Wednesday, August 20, 2008

Artificial intelligence and Integral properties in experiments

Behavior of digital technologies and programs algorithmically opposite the work of our brain. They are unbundling the time and space on discrete samples, while we carefully collect all of krupitsam and integrate or, more simply put, sum.

For example, in sound software and hardware of a complex sound wave recorded in the form of discrete digital taps. After she may share a number of simple sinusoid and kosinusoid using Fourier transform. We, on the contrary, take each of the frequencies separately with the help of receptor cells, after which all summed brain, that is a complex wave created as part of its action.

Regarding the video, it is understandable. And the speed of reaction of our bodies is very high, much faster than the speed image processing. Therefore, the rapid shift in the beam we merges in light line. In a series of materials on the basics of digital video, I propose to hold its experience for those who are at least somewhat familiar with Adobe Flash. But, by the way, exactly the same can be done in any video editing. The goal - the definition of speed reaction to the frequency shifts of personnel. In fact, a very simple and takes about ten minutes. Establish editor formiruete a continuous sequence of the three cadres - red, blue and green. Run reproduction. With little importance fps (frame per second - the number of frames per second), you clearly share three changing colors, and when it merges all the increase in white. It is on this principle going to do first color film, introduce same colour television, and now used in the DLP-projectors with color wheel. A principle - Integral properties of the brain. However, it is used at a simple level - movies and videos with their change of static frames.


The old experiment with images. Continuation

However, as the old ... Your humble servant has done it several years ago. The essence is this. Have been drawn up some mathematical models of letters, rather mathematical models of their writing, so that could be distinguished. Originally planned that the letters themselves will be shown on cards in a video camera, but their program would guess. Naturally, in the first phase would simplify everything, so as a basis not used video, but simply binary images. The program compares symbol downloaded from its base models and choose the most appropriate, based on the percentage received, calculated on multi-criteria.

Naturally, I do not set ourselves the task of writing the text recognition algorithm, was interested in another - images. After a successful otkatki in different fonts, different angle and rotation, etc. received as a model of binary images have given ordinary objects-primitives - namely, geometric shapes, all trivial subjects in the same binary graphical form. The objective of the programme was to identify, or rather, the answer to the question: what it seems? That is, seeing Christmas tree, the program issued: "It resembles the letter A" and so on.

After otkatki this programme at the images of binary files with the transition to video processing has neproblematichnym. True, had povozitsya with optimization, because in the initial stages of algorithms to the main criterion was their reliability, ie, issuing the correct result in different conditions, but not speed extradition result. Optimizing expressed not only in the work on the algorithm itself, but in calculating the optimal frequency of cultivated cadres. Immediately, it had povozitsya, therefore, became the next question ...

Experience. Suppose I have already shown the triangle program, received a reply that it resembles the letter A. Then I show the same picture again and ... computations are performed again. And can a program to issue the result almost immediately, if tcpd results in their own experience as this makes people? Save the image in order to then make comparisons meaningless. This is exactly the same as store database with the value sinuses for different angles, rather than apply the formula. In addition, for comparison concrete block analysis is needed - even one additional algorithm and a lot of memory to store images.

The decision found quite easily. When the image is processed and correlated with mathematical models of letters, then each of the intermediate criteria appears certain value and then in the amount we receive the required percentage match. So, these factors can be used to base experience. That is, of them became drawn up a base. And, for example, the same picture with a triangle was shown at different tilt, with little change projections, that is, in each case, we have received all the new experiences and new facilities that are downloaded into the database, and with accuracy approksimirovanii experiment became successful, then there are some Results were evacuated when compared with the base.

Many now noted that this method is ineffective. Because a similarity calculation simply happen faster than calculating the similarity plus comparison with the base. That is why this idea is often otmetayut. And here again is worth recalling that the main task of experiments - to teach the computer to work with images. The selected model plus rates - the basis for a new mathematical model. A bit unclear? The computer determined that the triangle like a letter A with some errors. And last clearly defined in the base ratios. And if the computer will ask the question of what in fact he has identified? Are you using the word "triangle" and all! A new mathematical model to determine the ready. Moreover, when analyzed error rates (more precisely, matches) for the different figures, it turned out, and that among them (the values of these ratios) there is some mathematical dependence, which can be replaced by formula. Thus, an effective system of experience was still be created at the rudimentary level.

So, experiment fully completed, resulting in, among the main tasks was implemented and primitive learning system for pattern recognition.

Of course, if taking into account the way in which people developed, starting in infancy, it all comes from understanding primitives, the same geometric figures and 3D-objects. And since we originally took another lift-off, the experience with the letters can be considered a logical conclusion. And even withdrawn definition:

The image - is the object of perception, the main structural unit.

