End-2-End Machine Learning on GCP
summary
Builds a classifier locally, uses transfer learning to replace the top layer and retrains in ~10 minutes, exports an export saved_model and deploys it to Google Cloud ML engine with a prediction API, and demonstrates on-device inference with TensorFlow Lite. Shows scaling options: estimators → tf.keras → tpu estimator, and claims Cloud TPU — ResNet-50: 8 minutes to 93% accuracy as an example of speed. Mentions TFX, kubernetes, AutoML/Neural Architecture Search and continuous retraining pipelines.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:10What is machine learning
- 8:23TensorFlow foundation
- 14:41Building models step-by-step
- 22:23Optimizing models and datasets
- 25:09Serving models in production
- 31:03ML infrastructure on GCP
- 34:54Future of ML with TPUs and AutoML
- 36:58Q&A
related talks
36mAccelerating Machine Learning at Scale with
13mTaking Machine Learning to the Next Level
43mMachine Learning at Scale
44mBuilding smarter solutions with no expertise in machine learning
42mCreating Smarter Applications
32mGPU Machine Learning From Laptop to Cloud
04mState-of-the-art Machine Learning research with Google tools
36mAccelerating High-Performance Machine Learning at Scale i...