Abstract: In this talk, I highlight 10 powerful, yet unknown, aspects of TensorFlow based on years of profiling, tuning, and debugging CPU/GPU/TPU-based TensorFlow in production.
Key takeaways include techniques for profiling and tuning TensorFlow Core, TensorFlow Serving, and TensorFlow Lite.
In addition to model training optimizations such as batch normalization and XLA, I demonstrate post-training optimizations including 8-bit quantization and layer-fusing.
Data Analysts, Software Engineers, Research Engineers, Data Scientists, Application Engineers, DevOps
""If you can't observe, you can't improve."" Attendees will learn how to profile, monitor, and improve TensorFlow models through the entire AI pipeline including both model training and model serving.
The audience should be familiar with the basic concepts of neural networks and TensorFlow.
Bio: Chris Fregly is Founder and Applied AI Engineer at PipelineAI, a Real-Time Machine Learning and Artificial Intelligence Startup based in San Francisco. He is also an Apache Spark Contributor, a Netflix Open Source Committer, founder of the Global Advanced Spark and TensorFlow Meetup, author of the O’Reilly Training and Video Series titled, "High Performance TensorFlow in Production with Kubernetes and GPUs." Previously, Chris was a Distributed Systems Engineer at Netflix, a Data Solutions Engineer at Databricks, and a Founding Member and Principal Engineer at the IBM Spark Technology Center in San Francisco.