https://twitter.com/OpenAI/status/1037765547427954688

In this tweet OpenAI updated about updates made to their Dota 2 agent. Here are two of the points listed.

  1. Double size

  2. from old parameters

How exactly is this done in implementation? If you double the model’s size, how exactly can the new layers with double the parameters be initialized from the old set of parameters?

What strategies are used? Is it as simple as concatenating each layer’s ’s with itself to double them?





Source link
thanks you RSS link
( https://www.reddit.com/r//comments/9dpznk/d__to_initialize_a__parameter_from_a/)

LEAVE A REPLY

Please enter your comment!
Please enter your name here