Skip to content

Variable Sample Sizes on CIFAR10 + Swish6 Activation Function
Compare
Choose a tag to compare
@joaopauloschuler joaopauloschuler released this 11 Nov 19:03
· 307 commits to master since this release
v1.0.5
6eec0ac

This release has:

  • You can now define the validation sample size when loading CIFAR-10 and CIFAR-100: #73 . By tweaking this number, you may find bigger test accuracies when training your models.
  • There is a new activation function TNNetSwish6. This activation function is a fantastic replacement for ReLU6 as it gives higher accuracy while maintaining numerical stability.
Assets 2