Where did the history of "deep learning", which makes AI a reality, began?

The development of computer technology has been indispensable for AI (artificial intelligence) technology which has developed rapidly in recent years, but in fact it was advocated from an age that is surprisingly old enough. Such AI and make it possibleDeep learningThe history of the movie "History of Deep LearningIs looking back.

History of Deep Learning - YouTube


"Deep learning" is an alias, also referred to as "Deep Artificial Neural Networks", has the nickname of "neural network" is born from here.


Became the basis of the algorithm, is a model of neural network "neuron" in the human brain.


Its history is old, at least in 1943 it is believed that the concept had already existed. At that time it was still the era of the Second World War.


Neurons undergo electrical stimulation from another neuron exceeds a "threshold (vital blood)" constant, the neuron becomes excited, also referred to as "fire", a signal to emit an electrical signal called impulse I will communicate. This transmission layer (layer) has several layers, and this leads to the complexity of the brain.


It is a mechanism of neural network to reproduce this signal transmission as an algorithm. However, the number of still-layer about the time that this idea began been proposed can not be sufficiently reproduced, as an algorithm to reproduce the human brain was incomplete.


Therefore, now it is a computer scientist called "father of artificial intelligence"Marvin MinskyDr. also stated the limits of the technology at the time and was even showing a skeptical view.


Mr. Minsky clarified the limit after that, as its content was misunderstood / exaggerated, the first "beginning" in the mid 1970sAI's winter periodIt will be greeted.


However, even in such winter the end will come. It is that of the 1980s from the late stand 1970.


Nevertheless, at that time the processor's ability to use for learning was not high enough yet.


Evolution of AI is improper and improper in processor capacity. As a result, neural network experts constantly send daily demands for high performance processors.


However, a groundbreaking idea appeared.


It is a concept of "convolution neural network" used for image analysis and "recurrent network" to process information many times.


A typical example of this is letting computers learn handwritten numbers and make them recognize them.


This attempt was accomplished by Mr. Jan Lecan, who belonged to Bell Laboratories at the time.


Mr. Lecan is currently leading AI research at Facebook and New York University.


And in 2006, Jeff Hinton proposed a concept of "deep learning" with sufficient multilayers by combining shallow neural networks.


And the research team of Hinton et al. Will achieve remarkable results. In 2009, the team proved that neural networks can be learned sufficiently in just about three weeks, and established the most advanced speech recognition technology.


That ability is enough to accomplish the technology we have cultivated over the past 10 years in as little as 3 weeks. This brings a strong impact to the world of AI, and the era of "tertiary AI boom" comes.


Mr. Hinton joined Google in 2013 with recognition of its achievement. Also, Mr. Lecan belongs to Facebook as mentioned above. In this way, the cutting-edge deep learning researchers are advancing the world's most advanced research in private companies.


Accuracy of speech recognition technology by deep learning is extremely high, "precision 95%" is also superior to human ability.


In this way, the human race has made robots "eyes and ears" with the same or superior ability as human beings. It is considered that the evolution of AI will continue to accelerate rather than stay constant, and it is no longer a mistake that what AI that exceeds human ability will bring is beyond human imagination. It seems to be said that there is nothing.