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
如何让人工智能AI解释自身行为? How can artificial intelligence AI explain its own behavior?
如何让人工智能AI解释自身行为?
How can artificial intelligence AI explain its own behavior?
Artificial intelligence has penetrated into every corner of our modern life, artificial intelligence can Facebook you to paste the name posted pictures of friends label, may help you choose to see in the picture on the Instagram, NASA and materials scientists and researchers began to use artificial intelligence to help find new science and space search.
However, there is a central achievement in this technology. No matter in social media or in the process of using Mars rover, the order maker of building it doesn't know the reason why AI makes every decision.
Modern artificial intelligence is still a new thing. In the past 5 years, the investment and Research on artificial intelligence of large technology companies has just come to the end. Before that, the decades - old practice of artificial intelligence eventually lost proof in 2012. The Enlightenment of the human brain, a group of artificial neural network depends on the "neurons" between the countless great cohesion, perhaps is calculated a string of small, cohesive system similar to that of neurons in the brain. But this software architecture has brought us a new trade-off: because millions of transitions are so complex and grand, researchers cannot accurately determine the outcome of the ongoing connection, and they will only lose one output result.
"With the machine learning in the society is becoming more and more widespread and increasingly high risk, people are beginning to realize we can't put the system as a reliable and fairly safe," Hanna? Wallach told Quartz in an email, he is a senior researcher at Microsoft, also is the spokesman for the conference. "We need to know what their internal attacks are, and how they are used."
Artificial intelligence to perform key obligations
In NASA's Jet Propulsion Laboratory, AI can enable Mars probes to operate semi autonomously on the surface of unknown planets. Artificial intelligence is also used to discontinue the process of carding thousands of photographs taken by the detector when it returns to the earth.
Kiri Wagstaff is an artificial intelligence researcher and spokesman of JPL. He said that because every decision brings huge risks, we need to understand every resolution before AI is applied.
"Suppose there is a spaceship on Mars orbit, which means that it is 200 million miles away, and it needs hundreds of millions of dollars, or even one billion dollars. Everything is hard to save, assuming what is going on. " "Without a lot of money," Wagstaff said, "there is no way to repair, visit, and exchange these things. Therefore, if we want to make the machine learning play a role, then let the machine perform these obligations of the people need to understand what it needs to do, why to do this behavior, because of the assumption that the robot doesn't know why to choose, why people will trust it to control their expensive or Mars orbiter?"
Wagstaff is trying to through a variety of NASA spacecraft in space to capture the image construction of artificial intelligence on their suspension of classification, because the number of these pictures can reach millions, so that artificial intelligence to identify humor photos in this huge database, then we can save a lot of time to find what we want to see these photos but the premise is that artificial intelligence an image of "humor" is what.
For Wagstaf, he felt that the purpose of understanding artificial intelligence was the key to the completion of a specific algorithm. Let's suppose that in executing machine learning process, there are computational errors in how to use images, which means that the obligation of data transfer is worth millions of dollars.
Wagstaff said: "when you see a computer to say 'this is very humorous, let me have a look at this picture", you will be in a uncertain shape, because you have seen the broken million photos, you don't know what this is humorous, perhaps this is why humor." "Is it because of its color, because of its shape, perhaps because of the space display order of the objects in the scene?"
Hidden knowledge
In 2007, Andrew Gordon Wilson, a professor of artificial intelligence at Cornell University and a team developed a new PET scanner. Because some particles do not behave like other ordinary particles on this machine, his duty is to track the process of a particle passing through a box of xenon.
His advisers suggest trying to use neural networks to stop monitoring, and neural networks are still a relatively vague concept in advance. Using this technique, Wilson can use the light emitted by the particle to locate it in the xenon.
After he lost the answer he wanted, Wilson said that using this algorithm to understand how light represents the internal rules of particle location, it will open a new way for the next discussion.
Wilson said, "to some extent, a model is the practice we observed. We can not only use models to stop prediction, but also better understand why this prediction is correct and how these natural processes work.
Unscramble the talent
But Wallach, a Microsoft researcher, says one of the biggest challenges is how to define it as complex as it is to open up a new range of interpretative skills.