Computers subject to so-associative thinking, something that human truth, he needs to learn to summarize, look at every object. In the form, which shall II systems, software and various software and hardware systems now, machine intelligence will be unable to distinguish from wood pillar or from nearby homes. Also, referring to the images, is the need of abstraction, because the object can be considered as a brick, and brick house.

Later, I developed this program to a level of recognition of the text that was included in the next experiment.


Semantics

Semantics - that understanding the meaning of the word, their combinations, phrases, suggestions, paragraphs and so on. In our case, all quite understandable. In the engine - not quite. The most vivid example - the programming language. For English-speaking community, many of them do not cause difficulty in learning, because based on customary language. When you say the phrase: "If A equals B, then C is 10", the code is written almost the same, taking into account the syntax. That is all there is understandable.

But ... very often in the discussion of issues II and even in many books on this subject can be found, in my opinion, a mistake. Namely: there is no border between the machine (architecture, performing actions), which, in fact, understands only binary code, and II, which is nothing more than a program. Yes, you can recall a remarkable example of the Chinese room, but still, one way or another, people are communicating with the software. The objective of the latter is: to understand and make decisions, perform conceived. Another question is to understand the human machine is not at the level of a specific query language, and more expansion and in context. For example, the word "eat". On the one hand, the verb "is" in the plural in the past tense, on the other - a noun "spruce" in the plural. Incidentally, in the Russian language such "zabav" for the programmes there are a lot of interpreters. "We sat down under the spruce and fir":). In the future under the "machine" we will understand a set of software and hardware.

In any language of logic implies a subject and éto, is the same and in the same programming languages and scripts. The latter can write their own version of interpretation in almost anything. For example, you can easily create a program that can perform operations on a literal writing in Russian: "A letter Proiznesi", "Say the word" Aboriginal ". Internal structure implies the transfer of the transition to standard commands, while distinguishes type" letter "of the type "floor" and after "proposal" and "text". Option scoring algorithm is achieved through TTS (text-to-speech - text-to-Speech), which was successfully implemented. Besides, the programme described in the previous section, was possible to read the text of a normal sheet of paper. It does not matter, take photographs, he was filmed on a camcorder. That is, the computer read the text, and this is now no surprise.

After the time has come to the application and STT (speech-to-text - described in the text), that is the job can make a voice. Think about whether their actions program? At least he understands that it wants. In doing so, have been included options for the same ugadyvaniya simple images at a level of humanity - the program can say: "I can not parse", "Maybe it ..."," Most of all it reminds ...". Naturally, everything inside was simply the result as a percentage, but people do not behave the same?

And we have a specificity of thinking, which involves attempts to explain the inexplicable things through clear and association. Suffice it to recall the famous face on Mars (rock, which somehow reminds person). If the computer database will be data on the structure of the human person, he will hold the same association.

In general, for further experiments opened an entire area, which investigates the properties of Integral II. For example, we read the text, at a minimum, operate or words, saying otherwise, images or objects of perception.

To date, test module verbal productions of the elementary mathematical tasks. For example: "Build a rectangular triangle with katetami three centimetres and four centimetres. What would equal hypotenuse?". That is the task given literally and in the voice mode, in the artificial semantics provides a certain subject area - mathematical, therefore, for example, the word "three" not associated with the verb of "friction", but with numbers. In the future, calculated to take two other subject areas (chess and economics), the machine would automatically determine to which of them to relate their perception. For example, in ozvuchennom sentence she could understand that we are talking about math because of the availability of "Triangle," a rectangular triangle "," katet "," centimeter "," hypotenuse. " That is, it must first review the text at several proposals, and after taking up challenges.


In conclusion

As you can see, flood II can be anything, and in fact it can accept images and associate them with existing knowledge base. The only thing that can cause difficulties in implementation, this definition three-dimensional objects (need laser scanner system or laser scanners) and, more importantly, understanding the materiality of others. That is, we differ cardboard box from a wooden box, even if they have identical shape and size, but as explained by car? Ideas, of course, there are (at least the same sound, each material has its own specifics absorption), but is it necessary to explain?

Incidentally, when the possibility of creating algorithms for the recognition of handwriting, your humble servant used to understand the principles of one well-known program, I think, guess what. The first thing I did, it took a medical card (themselves know how to write doctors). A small experiment has shown that it almost does not lend decoding and at the programmatic level. Also, I found a picture with handwritten pages V. I. Lenin (formerly entire book of such photographs were published manuscript pages). The results were slightly better than those of doctors.

And, generally speaking, if I told you about all the actions made in our experiments in terms of scientific terminology II, you would have met with all the basic concepts underlying basis in this discipline.

1) "Get Money for Clicks" NameDrive.com - Fastest Growing Domain Parking Company in the World.
2) Search your domain name wishing to have! FREE DOMAINS - yourname.co.cc

No comments:

Post a Comment