We have an artificial neural network to detect patterns in input. training the system to learn specific pattern when the position of pattern is constant is easy and it can be done by feed forward neural network and using back propagation. But the question is that how can we make it position invariant so it can detect the pattern in any place of input ?
Dear friend
Greetings.
You can do it using some dynamic sigmoid function for the learning functin.
This will solve the problem.
All the best.
Best regards
Dr.Indrajit Mandal
Dear friend
Greetings.
You need to use a dynamic learning sigmoid function for the same. I have used it in some of my published papers.
I hope it will help you.
Best regards
Dr.Indrajit Mandal
Hello Dr. Mandel,
Can you send me some reference about dynamic learning sigmoid function ?
Best
The so called Convolutional Neural Networks will do that for you (also known as weight sharing)
weight sharing just can do on a little bit shift . it can not do on big shift.
- Badges
- Science topic