Google AI can alreadyinventionOwn encryption algorithm, now it can create their own AI. The Google Brain team was announced earlier this yearAutoML, AI can generate their own AI. Researchers try to create with AutoMLPerformance AI to defeat human AI. They used an enhanced learning approach to automate the design of a machine learning model. AutoML serves as a controller neural network to create a sub-AI network that performs specific tasks.
The task of this sub-AI, called NASNet, is to identify real-time targets such as people, cars, traffic lights, handbags and backpacks from the video. AutoML evaluates NASNet's performance and then uses the information it gets to develop improved versions that are repeated thousands of times. Researchers used NASNet to test the ImageNet image classification and COCO target recognition datasets. In ImageNet's test, NASNet achieved a correct rate of 82.7%, 1.2% higher than the latest results from humans, 4% higher efficiency and less computational requirements.