Inventory of artificial intelligence applications that have hit the market in the Gengzi Year

By Super Nervous

Executive summary: Hypernervous has been paying close attention to and reporting on major events in the AI ​​world. What are the unforgettable AI events in the past year? We have summarized a list.

Keywords: 2020 AI event

In 2020, the application of AI will be more extensive. While exploring cutting-edge emerging industries, it will also go deep into traditional industries to help them transform. It is growing vigorously and gradually becoming an indispensable part of our lives. AI has almost become a household name.

As we begin a new chapter in the new year, let us review the past and review the achievements and development trends of AI in the past year.

AI takes you and me through

One of the hottest projects in the artificial intelligence world this year is the Weibo blogger @大谷Spitzer using AI to restore historical videos and character portraits.

First, in May of this year, he used the DAIN frame interpolation algorithm, ESRGAN, DeOldify and other AI technologies to color, repair the frame rate and super-resolution on a black and white video of old Beijing a hundred years ago, bringing the whole people through and feeling Beijing a hundred years ago. Street view.

Since then, he has successively carried out the "restoration" of some historical portraits. In December, I used AI to "brain" the portraits of emperors of the Song, Ming and Qing dynasties, and once again became a popular circle of friends. PaddleGAN, AI Studio, Artbreeder, and the face recognition technology SeetaFace developed by the Chinese Academy of Sciences have all contributed to these brain supplement work.

These technologies will also achieve more value in film and television, artistic creation, and historical and cultural dissemination in the future.

AI repairs the street view video of Beijing in the Republic of China, the fire is everywhere and the whole network will take you through

AI brain replenishes the appearance of Emperor Song, Ming and Qing, four models conceive real facial features

Use the robot to give instructions to the dog

Different types of dogs can undertake complex tasks such as search and rescue, navigation, transportation, companionship, driving, guarding, and house demolition in the society. So can they understand and obey the instructions of the robots?

The Social Robotics Laboratory of Yale University recently published a new study in IEEE’s "Spectrum". In the experiment, the instructions issued by the robot to the pet dog are more obedient than the instructions issued by the human to the pet dog (strange Knowledge points have been added. Jpg).

Yale University's new research publication IEEE: Dogs can better obey robot instructions

Paper address:

https://spectrum.ieee.org/automaton/robotics/robotics-software/dogs-obey-commands-given-by-social-robots

Mosaic becomes HD in seconds

What kind of experience is it to convert a “mosaic” pixel-level headshot into a high-definition photo?

Researchers at Duke University used AI algorithms to give the answer. The algorithm they proposed is called PULSE (Photo Upsampling via Latent Space Exploration).

The algorithm can convert fuzzy, unrecognizable face images into computer-generated images, with more detailed and realistic details than ever before, and even hair strands and crow's feet are clearly visible. With PULSE's brain replenishment ability, it is not a dream to say goodbye to touching image quality.

Duke University proposes an AI algorithm to save the slag image quality mosaic into high-definition in seconds

Paper address:

https://arxiv.org/pdf/2003.03808.pdf

Machine learning predicts the risk of physics failure

The new year is over, which means that the final exam results are also available. Can a physics class that has troubled a large number of students pass it smoothly? Machine learning can predict in advance.

In August, West Virginia University and the California Institute of Technology academics, a study published in the arxiv.org "Using Machine Learning to Identify the Most At-Risk Students in Physics Classes", made by machine learning algorithms, according to the usual Classroom performance, homework scores, etc. can be used to evaluate the graduation scores of students in the basic physics course.

However, what is embarrassing is that its prediction accuracy is almost the same as the random probability, only 57%... So, whether you will fail the subject or not, it is up to your own feeling!

Before the final exam, the algorithm said that my physics must fail

Paper address:

https://arxiv.org/abs/2007.13575

Change your face in five minutes, perform a divine comedy

Some time ago, the Up masters with perfect skills on the Xiaopo station used the AI ​​face-changing project of the "first order motion model" to generate a variety of videos with refreshing style.

Let Jacky Cheung, Du Fu, Tang Seng, and the panda head emojis sing "damedane" and "unravel" with eloquence and affection...

Damedane at Station B: The essence is to change your face, you can learn it in five minutes

With this face-changing artifact, you can give any character any expression, even the character in the painting. Moreover, the model is open source, everyone can have it, and it can be learned in five minutes. Here is the tutorial:

https://openbayes.com/console/openbayes/public/containers/BwZQj5wr3Jp

Paper address:

https://arxiv.org/pdf/2003.00196.pdf

Github address of the original project:

https://github.com/AliaksandrSiarohin/first-order-model

Smart phone test for drunkenness

As we all know, you don’t drive when you drink, and you don’t drink when you drive. But some people often just can't control themselves.

In order to detect whether the old drivers are drunk, the lead researcher Brian Suffoletto of the University of Pittsburgh School of Medicine, in the paper "A Preliminary Study Using Smartphone Accelerometers to Sense Gait Impairments Due to Alcohol Intoxication", introduced the method of using smartphones to detect whether they are drunk. The test subjects only need to take a smartphone and take two steps to detect their breath alcohol concentration.

If you’re drunk or not, just take your smartphone and take two steps to find out

Paper address:

https://www.jsad.com/doi/pdf/10.15288/jsad.2020.81.505

Looking back on the achievements of AI this year, it turns out that it is indeed almost omnipotent. Some people say that in the past year, the development of AI is equivalent to the past three years. I believe that under the "troika" of computing power, algorithms and data, AI will also achieve more amazing developments. Let us continue to pay attention to what surprises and breakthroughs AI will bring to the world in 2021.

Scan the QR code to join the discussion group

Get more high-quality data sets

Understand the application of artificial intelligence

Follow Top Meetings & Papers

Reply to "reader" for details

More exciting content (click on the picture to read)

Guess you like

Origin blog.csdn.net/HyperAI/article/details/113855796