Closed arvoelke closed 6 years ago
Thanks @seanny123 for pointing me to: http://extremelearning.com.au/unreasonable-effectiveness-of-quasirandom-sequences/
It should be fairly easy to add in this new approach, which will improve the distribution of eval points, encoders, and semantic pointers. This works for any dimension, as opposed to the old method that regressed for dimensions > 40.
Thanks @seanny123 for pointing me to: http://extremelearning.com.au/unreasonable-effectiveness-of-quasirandom-sequences/
It should be fairly easy to add in this new approach, which will improve the distribution of eval points, encoders, and semantic pointers. This works for any dimension, as opposed to the old method that regressed for dimensions > 40.