Neural Networks

[steve_bank]

https://en.wikipedia.org/wiki/Artificial_neural_network
http://pages.cs.wisc.edu/~bolo/shipyard/neural/local.html

"...Artificial neural networks (ANNs) or connectionist systems are computing systems vaguely inspired by the biological neural networks that constitute animal brains.[1] Such systems "learn" (i.e. progressively improve performance on) tasks by considering examples, generally without task-specific programming. For example, in image recognition, they might learn to identify images that contain cats by analyzing example images that have been manually labeled as "cat" or "no cat" and using the results to identify cats in other images. They do this without any a priori knowledge about cats, e.g., that they have fur, tails, whiskers and cat-like faces. Instead, they evolve their own set of relevant characteristics from the learning material that they process.

An ANN is based on a collection of connected units or nodes called artificial neurons (a simplified version of biological neurons in an animal brain). ..."

Open the pod bay doors HAL

Looks like it is a developed technology with applications. Non algorithmic computing, which by inference is how our brains function.

We learn, modify, and optimize by repeated trial and error. Human creativity can not be reduced to an algorithm, IOW a Turing Machine.

Goedel thought a human mind analog could possibly be taught as we are from childhood.

 

[rjh01]

One issue with such things is that it can hard to tell what it has leant. For example all pictures of cats would have a keyboard also in the picture. However no picture without a cat has a keyboard. This fact is unknown to the humans. The neural network then learns to tell the difference between pictures containing keyboards from pictures without keyboards. Then a picture of a cat without a keyboard is shown and it says there is no cat in the picture. No human would be able to work out why it got it wrong. All what they can do is tell the network that it is wrong and hope the network can correctly identify the problem. Then the network is shown a picture of a dog and it thinks it is a cat.
A human would learn by being shown one live pet cat and being told it is a pet cat. When that human then sees a dog for the first time the human would know it is not a pet cat because it fails to meet the criteria they have taught about what is a pet cat. Then the human sees a lion and is told it is a different type of cat.

 

[Subsymbolic]

One issue with such things is that it can hard to tell what it has leant. For example all pictures of cats would have a keyboard also in the picture. However no picture without a cat has a keyboard. This fact is unknown to the humans. The neural network then learns to tell the difference between pictures containing keyboards from pictures without keyboards. Then a picture of a cat without a keyboard is shown and it says there is no cat in the picture. No human would be able to work out why it got it wrong. All what they can do is tell the network that it is wrong and hope the network can correctly identify the problem. Then the network is shown a picture of a dog and it thinks it is a cat.
A human would learn by being shown one live pet cat and being told it is a pet cat. When that human then sees a dog for the first time the human would know it is not a pet cat because it fails to meet the criteria they have taught about what is a pet cat. Then the human sees a lion and is told it is a different type of cat.

Gavagai!

 

[steve_bank]

Us humans are good at pattern recognition, probably a survival characteristic.It is hard to duplicate in software, a lot of 'noise' in the pictures.Same with audio. A skilled radio operator can pick out Morse Code with a lot of noise. Neural nets are better than traditional software solutions.