News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
Let the AI machine have a talent
Let the AI machine have a talent
Ian Goodfellow (Ian Goodfellow), who was known as the "father of GAN", was known as the top expert in the field of artificial intelligence.
.
Materials show that Goodfellow et al. In October 2014, in Generative Adversarial Networks, proposed a new framework to estimate the model through the estimation of the opposing process. In the framework, two models are exercised simultaneously: the generation model G for data dissemination, and the estimation model D for estimating the probability of the original self exercise number. The exercise program of G is to maximize the probability of D error, which corresponds to a game between two sides of the maximum set. It can be proved that there is a unique processing plan in the space of arbitrary functions G and D, so that G can reproduce the dispersion of exercise data, while D=0.5. When G and D are defined by multilayer perceptron, the whole system can be stopped by backpropagation. There is no need for any Markov chain or the expansion of the approximate reasoning network during the training or generation of samples. The experiment proves the potential of the framework through the qualitative and quantitative evaluation of the generated samples.
By letting the neural networks attack each other, Ian Goodfellow (Ian Goodfellow) invented the powerful artificial intelligence (AI) tools to give the machine an imaginary talent. Now, he and our others have to face the results of this tool.
One night in 2014, Goodfellow drank a drink with a newly graduated doctor. At Les 3 Brasseurs, a popular bar in Montreal, many friends pleaded for his assistance because they were developing a tricky project, the computer that could create pictures.
Generating the birth of the antagonistic network
The researchers used to use the neural network, that is, to simulate the lax model algorithm set up by the neural network of the human brain and to create my new data as a "generation" model. But the result is often not satisfactory: computer generated face images are often vague or appear to be missing such ears.
Goodfellow's friends put forward a plan to stop the complex statistical analysis of elements that make up photos, so as to assist the machine to create images. This requires a lot of digital operations, and Goodfellow tells them that this is basically not feasible.
But when he thought about this problem when he was drinking beer, he suddenly came up with a mind. What will happen if two neural networks are antagonistic? Friends are skeptical, so when he returned home, his girlfriend was asleep, he decided to try it. Goodfellow stopped coding in the first few hours, and then tested his software. He didn't expect to win the first time.
The technology developed by Goodfellow on that night is now called "GAN". This technology has aroused great excitement in machine learning, and has turned its developers into celebrities in the field of AI.
Over the past few years, AI researchers have achieved an impressive pause by using a technology called deep learning. Provide enough image to deep learning system, it will learn from it, for example, identify a pedestrian crossing the street. This way makes it possible for driverless cars and dialogs to drive Alexa, Siri and other virtual assistants.
However, deep learning can learn to identify things, but they are not good at inventing them. The purpose of GAN is to endow the machine with the natural imagination. In the future, computers will better enjoy raw data and calculate what they need to learn from them. This will not only enable them to draw or compose music, but also to reduce their dependence on human beings and learn and understand the world and their operation mode by themselves.
Often, AI programmers often need to tell the machine what to do in the data, for example, there are millions of pictures of pedestrians crossing the road. This method is not only high cost but also strong labor intensity. In addition, even if a slight deviation from the training, AI system will encounter twists and turns when disposing image data. In the future, computers will better handle the raw data and calculate the content they need to learn without being informed.
This will mark the great progress of AI's "no surveillance learning". Unmanned vehicles can understand many different road conditions in the absence of separate garage conditions, and robots can foresee obstacles that may be encountered in a busy warehouse and do not need to bypass it.
The magic of GAN lies in the competition between two neural networks.
Our ability to imagine and consider many different situations is an important part of us as human beings. In the future, when historians of science and technology recall today, they will probably regard GAN as an important progress in the invention of machine that has human cognition. Facebook chief AI scientist Jahn Lohe (Yann LeCun) called GAN "the coolest idea of deep learning in the past 20 years". Another AI big cafe and former chief scientist of Baidu (Andrew Ng) also said that GAN represents "important, basic progress", which will inspire the growing community of global researchers.
Goodluck is now a research scientist in Google Google Brain, which is located in Google headquarters of California mountain view. When I saw Goff de flow recently, he still seemed surprised by his position as "giant star", calling it "a bit incredible." Perhaps equally surprising is that he found that most of my time now is used to deal with those who want to use GAN to do evil.
