Google launches TensorFlow 2.0 Alpha
Featured
Google on Wednesday announced a series of announcements related to TensorFlow, the open-source machine learning library that's already been downloaded more than 41 million times. To make it even more accessible, Google is rolling out the alpha version of TensorFlow 2.0, which Google says is simpler and more intuitive to use.
On top of that, Google announced TensorFlow.js version 1.0 for the Javascript community. Google also introduced TF Federated, an add-on that lets users take advantage of decentralized data, as well as TF Privacy for fairer and safer training.
With TensorFlow 2.0, Google is making API components integrate better with tf.keras as the recommended high-level API for most users. This should help developers move more easily from data ingestion to transformation, model building, training and ultimately to deployment. Additionally, the launch of TensorFlow Datasets lets developers import many common datasets.
TensorFlow 2.0 will also feature eager execution by default -- this means ops will run immediately upon calling them. The new version also features automatic optimization of eager code with tf.function, intuitive Python control flows and improved error messaging.
Meawhile, Google is also introducing new add-ons, such as TensorFlow Federated. The open source framework allows for experimentation with machine learning and other computations on decentralized data -- where the data is generated. Typically, machine learning models need data that's centrally located, which can create challenges if the data is sensitive or expensive to centralize.
TensorFlow Privacy is an open source library that makes it easier for developers to train machine learning models using techniques based on the theory of differential privacy. It effectively offers a strong mathematical guarantee that machine learning models don't learn or remember details about specific users.
Google on Wednesday also announced TensorFlow.js version 1.0, which offers significant improvements for the Javascript community. For instance, MobileNet v1 is 9X faster in browser for inference compared to last year. There are also new off-the-shelf models for web developers to incorporate into applications, as well as support for more platforms where Javascript runs.