Neural Network: Computer That Has Intuition

Handwritten Character Recognition

Neural network is a computer model that simulates brain function. N x M of input modeled retina cells are calculated through matrix of nodes modeled synapse circuit, and then the output of the synapse circuit will be an answer for the input. What this technology differs from the traditional ones is that it learns by itself pairs of input and output. It repeats tuning the internal state; the value of joint strength between synapses; in order to adaptĀ  the correct answer for those pairs of input and output.
At the beginning of learning, it answers irresponsible output because the default values of the joint strength are defined at random. And with repetition of learning, it comes to answer correctly.
This technology is utilized to handwritten character recognition for example. It’s easily estimated that it would be very hard to recognize handwritten character because there should be fuzzy aspect depending on the writer. With neural network, those fuzzy aspect can be learned and it can recognize them in a higher rate.
So now we can call it “computer that has intuition”.

Higher level recognition: fuzzy logical case

More than 15 years ago, when I was a graduate student in a university, I wondered whether it could recognize higher level cases. I was studying architectural design, and there were many theoretics on figure. And everybody have believed those theories are logic. CalledĀ  logical recognition with fuzzy aspect.
I wondered that was true since human beings can recognize the logic even if they were not perfect. For example, you can recognize symmetry figure even if there are some little exception.
I tried to make a neural network program and found that it could recognized those “imperfect logical” theories!

Logical theory is metaphysical (smooth surface as well)

If they are really “logical”, they should be 100% symmetry that can be calculated mathematically.

This entry was posted in Topics and tagged , , , . Bookmark the permalink.

Comments are closed.