sigmoid neuron vs Acceptron

This app explains how a single sigmoid neuron works. You can have some insight for neural network.

Acceptron turns formal language theory into something you actually want to do every day. You're given a language in plain English: "accepts strings ending in 01," "contains 000 as a substring," "starts with 0 AND ends with 1". You draw the state machine that recognizes it. Place states, wire up transitions, mark your accepting states, and hit Check. If you're wrong, you're not just told "wrong", you get the shortest counterexample: the exact string your machine got wrong, and whether it should have been accepted or rejected. If you're right, you're scored on how close your state count is to the true minimum, computed for real from the target language, not hard-coded.