artificial intelligence - Continuous output in Neural Networks -

How can I set up a neural network so that they can accept and output continuous range of values ​​rather than discrete individuals? What do I remember from a neural network class some years ago, the activation function would be a sigmoid, which would create values ​​between 0 and 1. What should I do if I want to give the real valuable scalar to my neural network? I thought that if I want a value between 0 and 10, then I can increase the value of 10? What if I have negative values? Do these people usually do or is there any other way? What about input?

Thank you

Most of the work in the field of neural networks with continuous inputs and output Is included in the

P>

- As others have mentioned, if you are concerned about limited limitation of the sigmoid functions, you can use the non-sigmoid activation function on the output node. However, this can cause your output to arbitrarily grow, which can cause problems during training. Sigmoid activation work However, it is important to make sure that your sigmoid work is not very high: an active activation function means that the "useful" category of values ​​is small, which reduces the weight of the network. (This is primarily an issue with genetic algorithms, which uses a fixed weight modification strategy that does not work properly during small weights.)

• Cost per multiple nodes - Multiple nodes are spreading a single continuous value to represent continuous input There is a general strategy. With the cost of increasing the size of the network, it has the advantage of providing more "features" for a network to play with it.
  • Binning - a single input is spread over multiple nodes (for example, where each node has a base function with a different center, which will be partially activated by input ).
  • Representation of binary - divide a single continuous value into 2 ^N segments, without losing a smooth representation, Then feed the feed in the network as a binary pattern for N nodes. This approach is compact, but it is like brittle and the result in the input varies, which changes in non-continuous manner.