Some Idea from Non-linear Classification of Perceptron Neural Networks

Concerning with perceptron neural networks, non-linear classification is a problem.It base on the mathematical output of all input,It use some kinds of activation function, multilayer perceptron neural networks to solve the problem.But it makes it more complex,in addition,the learning algorithm precision may fall too.The activation function is a mathematical output,in fact,it's a deduction output.Like some other mathematical theorem,it needs so complex logic to prove.

Can deduction logic prove all of problem in reality?Can all assertion proved by logic?Maybe can not.Maths has a lot of axiom,it's output is proved by human experience,i.e. by induction.Maybe the induction is the key to artificial intelligence.The induction needn't complex logic operation,but it's structure needs more careful design.