GTC Silicon Valley-2019 ID:S9497:Distributed TensorFlow with Distribution Strategies
Magnus Hyttsten(Google, Inc.)
We'll discuss best practices for using TensorFlow effectively in a distributed manner. Machine learning models are only getting bigger and require more compute to train. We will explain how to use TensorFlow's new distribution strategy to get easy high-performance training with Keras models (and custom models) on multi-GPU setups as well as multi-node training on clusters with accelerators.