TensorFlow 1.0 brings machine learning to mobile devices
There’s a new TensorFlow 1.0 release candidate and it comes bearing full-blown features. In addition to making it easier for Java and Python users to put TensorFlow to good use, the added support for iOS and Android has now been optimized. Let’s see what else is new.
TensorFlow 1.0 features and improvements
The new release candidate comes with XLA (Accelerated Linear Algebra), a domain-specific compiler for linear algebra that optimizes TensorFlow computations. According to the overview, XLA brings improvements in speed, memory usage, and portability on server and mobile platforms. Although most users will not see large benefits from XLA at first, they can experiment by using XLA via just-in-time (JIT) compilaton or ahead-of-time (AOT) compilation.
There’s also a new Java API but users should keep in mind that it is incomplete and experimental and can change without notice. Also, new Python 3 Docker images have been added and several Python API calls have been changed to resemble NumPy more closely.
Good news for those who focus on Android: the release candidate comes with new person detection + tracking demo implementing “Scalable Object Detection using Deep Neural Networks” (with additional YOLO object detector support), as well as new camera-based image stylization demo based on “A Learned Representation For Artistic Style”.
Judging by the release candidate update, it seems that TensorFlow 1.0 might be focusing greatly on boosting Python interactions. That’s no surprise considering that Python remains the most popular programming language for machine learning.
Check out the list of bug fixes and other changes here.