A new family of activation functions that wins on benchmarks

Hey, I proposed a new family of activation functions with promising properties.
Looking for endorsement in cs.LG on arXiv.org.

Here is the benchmark:

1 Like

1 Like