Option to apply RLS learning rule at random intervals

[arvoelke]

In the full-FORCE paper, they apply it at a random interval with a mean of 2 ms. It's very slow if applied every time-step (like 100x slower than PES for 2000 neurons, since it's updating 4 million weights every time-step). Some sort of interval as a fixed value (in simulation seconds) or perhaps as a distribution would be better.