Monotonic Constrained NEURAL NETWORK
2 views (last 30 days)
Show older comments
I am new at the Neural Network field. Please be explicit. I have two input neurons, one output target and two hidden neurons. I want the derivative with respect to the inputs and the bias are positive. Is there a trick to constraint the Neural Network so that the derivative of the neural network's outputs with respect to the input variable is positive and the bias is positive. I believe I want that the products of the weights and the biases along all paths to be positive as long as the activation function is monotonically increasing .
1 Comment
Orion Wolfe
on 25 Mar 2016
Restrict all of the gains to be positive. This can be done by using a positive mapping on the gains. For example if the network is defined using tanh(w'*x+b) activation functions replace w with g(w) where g is R to R+ mapping g(w) = log(1+e^w) is one such mapping and the activation function become tanh(g(w)*x+b)
Answers (1)
Greg Heath
on 8 Jul 2015
If the derivative of the target with respect to the input is positive, just design a good net with as few hidden nodes as possible. It may take a lot of trials.
0 Comments
See Also
Categories
Find more on Define Shallow Neural Network Architectures in Help Center and File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!