Specialists from the Artificial Intelligence Institute AIRI have prepared and presented a database for training artificial intelligence (AI) to make independent decisions, the institute's press service reported.
The created training examples included information on 30,000 different tasks. They are adapted for use in the universal learning environment XLand-MiniGrid (a dataset for contextual and reinforcement learning).
The presented dataset is available to all scientists and developers and includes 100 billion fixed datasets. It can be useful not only for creating AI that will be much easier to configure for solving new tasks without the involvement of highly specialized experts, but also as a source of useful synthetic data.
The developers demonstrated the effectiveness of the database in the AI training process. It was built on the basis of two different approaches to contextual reinforcement learning - AD and DPT.
The result showed that the first neural network learned to make decisions based on the proposed solutions, while the second one failed to do so.
Earlier, www1.ru reported that the Higher School of Economics declared rules for using neural networks in the educational process.
Read more on the topic:
Russian researchers have trained a neural network to recognize people's faces from awkward angles