Google's artificial intelligence has more rapid learning skills
The newly discovered technique can help artificial intelligence achieve 87% of the performance of the outstanding players in the game.
According to Engadget, When it comes to artificial intelligence (AI), the value of every achievement is memorable. That's especially true for Google 's DeepMind who has proven itself by defeating world champion Lee Sedol, imitating human voices and reducing server electricity bills.
From DeepMind
Currently, a research group has announced a new "learning enhancement" method on the AI platform to speed up self-training without the need for direct human teaching.
First of all, DeepMind has better grasp the control of pixels on the screen. Google has noted that "similar to the way a child learns to control his hands by moving them and observing the movement". In this way, the child can find the best way to get high scores and participate in more effective games.
DeepMind has better grasp the control of pixels on the screen.
. to UNREAL
In addition, DeepMind can now find rewards from a game based on past practices."By reviewing the history of rewards more often, DeepMind is able to discover predictive features that are much quicker , " Google said. In an article, the company calls the concept of all of AI's capabilities under the heading Reinforcement Learning with Unsupervised Auxiliary Tasks.
These skills, along with DeepMind's previous intensive deep learning methods, created the UNREAL (UNsupervised REinforcement and Auxiliary Learning). That is only a small part, the biggest achievement is that DeepMind is starting to "teach itself like a human being". The team describes the DeepMind methods used similarly to how animals dream about positive and negative events (although they don't really know DeepMind has learned how to "dream").
In a 3D labyrinth environment called Labryinth , Google said UNREAL could still learn stages about 10 times faster. It can achieve 87% of the performance of an outstanding player in the game, and about 9 times the average player in a series of Atari games.
Overall, UNREAL has helped DeepMind significantly. But the team will have to wait and see if the achievements can be useful in outside game situations?
- What is artificial intelligence, deep learning, machine learning?
- Artificial intelligence has learned how to create another artificial intelligence
- Artificial intelligence: Development history and future potential
- This AI system will create images from the input text
- Artificial nerve coupling is self-learning
- What if terrorists get artificial intelligence?
- Congratulations to the designers, the people who are doing the job are hard to rob the robbery
- Apple first published research on artificial intelligence
- What is Artificial Intelligence? What is AI (artificial intelligence)?
- The first artificial intelligence possesses 'curiosity' just like humans
- Project Loon uses artificial intelligence to fix the location, catching the wind to fly high
- Artificial intelligence has been able to translate languages without a dictionary