Neural networks... probably doesn't belong here, but I need it in beginners terminology
I don't understand neural networks at all. I'd like to begin my own artificial intelligence program, but I can't seem to make sense of neural networks. Could somebody help me understand them? I will probably repost this in the windows c++ programming forum also.
Last edited on
You should not post in multiple categories. What on earth does neural networks has to do with windows?
There are many different kinds of neural networks. I think feedforward neural networks is one of the simplest kind of neural networks.
A neural network takes some input values and produce some output values. You can see a neural network as a function that takes some arguments (input) and return something (output). There are different ways to train the neural network to make it more accurate. One common method for feedforward networks is backpropagation.
The details of neural network is a too big subject to be described in a forum post like this. You should read a book or find some good source on the internet. Unfortunately I don't have any links to recommend.
If you like watching videos I think this one is quite good: http://www.youtube.com/watch?v=TH7WyX4E3dE (there are many parts)
They use Java but the theory should be true for any language.
Just to warn you: Neural networks are not a method that will just create a clever AI by itself. There is a lot of work, testing different input, different parameters etc. that has to be done. It can be really hard to know why a neural network gives bad output and therefore hard to fix. From my experience, if you can think of another way to make your AI that is usually a better way.
There are many different kinds of neural networks. I think feedforward neural networks is one of the simplest kind of neural networks.
A neural network takes some input values and produce some output values. You can see a neural network as a function that takes some arguments (input) and return something (output). There are different ways to train the neural network to make it more accurate. One common method for feedforward networks is backpropagation.
The details of neural network is a too big subject to be described in a forum post like this. You should read a book or find some good source on the internet. Unfortunately I don't have any links to recommend.
If you like watching videos I think this one is quite good: http://www.youtube.com/watch?v=TH7WyX4E3dE (there are many parts)
They use Java but the theory should be true for any language.
Just to warn you: Neural networks are not a method that will just create a clever AI by itself. There is a lot of work, testing different input, different parameters etc. that has to be done. It can be really hard to know why a neural network gives bad output and therefore hard to fix. From my experience, if you can think of another way to make your AI that is usually a better way.
Last edited on
+1 to Peter87
Neural Network + Beginner Terminology = DOES NOT COMPUTE!
Save your attempt at A.I. until you're very well versed in math, logic and programming.
NOTE: That's not to say you shouldn't start looking into it and playing around. Just don't expect to be having a personal conversation with your PC anytime soon...
Neural Network + Beginner Terminology = DOES NOT COMPUTE!
Save your attempt at A.I. until you're very well versed in math, logic and programming.
NOTE: That's not to say you shouldn't start looking into it and playing around. Just don't expect to be having a personal conversation with your PC anytime soon...
What AI method that is best suited depends very much on what problem you try to solve.
If you are doing a chess AI you might want to look into the minimax algorithm.
If you try to solve Sudoku puzzles you can use a backtracking algorithm.
If you try to find the closest path between two locations on a map you can use a search algorithm like A* or Dijkstra's algorithm.
If you are doing a Goomba AI in a Mario game, well, I guess it's just to check if there is something blocking the way and if there is, change direction.
If you are doing a chess AI you might want to look into the minimax algorithm.
If you try to solve Sudoku puzzles you can use a backtracking algorithm.
If you try to find the closest path between two locations on a map you can use a search algorithm like A* or Dijkstra's algorithm.
If you are doing a Goomba AI in a Mario game, well, I guess it's just to check if there is something blocking the way and if there is, change direction.
Last edited on
So if I want one that's going to run my pc, how would I do that? Think Jarvis, but eight million times less sophisticated. I was going to run it off of speech recognition, but I can use the stuff that came in my win7 until I'm ready to mess with grammars and all that. But essentially, one that learns things as it goes. The program opening, etc is easy that's all input->command style things, but I'd like to also have it learn things, such as sayings, such as replacing "open" with "start" after hearing me say they're the same. Any ideas for this?
Topic archived. No new replies allowed.