To understand whether talents can mean that AI experts know why the algorithm of Facebook data will show a specific post to each group, or is it a way to let you know yourself? Use of artificial intelligence
However, there is a central achievement in this technology. No matter in social media or in the process of using Mars rover, the order maker of building it doesn't know the reason why AI makes every decision.
Modern artificial intelligence is still a new thing. In the past 5 years, the investment and Research on artificial intelligence of large technology companies has just come to the end. Before that, the decades - old practice of artificial intelligence eventually lost proof in 2012. The Enlightenment of the human brain, a group of artificial neural network depends on the "neurons" between the countless great cohesion, perhaps is calculated a string of small, cohesive system similar to that of neurons in the brain. But this software architecture has brought us a new trade-off: because millions of transitions are so complex and grand, researchers cannot accurately determine the outcome of the ongoing connection, and they will only lose one output result.
"With the machine learning in the society is becoming more and more widespread and increasingly high risk, people are beginning to realize we can't put the system as a reliable and fairly safe," Hanna? Wallach told Quartz in an email, he is a senior researcher at Microsoft, also is the spokesman for the conference. "We need to know what their internal attacks are, and how they are used."
Artificial intelligence to perform key obligations
In NASA's Jet Propulsion Laboratory, AI can enable Mars probes to operate semi autonomously on the surface of unknown planets. Artificial intelligence is also used to discontinue the process of carding thousands of photographs taken by the detector when it returns to the earth.
Kiri Wagstaff is an artificial intelligence researcher and spokesman of JPL. He said that because every decision brings huge risks, we need to understand every resolution before AI is applied.
"Suppose there is a spaceship on Mars orbit, which means that it is 200 million miles away, and it needs hundreds of millions of dollars, or even one billion dollars. Everything is hard to save, assuming what is going on. " "Without a lot of money," Wagstaff said, "there is no way to repair, visit, and exchange these things. Therefore, if we want to make the machine learning play a role, then let the machine perform these obligations of the people need to understand what it needs to do, why to do this behavior, because of the assumption that the robot doesn't know why to choose, why people will trust it to control their expensive or Mars orbiter?"
Wagstaff is trying to through a variety of NASA spacecraft in space to capture the image construction of artificial intelligence on their suspension of classification, because the number of these pictures can reach millions, so that artificial intelligence to identify humor photos in this huge database, then we can save a lot of time to find what we want to see these photos but the premise is that artificial intelligence an image of "humor" is what.
For Wagstaf, he felt that the purpose of understanding artificial intelligence was the key to the completion of a specific algorithm. Let's suppose that in executing machine learning process, there are computational errors in how to use images, which means that the obligation of data transfer is worth millions of dollars.
Wagstaff said: "when you see a computer to say 'this is very humorous, let me have a look at this picture", you will be in a uncertain shape, because you have seen the broken million photos, you don't know what this is humorous, perhaps this is why humor." "Is it because of its color, because of its shape, perhaps because of the space display order of the objects in the scene?"
Hidden knowledge
In 2007, Andrew Gordon Wilson, a professor of artificial intelligence at Cornell University and a team developed a new PET scanner. Because some particles do not behave like other ordinary particles on this machine, his duty is to track the process of a particle passing through a box of xenon.
His advisers suggest trying to use neural networks to stop monitoring, and neural networks are still a relatively vague concept in advance. Using this technique, Wilson can use the light emitted by the particle to locate it in the xenon.
After he lost the answer he wanted, Wilson said that using this algorithm to understand how light represents the internal rules of particle location, it will open a new way for the next discussion.
Wilson said, "to some extent, a model is the practice we observed. We can not only use models to stop prediction, but also better understand why this prediction is correct and how these natural processes work.
Unscramble the talent
But Wallach, a Microsoft researcher, says one of the biggest challenges is how to define it as complex as it is to open up a new range of interpretative skills.
To understand whether talents can mean that AI experts know why the algorithm of Facebook data will show a specific post to each group, or is it a way to let you know yourself? Use of artificial intelligence