Sign in to follow this  
brwarner

Weighing Neural Nets

Recommended Posts

On AI_junkie I leaerned about Neural Nets and hoiw to train them using Genetic ALgorithims. How would I train them manually: In one tutorial on GameDev they programmed a neural net without a genetic a;lgorithim. Is it just a process of tinkering with values??

Share this post


Link to post
Share on other sites
I have now read this method but what if i have 2 desired outputs for the network like :

Inputs: 1,5 Desired Output: 9
Inputs: 2,5 Desired Output: 6

How would Backpropagation work here??

Share this post


Link to post
Share on other sites
Quote:
Original post by brwarner
I have now read this method but what if i have 2 desired outputs for the network like :

Inputs: 1,5 Desired Output: 9
Inputs: 2,5 Desired Output: 6

How would Backpropagation work here??


It would learn both values. I don't see what's special about this example.

Share this post


Link to post
Share on other sites
I don't understand what you're saying, what do you mean by manually train them. In a fully connected network every weight change will effect all values downstream, unless you're having something trivial like two input and one output node, this is an impossible task other than to do it by th methods found in the literature, backprop, secong order (conjugate gradient, levenberg marquardt etc), and genetic algorithms. for the latter to work though you'll need some kind of measure of preformance for it to work. Simulated annealing is another one (but basically random search as is the GA by the way). For a network to produce 6 or 9 as outputs, you'll need linear output units, normally it's between 0 and 1 or -1 and 1 (you could scale it up of course)

Quote: I have now read this method but what if i have 2 desired outputs for the network like :

Inputs: 1,5 Desired Output: 9
Inputs: 2,5 Desired Output: 6

How would Backpropagation work here??

With the right structure (see above) the inputs 1,5 would generate 9 and 2,5 would generate 6 after training. For input 2,5 you would probably get something like 7.5 (that's called generalisation, and is the objective of BP-learning) But since in you're training you kept the second input constant at 5 (and there are no other samples), some thing like 1,3 would give some unpredictable (and utterly useless) result.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this