Modern Keras Design Patterns

on YouTube ↗14m

" summary

A practical walkthrough of building a variational autoencoder in Keras: adds a sampling layer that draws z from z_mean and z_log_var, uses the Functional API to produce two outputs, registers the KL divergence via model.add_loss, and shows using a GradientTape custom train step when subclassing. It also notes that model.fit still works and the approach scales to MirroredStrategy or TPUStrategy, with full code available in the TensorFlow/Keras links.

AI-written summary · about

" chapters (8)

1. 0:00Keras design patterns
2. 3:06Autoencoder basics
3. 4:32Variational autoencoders need
4. 6:19Custom layers Functional API
5. 7:47Building variational autoencoders
6. 9:34Complete VAE
7. 11:19Subclassing API training
8. 12:22Distributed training

" related talks

• 01h 17mDeep generative models for image...Dimitra Gkorou, Koen Vannisselroij, Shama Khalil, Sonali Fotedar · PyData Eindhoven 2019
• 57mTutorial on Deep Learning IIIRuslan Salakhutdinov · Foundations of Machine Learning Boot Camp 2017
• 21mWhy TensorFlow eager execution mattersSigrid Keydana · rstudio::conf 2019
• 17mProcedural Modeling Using Autoencoder NetworksRadomír Mech, Paul Asente, Levent Burak Kara, Mehmet Ersin Yumer · UIST 2015
• 01h 22mDeep Probabilistic Methods with PyTorchChris Ormandy · PyData London 2018
• 01h 37mTen Steps to KerasValerio Maggio · PyData London 2017
• 09mPrimal--Dual Optimization and Application to SamplingAdil Salim · Meet the Fellows Welcome Event
• 01h 00mTutorial on Deep Learning IVRuslan Salakhutdinov · Foundations of Machine Learning Boot Camp 2017