Machine Learning Innovation – How Would Non-Linear Connections Work?

Playing around with a new machine learning library. Neural Networks in Python, NeuPy. I came across an example of connection visualization.

The connections does not seems all that complex. Hmm here might be room for complex improvement. Playing around with ’heal section” under GIMP I created a non linear mock image of the connections. Maybe this can serve as inspiration.

So how can non-linear connections work and improve machine learning?

Notice the healed section of non linear looking connections.