Facebook uses these tools for optimizing Instagram’s back-end infrastructure and improving the response rates of user surveys.
Facebook has launched two new open source AI tools – Ax and BoTorch – to simplify artificial intelligence (AI) model optimization.
BoTorch, which is based on PyTorch, is a library for Bayesian optimization, while Ax is a general-purpose platform for managing, deploying and automating AI experiments.
These tools have been used by Facebook for optimizing Instagram’s back-end infrastructure and improving the response rates of user surveys.
As claimed by Facebook, BoTorch makes AI model optimizing easier and faster for data scientists to get to a production-ready model. On the other hand, Ax provides all the management functions around BoTorch’s ability to find the optimal configuration for these models and allows developers to focus on getting their services production-ready.
The Facebook AI team said that unlike BoTorch, the other existing Bayesian optimization tools are difficult to extend and customize, and that they didn’t meet Facebook’s needs.
Ax is framework-agnostic
Ax can automatically pick the best optimization strategy, either it could be Bayesian optimization, bandit optimization (another classic optimization strategy) or another algorithm. That means Ax is framework-agnostic.
While it uses BoTorch, researchers can also plug in their own code using the services PyTorch and NumPy interfaces.
Joe Spisak, Facebook AI’s product manager for PyTorch, noted that they decided to open source BoTorch because they wanted to collaborate with the top researchers in the field. They worked with Cornell University for the first release.
(With inputs from TechCrunch)
