3 ms·
In many cases you can train multiple instances of a neural net simultaneously in increments, average the weights together, and repeat. Making the underlying com
by hedgehog 11y ago
In many cases you can train multiple instances of a neural net simultaneously in increments, average the weights together, and repeat. Making the underlying compute more efficient is an interesting problem.