I'd like to share with you 3 works in progress about Minic.
1) I was able to train a smaller net which is stronger than noisy notch at short TC by about 20Elo and looks even at longer one. This soon may be released as Minic3.05 + Nibbled Nutshell. But I need to confirm first that progress can still be made on that new architecture.
2) I am now using a new fen generation process that seems promising. More on this later, just teasing.
3) I'm almost done porting Minic to a distributed memory model. This will fun for those how have a big cluster available
Using the same concept as for threading, namely a lock free somehow shared transposition table is it quite easy to achieve this. One sided mpi comms (window) help at some point but playing it smart it is also possible with more standard asynchronous two sided comms.