As refining the MLP beyond what it already does seems no easy task, I thought I would move to the previous problem: To make sure that there is enough temporal diversity in the percepts for the predictor to learn from longer temporal sequences. I’m running a proper test now, but following are a few frames […]
↧