Diving into TensorFlow 2.0 Ecosystem 

TensorFlow has matured into an entire end-to-end machine learning ecosystem. With eager execution and API cleanup, writing code and training model in TensorFlow 2.0 is fun and easy. Use TFX to manage a production pipeline, and use tflite and tf.js to deploy on edge devices.

Talk Description 

TensorFlow has matured into an entire end-to-end machine learning ecosystem. With eager execution and API cleanup, writing code and training model in TensorFlow 2.0 is fun and easy. But deep learning is more than just training a single model. Use TFX to create and manage a production pipeline, so that engineers, data scientists and machine learning researchers can collaborate seamlessly without worry about unnecessary technique details. TensorFlow Lite, Tensorflow js and kubeflow make deployment simple in various scenarios. In this talk, we are going to cover following topics:

  • Tensorflow 2.0
  • Eager mode and API clean up
  • Different ways to build a model, from simple to arbitrarily flexible
  • Different ways to train a model, from simple to arbitrarily flexible
  • Distributed training with GPU and TPU
  • TFX for production pipeline
  • Core concept and pipeline component
  • Validation, transform, and analysis
  • Apache Beam, airflow, and kubeflow
  • TFLite and TF.js

Speaker 

TJWei is an associate professor of NCTU College of Artifical Intelligence. He is also a Google developers expert in machine learning and active in open source communities, founding Hualien Python User Group and Google Developers Group Hualien.