According to foreign media reports, Facebook recently launched the ReAgent reinforcement learning toolkit, for the first time to achieve offline evaluation (policy evaluation) by collecting offline feedback.
ReAgent has now partnered with Microsoft Azure’s top AI platform.
Facebook AI Research recently introduced ReAgent, an enhanced learning toolkit for building decision-making AI that can receive feedback. ReAgent can score user behavior and use the user’s input behavior (such as clicking on recommended content) as training data.
ReAgent is a small C++ library that can be downloaded from GitHub and embedded in any application. The toolkit includes a set of introductory decision AI models, an offline module for model performance evaluation, and a platform for deploying AI into production using the TorchScript library.
Horizon is an intensive learning platform that was used in Facebook’s open source production in November 2018 and is now part of ReAgent.
Building a large-scale real-time decision making AI module becomes simple
Srinivas Narayanan, head of applied research, said at Facebook’s @Scale conference recently that ReAgent now personalizes billions of Facebook decisions, such as user notifications on Facebook and Instagram.
Facebook once mentioned in the blog: “ReAgent is the most comprehensive and modular open source platform for creating AI-based reasoning systems, and is the first to evaluate the strategy by collecting offline feedback. Improve the model.”
“ReAgent makes it easy to build large-scale real-time decision-making modules, while also making the creation and evaluation of research projects and strategies in production applications civilian (democraTIzed).
Working with Microsoft Azure
To continue to improve ReAgent, Facebook has released documentation for its application to cloud services such as Microsoft Azure. Microsoft Azure launched its own intensive learning service earlier this year.
In a conversation at the Transform conference this summer, OpenAI CTO Greg Brockman and chief scientist Ilya Sutskever believe that the core of future AI development is reasoning and explainability.
Google and Microsoft, both of the AI giants, have their own open source AI development platforms, such as Google’s Google Cloud PredicTIon API and Microsoft’s Microsoft Azure Machine Learning.
Facebook has also released its own open source AI toolkit. The threshold for research and development in the field of AI is constantly decreasing. Some small and medium-sized enterprises can join this competition, and this field is constantly injecting fresh blood and full of vitality.
Anyway, full competition, exchange and integration will have more inspiration.