how to get fix output for neural network?
3 views (last 30 days)
Show older comments
hi, i am new to neural network. as i run my program every time different outputs are there.so can any one tell me how to get same output which is best. thanks
0 Comments
Answers (2)
Greg Heath
on 4 Mar 2012
In general, weight initialization and data division involve random numbers. Therefore
1. Intialize the RNG to the same state whenever you want to reproduce results.
2. Train in a double loop. I typically use 10 trials for every value of H (number of hiden nodes).
3. Tabulate the results.
4. Find the best result. You can retrain to get the weights by counting how many times the RNG state has changed since initialization.
5. A better generalizing alternative is to find the several best designs and average their outputs. Search the term "ensemble".
Hope this helps.
Greg
Greg Heath
on 7 Mar 2012
RNG = random number generator
ENSEMBLE = a combination of neural nets.
Search the term in comp.ai.neural-nets and comp.soft-sys.matlab.
Hope this helps.
Greg
0 Comments
See Also
Categories
Find more on Deep Learning Toolbox 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!