do:
for do:
Sample
Evaluate individuals and
Spread returns to each other worker
for do:
Compute normalized rank from the returns
Reconstruct from the random seeds of the other workers
(maybe using Adam)
Now, all that remains is to implement this algorithm.