- What is artificial intelligence?
- Neural network intelligence myth or reality?
- Information access protection
- Life in the cloud
- Cloud computing
- Anonymizer as a means of protecting electronic money
- Creating a company blog
- Viruses on websites
- Endless expanses of the Internet. Email.
- Introduction to OpenCL
- Information security of a small organization
- Video conferencing systems
Kretov Nikolai Nikolaevich: Is neural network intelligence a myth or reality?
At present, when in various production activities a replacement for human hands is needed, which can assess the situation and make, sometimes for a person, very complex decisions, there is a struggle on the market between manufacturers of high technologies that are capable of somehow performing intellectual tasks.
Already in the West and in other highly developed countries, such technology as neural networks is used. According to many scientists, neural networks are the future of computing. So what is a neural network?
A neural network is interconnected simulated neurons (a model of neurons located in the human brain). This structure allows you to analyze the input data and output the calculation result. The number of network inputs is equal to the amount of data that it must process, and the number of outputs depends on the form in which you need to get the solution to the problem. With the help of neural networks, you can solve almost any problem.
A frequently mentioned example of neural network problems is the problem of pattern recognition. The picture (let's say black and white) shows the letter . Our task is to recognize this letter. This is a typical task for neural networks and is very difficult to solve using any other technologies. And it’s very easy to translate this task into a neural network language. Let’s assume we have a black and white image of 30 * 30 pixels, so our neural network will have 900 inputs, each of which receives a bit signal equal to either 0 or 1 (depending on the color of the given pixel). There will be 33 outputs (one output - a letter), as a result of calculations, a signal should appear at the output that denotes the letter shown in the figure (the strongest if we use analog signals).
But the neural network itself is not able to correctly solve the problem, so for starters it needs to be trained. That is, to adjust for this task (adjust the weights). Learning a neural network is like teaching a child: you show one letter and look at the outputs of the network. If the output is incorrect, we correct the necessary weights and so on until the moment when the neural network gives a lot of correct answers. When the neural network is trained, it can give the right decisions even when the situation is unfamiliar to it (the letter is slightly distorted).
We have considered the topology of the simplest neural network.
This is a very promising direction in computational science, which in the future will possibly replace a person in many of his activities.
References to literature.