Machine Learning Zero to Hero
summary
Code shows training a rock paper scissors classifier with TensorFlow Keras using 150x150x3 input images, a convolution + pooling stack feeding a 512 dense units layer, trained for ~15 epochs and reaching about 96.8% accuracy on validation. Workflow includes loss + optimizer via model.compile, realtime inspection with TensorBoard, scaling with Mirrored and MultiWorker strategies, exporting as SavedModel for TensorFlow Serving, and converting to TensorFlow Lite for on‑device inference.
AI-written summary · about
chapters (13)
- 0:00Start
- 0:50Imagine a Rock Paper Scissors Game...
- 11:41Convolutions
- 12:50Convolution Example
- 14:23Pooling Example
- 23:43Structured data parsing
- 25:21Build your own layers
- 28:04Built-in performance profiling
- 28:34Training can take a long time
- 29:36Distribution through data parallelism Update model
- 30:09Performance on Multi-GPU
- 32:23To Saved Model and beyond
- 33:48Big models fit in small packages
related talks
01h 06mDeep Learning with TensorFlow
51mA Practical Introduction to Convnets for Image Classification
48mLeveraging the Power of C++ for Efficient Machine Learning on Embedded Devices
39mLeveraging the power of C++ for efficient machine learning on embedded devices
01h 01mGetting Started with Computer Vision
27mMachine Learning on the edge using TensorFlow Lite
37mJavaScript Machine Learning from Magic to Science
42mCreating Smarter Applications