Google has a DeepMind Lab open AI training platform

On December 6th, according to Bloomberg News, Google DeepMind, an artificial intelligence division of Alphabet, will open its labyrinth game platform for many experiments to other researchers and the public.

DeepMind said on Monday that it will release the entire source code of its training environment on the open source community GitHub. The training environment, formerly known as Labyrinth, is now renamed DeepMind Lab. Anyone will be able to download the code and customize it to help train its own artificial intelligence system. They will also be able to create new game levels for DeepMind Lab that can be uploaded to GitHub.

Opening the AI ​​test platform to the public is another example of DeepMind's decision to open up its research. Last month, the company announced a partnership with Activision Blizzard to turn the popular video game Starcraft II into a test platform that any AI researchers can try to build. AI system for complex games.

Shane Legg, one of DeepMind's co-founders, said in an interview that putting the DeepMind Lab code on GitHub will let other researchers know if its developer's own breakthrough can be replicated. These scientists will be able to evaluate the performance of their own AI agents in the tests used by DeepMind. "If they want to, they can try to beat us," he said.

Tesla CEO Elon Musk, well-known venture capitalist Peter Thiel and Silicon Valley Venture Accelerator Y Combinator founder Sam Altman co-founded the AI ​​Research Organization In April of this year, it opened its own AI training platform OpenAI Gym.

On Monday, it also announced that it will unveil an interface called Universe that allows AI agents to "use computers like humans: by watching screen pixels and operating virtual keyboards and mice." Simply put, it is for AI systems to learn A mediator who plays games or runs the skills needed for other applications. Researchers can use OpenAI Gym's tools to measure the performance of these systems.

When Universe is launched, it will offer 1,000 different game and training environments. OpenAI previously called on video game developers to allow OpenAI to be included in their games.

Some people think that DeepMind's decision to open DeepMind Lab to the public is because of competition with OpenAI, or because Google has been criticized for its own AI breakthrough. Legge denied this. "The machine learning research community has always been very open. We have published 100 research papers each year, and we have previously opened up a number of our own AI agents."

DeepMind is known for creating AlphaGo, the world's Go champion Li Shishi. This achievement is considered to be a major breakthrough in the field of computer science. Because Go is very complicated and has too many moves to be grasped by brute force. Therefore, the AI ​​system must have the intuition of human beings to make good Go.

AlphaGo is not trained on the DeepMind Lab platform, but the company has its DeepMind Lab environment for several leading-edge projects. In those projects, the system must have the ability to sense, remember, plan, and navigate. They include a recent trial in which the company's researchers drastically reduced the time it takes to train AI agents to shuttle through the game environment and look for digital apples scattered throughout the maze.

Legg said that DeepMind Lab is better than other AI training environments because the game environment is more complicated. DeepMind Lab's AI agent controls hovering through "first person" vision. Legge pointed out that AI agents can look and move in any direction, but other AI training environments can't do this.

Flange Sensor

Flange Sensor

Flange type NTC temperature sensor we made with high precision, reliable NTC Thermistor connected to qualified FEP, PVC insulated wire, heat conductive and water resistant materials into custom shapes so as to facilitate long-distance temperature measurement and control. Our sensors that have beautiful appearance and fix convenient have been applied for dryer, rice cooker, automotive air conditioner, water heater, toaster, water closet and more. Temperature range of this sensor is from -30°C to 200°C.


Flange Sensor,Washing Machine Sensor,Toaster Sensor,Rice Cooker Sensor

Feyvan Electronics Technology Co., Ltd. , https://www.fv-cable-assembly.com