3 ms·
Or with batchnorm for that matter. The multiplicative part of BN can be folded into weights after training, and the additive portion can be folded into bias. No
by m0zg 7y ago
Or with batchnorm for that matter. The multiplicative part of BN can be folded into weights after training, and the additive portion can be folded into bias. Nobody would know you used BN in the first place. Dropout disappears altogether at inference time.
- mkagenius 7y agoMost of the time litigation is just to harass and tire the weaker opponent, truth may not even matter.