Make Neural Network run faster (MATLAB)

Question

I have a MATLAB code for neural network and it takes around 3 mins to train the network.(it trains around 85 files). When I use the profile viewer it says that trainlm has a self time 143.166s which 83.3% of that time comes from dWB = -(jj+ii*mu) \ je;

Is there a way of making it faster? Should I use data binning?

Thanks in advance!

user85109 · Accepted Answer · 2013-07-26 13:37:51Z

1

That one line is a solution of a linear system of equations. The limit here is the number of unknowns, so the size of your network which defines the number of unknowns.

No, you cannot really reduce that time easily, unless you are willing to reduce the number of unknowns, or buy/find a faster computer.

answered Jul 26, 2013 at 13:37

user85109

Sign up to request clarification or add additional context in comments.

2 Comments

Chrysovalando Over a year ago

oh.. my supervisor told me that I can use Data binning but I really don't know what data binning actually does and if it will help me reduce the time.. thanks :)

user85109 Over a year ago

Then maybe you need to talk to your supervisor?

C.Colden · Accepted Answer · 2013-07-26 12:19:27Z

0

Have you already checked whether you can somehow parallelize your code using parfor or the parallel computing toolbox in general?

answered Jul 26, 2013 at 12:19

C.Colden

6271 gold badge8 silver badges29 bronze badges

9 Comments

Chrysovalando Over a year ago

well no.. I don't really know how do to that :/

C.Colden Over a year ago

Are you using for loops? If so, try you use parforloops. The following link could help you:mathworks.ch/ch/help/distcomp/parfor.html

Chrysovalando Over a year ago

I do..But it seems that it takes a long time not at the loop but when I have net = train(net,y,T');

C.Colden Over a year ago

It depends on how you bin your data. If bin it and then summarize for every bin, yes, you will lose some data. If you bin to use different cores on your computer, then you will not lose data. If you can somehow use parfor that would actual help, since every core would would run independently each loop. So your net=...would run on all cores in parallel.

C.Colden Over a year ago

btw avoiding forand while at all would also speed up your script. Functions like arrayfun etc are much more efficient...

|

Collectives™ on Stack Overflow

Make Neural Network run faster (MATLAB)

2 Answers 2

2 Comments

9 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

9 Comments

Your Answer

Sign up or log in

Post as a guest

Related