Google has announced the release of DeepMind Lab project which is available under the terms of open source licensing for the AI community. The tech giant made this announcement yesterday during the Neural Information Processing Systems(NIPS) conference which was held in Barcelona.
With the help of DeepMind Lab, computing platforms such as Virtual Reality(VR) and Artificial Intelligence(AI) can be bought together in multiple dimensions. To achieve this DeepMind Lab uses its own AI called Machine Learning(ML) which then consists of an even more advanced type of ML called Deep Reinforcement Learning (DeepRL).
Google had also created agents that had defeated a world champion in the game ‘Go’ and was also capable of playing Atari 2600 games by itself with amazing skills. All of these were achieved as the DeepRL agents had the ability to learn from the previous experience and was able to apply it to new situations, these can be called as supervised Machine Learning.
However, these type of Machine Learning needs some labeled data sets in order to perform and learn, hence learning without supervision is the next step. This is where DeepRL comes in as it is capable of learning things like a human or an animal.
Games can be the main framework for the DeepRL due to the inherent progressive score which acts as a reward. The agent will be able to learn from the past success and failure of a particular game and will be able to assess the probability of its next move, by taking the inputs from past experiences. It was also stated that DeepRL which is trained on 3D VR games can be used for the understanding of 3D reality.