Facebook announced the launch of the open source version recommended depth learning model (DLRM), which is one of the most advanced personalized recommendation AI model, and can be used in a production environment. This model can be used to achieve PyTorch Facebook's, Facebook's distributed learning framework Caffe2 and Glow C ++.
Recommendation engine to a large extent determines what people see every day, whether it is the content of social media sites such as Facebook, Amazon and other e-commerce sites, or recommend the game on the Xbox home page. Just last month, Amazon will also apply to AI personalized shopping recommendation system on AWS.
The end of May, more than 20 Facebook AI researchers reported in an article on arXiv paper , we explain how to use the embedded table mapping model DLRM categorical data were expressed, most of which calculation is performed by the prediction function Multilayer Perceptron (MLP). DLRM paper introduces the model and compared with the existing model is recommended to fully display its properties.
Facebook Artificial Intelligence Research (FAIR) has a lot of work to its open source, it can help provide DLRM broader AI community to address the challenges brought about by the recommendation engine, such as the use of neural network to classify the data and some of the higher-level attributes associated free.
Manufacturers DLRM recommendations for speed and accuracy recommendation engine performance benchmark with the model. DLRM benchmark for performance evaluation experiments and are written in Python, and synthetic supports random input.
Facebook scientists Dheevatsa Mudigere and Maxim Naumov said in a blog post, the results will be shared publicly DLRM optimize the performance of the system in the future.
Facebook in recent weeks by the AI model or other open-source framework also includes PyRobot, robot frame work with PyTorch; and PyTorch Hub, a workflow and API, designed to encourage the reproducibility of the AI model. There Ax and BoTorch, tools of machine learning and Bayesian model optimization experiments for launch in May with PyTorch 1.1.
Facebook's recommendation tool has been controversial in the past. Last year, Keras deep learning library creator FrançoisChollet in a post called, the "conscience of AI researchers should not work in Facebook", he advocates "do not use AI as a tool to manipulate the user; the contrary, as a tool to provide AI to the user, so that they have more autonomy. "
Source: VentureBeat