Keras 3: Deep Learning made easy
summary
Keras 3 is a ground up rewrite that restores multi-backend support for JAX PyTorch and TensorFlow plus NumPy inference and a backend agnostic .keras save format so a model trained in JAX can load under TensorFlow for deployment via Connect and TensorFlow Serving. It adds an ops family of ~200 backend agnostic tensor functions, over 100 built in layers, a tabular API, RNG and distributed training with model parallelism and quantization. Simple MNIST examples go from 92% to 99% accuracy as layers are added.
AI-written summary · about
chapters (5)
related talks
01h 14mKeras (3) for the Curious and Creative
43m: Shortcut to AI mastery
19mMaking Deep Learning Easier - Updates on Keras
31mMaking machine learning accessible using Keras
01h 37mTen Steps to Keras
42mDeep learning in Python: how to choose a framework?
32mGetting Started with TensorFlow 2.0
12mWhat’s new in PyTorch 1.3