Cutting Edge TensorFlow: New Techniques
summary
Keras Tuner automates hyperparameter search and claims a tuned ResNet variant improves a logo-classifier from ~79% to 100% accuracy while shrinking 44M→24M params after giving the tuner a $500 budget. TensorFlow Probability adds distributional layers and Bayesian weights, processes 6 TB, explores ~250,000 models and reports a 50% drop in false positives and 200% drop in false negatives on jet‑engine anomaly detection. TF‑Ranking trains on ~1B query‑document pairs, gains >3% by adding sparse text and ensembling with lambdaMART. TensorFlow Graphics ships differentiable 3D transforms, cameras, materials, graph convolutions and is pip installable.
AI-written summary · about
chapters (13)
- 0:00Introduction to projects
- 0:35Keras Tuner overview
- 3:33Keras Tuner brand detection
- 6:25Tuner cloud services
- 7:20Probabilistic programming
- 13:30TensorFlow Probability tools
- 17:20Introduction to ranking
- 20:04TF Ranking loss functions
- 22:43TF Ranking implementation
- 25:58TF Ranking resources
- 26:44TF Graphics overview
- 30:14Differentiable graphics
- 34:053D convolution and closing
