On January 20th, the strongest brain in the fourth quarter came to a close. This season, due to the addition of a special member to the player, the AI ​​robot from Baidu was in a three-game "man-machine war." In the two wins and one draw, this result completely crushed the three human players who represent the limits of the human brain. Every detail in these three games has become the focus of hot discussion on social media. On the one hand, the public is amazed at the power of artificial intelligence, or it will replace human beings; on the other hand, it is also mixed with too many conspiracy theories. Emphasize the unfairness of this man-machine war.
If the former cause is derived from technology, especially the ignorance of the development of artificial intelligence; then the latter sound is not only the ignorance of artificial intelligence, but also the ignorance of human existence and future, which is stupid.
The chessboard and the human brain are the rulers of artificial intelligence.
In fact, despite the fact that the field of artificial intelligence was so hot in 2016, there were only three shocking events in the field of artificial intelligence from January 2016 to the present one year:
• On January 24, 2016, the artificial intelligence pioneer Ma Wensky died;
• On January 27, 2016, Google DeepMind published a paper in Nature, officially announcing the cracking of Go;
• On January 20, 2017, Baidu's artificial intelligence robot defeated the third human player in a small degree and won the victory in the three-game man-machine battle;
The reason why the three events of time and space are not related to each other is that the three events are the end and a new era of artificial intelligence. Since more than 60 years ago, whether it is Minsky or several other pioneers of artificial intelligence, the first question facing these top scientists and mathematicians is: How to measure artificial intelligence?
In the 1920s, American psychologist Louis Lean Thurstone found that respondents were more likely to answer questions of relative or comparative significance when answering questions, such as questions like "Which picture do you prefer, A? B?" is much easier and simpler than simply answering "How much do you like A painting?" This theory is called "Law of ComparaTIve Judgement". By letting people compare two of the multiple objects at a time, one can finally calculate the measurement score (distance scale) for each object.
The scope of its application is very wide, and artificial intelligence researchers are no longer worried about defining "intelligence". They only need to put the machine and human being in the same environment to continue the game, using human intelligence to measure the intelligence of the machine. Chess games are first used to test the intelligence of machines because chess games are a kind of "perfect" information game. For players, whether they are human or machine, the information they face is transparent and equivalent. - It is just a chessboard and a chess piece.
This sly plot began in 1956 when IBM engineer Arthur Samuel created an app for checkers and used reinforcement learning to train the app. In 1962, Arthur Samuel's checkers program defeated Robert Nealey, the nation's strongest amateur at the time.
The next two most intriguing stories are Kasparov and the Dark Blue Century War and Li Shishi Battle AlphaGo. With the development of mass media such as TV, Internet, and social media, people all over the world have seen East and West. The top characters in the two major chess classes bow down and admit defeat.
Artificial intelligence has proved its ability in chess and Go, and the strongest brain that challenges humans has become another yardstick for measuring artificial intelligence.
In the strongest brain competition, the three games covered face recognition, speech recognition and video (motion blurred image) recognition. These "skills" are formed by the long-term evolution of human beings. Baidu chief scientist Wu Enda explained human face recognition ability: "A 3-year-old child sees her mother, regardless of whether her mother is smiling, angry, squinting, closed eyes, long With hair, short hair, and what to wear, children can easily recognize that this is a mother."
More importantly, this recognition of human beings is almost instantaneous, and even today, the world's top scientists cannot understand the true principles behind this. To make computers have this ability, scientists have put forward a lot of ideas in the past 50 years, but until these years, image recognition has truly achieved technological breakthroughs.
Similar to image recognition, the technological development process of speech and dynamic image recognition has also undergone a long process. How much artificial intelligence technology relying on the new algorithm is far from the human brain, especially those with super-powerful human brains. The competition provided the best observation angle, and the results of the competition fully demonstrated that artificial intelligence has surpassed humans in some areas.
Current artificial intelligence only focuses on specific areasIn the early years, Li Yanhong said after attending the "Best Brain" program as a guest: "Some things that are difficult for human beings are very simple for computers." This is not an exaggeration, such as the field of Go, Compared with the growth speed of human chess players, the evolution speed of computers is "suffocating". The Master who swept the masters of China and South Korea at the end of 2016 is also an evolutionary version of AlphaGo. It is necessary for Alpha Go to "go into the professional chess session" but only a little more than a year. At the time, this speed of learning and evolution is beyond the reach of human beings.
The current boom in artificial intelligence has benefited from the rise of deep learning over the past few years. The main focus is on three areas: image recognition, speech recognition, and natural language processing. Investor David Kelnar provides two comparisons of image recognition and speech recognition evolution speeds:
Taking image recognition as an example, in the 2012 image classification competition ImageNet, the research team supported by deep neural network won the first place, and the error rate was reduced to below 20%, allowing giants such as Google and Facebook to Shocked, Google later bought the team and also let Geoff Hinton, the deep neural network "Godfather", enter Google. With the help of deep neural networks, Google's image recognition level has been greatly improved, and the error rate has been reduced to less than 10%.
In terms of speech recognition, in the “Top Ten Breakthrough Technologies in 2016†selected by MIT Technology Review, ConversaTIonal Interfaces was successfully selected, and it is necessary to make machine dialogue effective. If you understand people, you should also make appropriate feedback. The MIT Science and Technology Review believes that the speech recognition engine developed by Baidu Silicon Valley Lab, Deep Speech 2, has a large deep neural network. Based on end-to-end deep learning technology, it can learn how to connect sounds and sentences based on millions of transcription language libraries, and the speech recognition rate is extremely accurate. The current speech recognition accuracy rate is 97%.
iDealTek-Electronics has accumulated rich application experience and design technical knowledge in the field of High-voltage power supplies, from Linear High Voltage Power Supplies based on power frequency transformers to Switching High Voltage power supplies based on IGBT components, from 60KV, 1KW High Voltage Power Module, to 5KV ~ 40KV, 1KW 2U Laboratory High-voltage power supplies, and to 5KV ~ 60KV, 2KW 4U and 5KV ~ 100KV, 6U rack-mounted capacitor charging high-voltage power supplies, and then to floor-standing cabinet-type high-current high-power high-voltage power supplies, all our high voltage power supplies are featured for high reliability, excellent high-voltage output stability and low-ripple electronic characteristics.
Through the reliable and durable operation panel on the front panel of the high-voltage power supply, the output voltage and current can be easily set and controlled. The high-precision LED or LCD ensures intuitive and high-precision high-voltage output measurement functions. Our high-voltage power supplies can be equipped with a wealth of remote-control interfaces, such as RS232 / RS485 /DB15 / DB25 / DB50 analog signal interface, etc. for remote high-voltage enable and inhibit, high-voltage output control programming and data monitoring.
Our HV power supplies have complete built-in protection functions, such as overvoltage protection, ARC protection, load discharge protection, over heat protection, etc. The protection mechanism can start in transient time to ensure the safety and reliability of the power supply itself and customer loads under high-voltage output.
At present, our high voltage power supplies are widely used in high-voltage laboratories, capacitor charging, electron beam, ion implantation, FUSION power generation and other industries.
HV Power Supplies, High-voltage Power Supplies, High Voltage DC Power Supplies, HV PSU, High Voltage Power Module
Yangzhou IdealTek Electronics Co., Ltd. , https://www.idealtekpower.com