As neuroscience reveals the importance of consolidating memory in dreams, Google's artificial intelligence (AI) company DeepMind has created a new technology that allows robots to dream, with the goal of improving learning efficiency. Not surprisingly, the starting point for AI technology dreams is mainly from video game scenes such as Atari.
DeepMind was the first to successfully make a video game including teaching AI to play ancient games such as Breakout and Asteroids. But the game referred to here is ultimately to let the robot dream, do the same thing with humans, and let the robot play an important role in the actual learning and information storage process.
To understand the importance of making robots dream, it is important to understand that dreaming is helpful to mammals (such as our own) brains. When scientists try to understand the role of dreaming from the perspective of neuroscience, they find that most of the content of dreaming is negative or threatening. You can try to do a dream diary for a month, and you will find that this is true.
It turns out that when human dreams of embarrassing things or threats appear, AI dreams of rearranging the various chapters of the game, and the process is reciprocating, but guiding the robot to dream is to let AI learn like experiments by humans. Use AI technology to guide robots to experiment and analyze different behavioral processes and their impact on the results.
So what are the challenges that robots may face? At present, the world's most advanced AI is only in the main video games, such as StarCraft II and Maze. Through "dreaming," AI can highlight some of the most challenging parts of the game, looping back and forth until it gains expertise, rather than meaninglessly rehearsing the game without affecting the player's score. Using this technology, DeepMind researchers are able to accelerate learning at 10 times faster. This speed may be faster as AI technology improves.
You may also ask why AI "dreaming" is necessary. Because robots already control human behavior in most games, such as chess and Go games. To grasp this, it is necessary to distinguish between using supervised learning (analysing data through AI and finding the appropriate model) and unsupervised learning. To date, most of the impressive technologies implemented through AI have been implemented using supervised learning, with "training data" provided by programmers, and AI learning to detect data patterns. This is a fairly simple method of training robots, but it is by no means a method of human learning. Instead, it uses an unsupervised learning method more similar to what programmers say. This type of learning takes more time than supervised learning because it involves the existence of a series of variables such as experiments.
Llc Transformer,High Frequency Llc Transformer,Low Frequency Llc Transformer,Low Frequency Power Llc Transformer
Huizhou Show-Grand Electronics Co., Ltd. , https://www.sgtransformer.com