Cycling Game (Mini Neural Net Demo) (doc.ic.ac.uk)

A game where each cyclist is a mini neural network, and they try to figure out an optimal strategy.

I like this a lot.

Trying to understand: so there is a single layer between the inputs and the output? And the inputs are all computed…

Interesting that there is (I assume a running) average 100m gradient and a separate 1000m gradient. I suppose so brains can differentiate between brief climbs/descents and longer climbs/descents.

Wondering now about an input for "Less than 100m to finish" to allow for last second sprinting. Or "Rider approaching behind within 10m".

I'd love to know what I'm looking at. The description doesn't really help me understand how/why/what it means as the numbers are change.