Loading...
Qilu.com Lightning News, May 2 With the rise of artificial intelligence technology, a new "digital profession" - artificial intelligence trainer has quietly emerged. How to make AI "better" understand the world, today enter this new profession and talk about the story of training AI.
During the May Day holiday, many people drove to Zibo to eat barbecue, and the traffic flow of Internet celebrity check-in places such as the Eight Bureaus and Haidailou surged, and drivers found that the frequency of green lights at intersections here was more than that in other places. Si Si, a tourist from Hebei, said: "Following the prompts of the Zhixing Zibo APP, the green light passed through several intersections, although there were many cars on the road, but the passage was very smooth. ”
Behind the smooth flow is because the traffic light has a "smart brain". The local government has introduced a large AI model into the traffic signal system, using 5G technology to collect traffic flow, traffic lights and other road condition information in real time, and AI can automatically control the traffic lights after judging the traffic conditions. Sheng Xiaoling, deputy director of the artificial intelligence department of the research and innovation center of Isaac Technology Co., Ltd., said: "For example, there is a section of road that is actually empty, but it also has a red light of about 30 seconds, and this red light can be canceled at this time, which will greatly improve the operation efficiency of transportation." ”
For AI to judge the size of the traffic flow, it must first let it identify the motor vehicle at complex intersections. Instead of distinguishing with the naked eye, the engineer first had to extract the frame of the intersection surveillance video and digitize all the information on the photo. Sheng Xiaoling, deputy director of the artificial intelligence department of the research and innovation center of Isaac Technology Co., Ltd., said: "Some of the main characteristics of this car, including color, model, brand, year model, and some movement directions and trajectory, will be turned into some numbers like 0 and 1 for structured analysis, and then stored in the database." ”
"Hey" the data teaches the AI to recognize, and then it processes the data automatically. However, in complex traffic environments, AI needs more training to control signals. Engineers input a large amount of historical traffic data, allowing the AI to accumulate experience and perform simulations. Sheng Xiaoling, deputy director of the artificial intelligence department of the R&D Center of Isaac Technology Co., Ltd., said: "In most cases, training is not a yes or no judgment, it is a score, it is a process of continuous trial and error, and continuous experimentation. ”
With the repeated scoring training, the AI's judgment of traffic lights is becoming more and more accurate. From improving an intersection to optimizing the traffic of the entire city, the massive amount of road condition information can sometimes make AI "indigestion". Liu Xiaowen, an artificial intelligence big data development engineer at Isaac Technology Co., Ltd., said: "We need to clean the data to distinguish and classify the data of motor vehicles and non-motor vehicles, and we only need to feed the data of motor vehicles into the large model, which takes up fewer resources and can be calculated faster." ”
At present, this AI model has autonomously controlled 1,035 traffic lights in Zibo, and the local road traffic efficiency has increased by 28%. This group of young people has made AI understand the world better, and they have also grown up with AI, and their careers are more professional in subdivisions, from text trainers to picture trainers to video trainers, more and more young people are choosing this promising industry. Sheng Xiaoling, deputy director of the artificial intelligence department of the R&D Center of Isaac Technology Co., Ltd., said: "With the help of AI, our lives may be easier and easier. ”