This article is produced by NetEase Smart Studio (public number smartman 163). Focus on AI and read the next big era!
[Netease Smart News December 12 News] Five years ago, researchers had a sudden but significant leap forward in the accuracy of the software that could interpret images. The artificial neural network behind it underpinned us in the artificial intelligence industry. Seen prosperity. However, we are still far from the reality described in The Terminator or The Matrix.
At present, researchers are trying to focus on how to teach the machine to do one thing to the extreme. Unlike brains where the human brain processes multiple things at the same time, robots must "think" in a linear manner. In any case, artificial intelligence has defeated humans in some areas. Deep neural networks have learned to talk, drive cars, play video games, play chess, draw pictures, and help discover scientific discoveries.
In the following six areas, artificial neural networks have proven that they can surpass human intelligence.
1. Image and object recognition
Records indicate that machines are far more capable than humans in terms of image and object recognition. In an experiment that tested the software's ability to recognize a toy, Geoff Hinton invented the Capsule Network with an error rate that was almost half that of the previous minimum error rate. Increasing the number of capsules during different scans allows the system to better recognize an object, even if this view is different from what was previously analyzed.
Another example comes from a state-of-the-art network. It is trained on a tagged picture database and can classify objects better than a doctoral student. These doctoral students received more than the same tasks. 100 hours of training.
2. Video games
Google’s DeepMind uses a deep learning technique called “deepenhardening learningâ€, in which researchers use this method to teach computers to play Atari’s brickbreaking game Breakout. They did not teach or program this computer in any particular way. Instead, it controls the keyboard while looking at the score. Its goal is to get the highest possible score. After playing for two hours, the computer became an expert in this game.
Deep Learning Community is conducting a competition to train computers to beat humans in almost every game you can think of, including Space Invaders, Doomsday, Pong, and World of Warcraft. In most games, deep learning networks have outperformed experienced players. Computers do not play games by programming. They only learn by playing games.
3. Speech generation and recognition
Last year, Google released WaveNet and Baidu released Deep Speech. Both are deep learning networks that automatically generate speech. These systems have learned to imitate human voices, and their levels have continued to increase over time. It is much more difficult to distinguish their speech from real people than people think.
University of Oxford and Google’s DeepMind scientists created an in-depth network. LipNet achieved 93% correct reading of people's lips, while ordinary human lips readers only achieved 52% accuracy. A team from the University of Washington used lip synchronization to create a system that synchronized synthesized audio with existing video.
4. Imitation of artwork and style
Neural networks can study the strokes, colors, and shadows of a particular artwork. Based on this, it can transform the original artwork into new images based on analysis.
DeepArt.io is an example. This company developed an application that can use deep learning to learn hundreds of different styles that you can apply to your photos. Artist and programmer Gene Kogan also used style conversions to modify the portrait of Mona Lisa based on the style that the algorithm learned from Egyptian hieroglyphics.
5. Prediction
Stanford University researcher Timnit Gebru selected 50 million Google Street View images to explore what a deep learning network can do. As a result, computers have learned to locate and identify cars. It detected over 22 million cars, including their manufacture, model, size and year. One of the insights gained from this system is where the starting and ending points of the constituency are. According to this analysis, "if the number of cars encountered in a 15-minute drive is higher than the number of pickup trucks, then the city may vote for Democrats in the next presidential election (88% chance)," said Timnit Gebru. His co-author writes.
Another example of a machine that provides more accurate predictions than humans comes from Google's Sunproof project, which uses aerial photographs from Google Earth to create a 3D model of your roof, separating it from the surrounding trees and shadows. Then, it uses the sun's trajectory to predict how much energy your roof's solar panel can generate based on the positional parameters.
6. Website Design Modifications
The artificial intelligence integrated in the Website Builder can help the website to update and make useful changes to the website, faster and more accurately than humans. The basic technology of this kind of system provides the general user's opinion regarding the appearance of the website, this can tell designer's website design is good or bad. Today, website builders either use deep networks to modify their designs or plan to use them in the near future. The technology can analyze different models and create better results based on previous conversion rates and other important indicators.
Although we are still a long way from achieving matrix-level artificial intelligence, companies are striving to rapidly increase the level of intelligence in neural networks. The above-mentioned project is only a shallow application of this technology. New ideas and improvements are constantly emerging, proving that the machine is constantly surpassing human performance in completing tasks.
(Selected from: VentureBeat compilation: NetEase see outside smart compiler platform review: Li Qing)
Pay attention to NetEase smart public number (smartman163), obtain the latest report of artificial intelligence industry.
Electrical Insulation PVC Tape
Electrical Insulation Pvc Tape,Pvc Electrical Tape,Pvc Electrical Insulation Tape,Best Electrical Insulation Pvc Tape
Longkou Libo Insulating Material Co.,Ltd. , https://www.liboinsulation.com