Blog

Notes on usage of the classic sigmoid and tanh activation functions

Sat, Jan 18, 2025

We never achieved satisfactory results in training neural networks containing ONLY  sigmoid and tanh activation functions. However, by exposing a "slope" parameter of the functions and making it a learnable parameter of the network changed the situation.

Continue reading...

Learning trading rules with MLP with enforced sparsity condition

Publication date: Tue, 11 Feb, 2025

In this post we explain how zero cost tweak of the fully connected MLP can better fit to your trading rule original design.

Continue reading...