Elephant problems in the research and development of pattern recognition

Elephant problems in the research and development of pattern recognition

Elephant is a fable from India, this story is about a group of blind people standing under an elephant to touch the different parts of the elephant, and then come to different conclusions, but no one can say that an elephant, in fact, in the pattern recognition practice, often can encounter elephant problems, the computer is like a blind person, There is no visual system like the human, its stupid incomparable, to recognize a thing is a very troublesome matter, and also often travel wrong, why? That is because so far we have not been able to design an animal-like identification system, we have designed a recognition algorithm is too limited, always from the local characteristics of things to judge a thing, However, it is not easy to take features, like the blind, he touched, he can hardly know when he will touch to determine.

Elephant problem is the problem we must face in the research of pattern recognition. Whether we take the recognition features to meet the needs of recognition, this is the prerequisite for us to analyze before the design of feature comparison algorithm, if the feature is not selected sufficiently, even if the feature comparison algorithm is designed to be more subtle and can not achieve the best results.

Recognition program testing also to avoid elephant problems, the establishment of test set must be based on the application, a comprehensive, large-scale collection of various possible test data, local test data, such as the local part of the elephant, can not judge the performance of the program recognition.

There is no absolute recognition rate, the computer gives people a sense of speed, stability, high accuracy, as if the recognition rate can not be absolutely difficult to understand, the identification problem is not a simple calculation problem, the recognition rate and the recognition algorithm design is good or bad, is a technical problem of artificial intelligence, The current artificial intelligence technology is far inferior to the human brain, even those small animals are inferior, not to mention people have to see the wrong time, for the computer that is often mistaken. For this reason, to do the identification technology to have the idea of excellence, they regard themselves as an elephant under the blind, to be realistic, modest attitude to improve the algorithm consistently.