Science and Technology Daily Beijing December 5th ç‰îŸ‡ ç‰îŸ‡ ç‰îŸ‡ å“’ç¶ å”¬â”• 唬┕ 唬┕ 㶠㶠㶠㶠㶠㶠㶠㶠ã¶Your own AI. According to recent news from Google’s official blog and Futurism News Network, the “sub AI†created by AI has beaten the AI ​​created by humans: In the test, the “Sub AI†system named NASNet has reached the correct rate. 82.7%, 1.2% higher than the results of the same AI products previously announced, and the system efficiency is 4% higher. In May 2017, researchers at Google Brain announced the development of automated artificial intelligence AutoML, which can generate its own "sub-AI" system. A few days ago, they decided to launch the biggest challenge to AutoML so far - try to use AutoML to create their own AI, to defeat the AI ​​designed by humans. Team members use a method called reinforcement learning to automate the design of machine learning models. This time, AutoML's "identity" is a controller neural network that develops a "sub-AI" for specific tasks. This newly-created "child" is called NASNet and can identify humans, cars, traffic lights, handbags, and backpacks in real-time. AutoML as a "parent" will evaluate the performance of "children" NASNet and use this information to improve "sub-AI" and repeat this process thousands of times. The team members tested the "sub-AI" NASNet on ImageNet (the computer vision system identification project, which is the world's largest image recognition database) image classification and COCO target recognition two data sets. They said that this is the two most recognized large-scale academic data sets in the field of computer vision, and its magnitude is so large that the test is very serious. As a result, in the ImageNet test, NASNet's accuracy rate on the verification set reached 82.7%, which is 1.2% better than the results of similar artificial intelligence products previously announced, and is comparable to the unpublished results reported on the paper's preprinted website. The system efficiency is increased by 4%, and the average accuracy of the largest model is 43.1%. Team members stated that NASNet will be used for various applications, and users can use this AI system for image classification and object detection. Editor's Editor Robots can build robots and AIs can design AIs. It's no surprise to think about it. As long as the goal is clearly defined, a powerful computer is certainly faster than the human brain and will sooner or later replace people. But this does not mean that AI can escape from self-improvement. Because AI was still stuck in a cage, it was occasionally put on the track and ran. When AI suddenly thinks about it and sets a goal for himself, then when can it compare to people? There is still a long way to go. Banknote Value Counting Machine Two CIS Banknote Value Counting Machine,Banknote Value Counting Machine Two CIS,Two CIS Suzhou Ribao Technology Co. Ltd. , https://www.ribaoeurope.com