.
Materials show that Goodfellow et al. In October 2014, in Generative Adversarial Networks, proposed a new framework to estimate the model through the estimation of the opposing process. In the framework, two models are exercised simultaneously: the generation model G for data dissemination, and the estimation model D for estimating the probability of the original self exercise number. The exercise program of G is to maximize the probability of D error, which corresponds to a game between two sides of the maximum set. It can be proved that there is a unique processing plan in the space of arbitrary functions G and D, so that G can reproduce the dispersion of exercise data, while D=0.5. When G and D are defined by multilayer perceptron, the whole system can be stopped by backpropagation. There is no need for any Markov chain or the expansion of the approximate reasoning network during the training or generation of samples. The experiment proves the potential of the framework through the qualitative and quantitative evaluation of the generated samples.
By letting the neural networks attack each other, Ian Goodfellow (Ian Goodfellow) invented the powerful artificial intelligence (AI) tools to give the machine an imaginary talent. Now, he and our others have to face the results of this tool.
One night in 2014, Goodfellow drank a drink with a newly graduated doctor. At Les 3 Brasseurs, a popular bar in Montreal, many friends pleaded for his assistance because they were developing a tricky project, the computer that could create pictures.
Generating the birth of the antagonistic network
The researchers used to use the neural network, that is, to simulate the lax model algorithm set up by the neural network of the human brain and to create my new data as a "generation" model. But the result is often not satisfactory: computer generated face images are often vague or appear to be missing such ears.
Goodfellow's friends put forward a plan to stop the complex statistical analysis of elements that make up photos, so as to assist the machine to create images. This requires a lot of digital operations, and Goodfellow tells them that this is basically not feasible.
But when he thought about this problem when he was drinking beer, he suddenly came up with a mind. What will happen if two neural networks are antagonistic? Friends are skeptical, so when he returned home, his girlfriend was asleep, he decided to try it. Goodfellow stopped coding in the first few hours, and then tested his software. He didn't expect to win the first time.
The technology developed by Goodfellow on that night is now called "GAN". This technology has aroused great excitement in machine learning, and has turned its developers into celebrities in the field of AI.
Over the past few years, AI researchers have achieved an impressive pause by using a technology called deep learning. Provide enough image to deep learning system, it will learn from it, for example, identify a pedestrian crossing the street. This way makes it possible for driverless cars and dialogs to drive Alexa, Siri and other virtual assistants.
However, deep learning can learn to identify things, but they are not good at inventing them. The purpose of GAN is to endow the machine with the natural imagination. In the future, computers will better enjoy raw data and calculate what they need to learn from them. This will not only enable them to draw or compose music, but also to reduce their dependence on human beings and learn and understand the world and their operation mode by themselves.
Often, AI programmers often need to tell the machine what to do in the data, for example, there are millions of pictures of pedestrians crossing the road. This method is not only high cost but also strong labor intensity. In addition, even if a slight deviation from the training, AI system will encounter twists and turns when disposing image data. In the future, computers will better handle the raw data and calculate the content they need to learn without being informed.
This will mark the great progress of AI's "no surveillance learning". Unmanned vehicles can understand many different road conditions in the absence of separate garage conditions, and robots can foresee obstacles that may be encountered in a busy warehouse and do not need to bypass it.
The magic of GAN lies in the competition between two neural networks.
Our ability to imagine and consider many different situations is an important part of us as human beings. In the future, when historians of science and technology recall today, they will probably regard GAN as an important progress in the invention of machine that has human cognition. Facebook chief AI scientist Jahn Lohe (Yann LeCun) called GAN "the coolest idea of deep learning in the past 20 years". Another AI big cafe and former chief scientist of Baidu (Andrew Ng) also said that GAN represents "important, basic progress", which will inspire the growing community of global researchers.
Goodluck is now a research scientist in Google Google Brain, which is located in Google headquarters of California mountain view. When I saw Goff de flow recently, he still seemed surprised by his position as "giant star", calling it "a bit incredible." Perhaps equally surprising is that he found that most of my time now is used to deal with those who want to use GAN to do evil.