Making Deep Learning Easier - Updates on Keras

on YouTube ↗19m

" summary

Keras shows concrete, usable changes: model.fit now trains the same model code across JAX TensorFlow and PyTorch backends, with multi‑device sharding defined by a device mesh and explicit layout maps; a clear workflow combines LoRA low‑rank adapters and quantize before fine‑tuning so only a few matrices are updated and serialized; a Keras Hub preset system delivers interoperable checkpoints with Hugging Face and Kaggle, and distributed embedding APIs target TPU sparse cores for large recommender lookups.

AI-written summary · about

" chapters (10)

1. 0:00Keras introduction
2. 1:04Keras ecosystem overview
3. 3:13Understanding model.fit and progressive complexity
4. 5:20Backend flexibility
5. 7:06Keras Hub introduction
6. 9:39Model presets
7. 11:46LoRA fine-tuning
8. 14:05Keras Hub integrations
9. 14:25Keras Recommenders preview
10. 18:43Summary

" related talks

• 20mKeras 3: Deep Learning made easyTomasz Kalinowski · posit::conf(2024)
• 43m: Shortcut to AI masteryDivyashree Sreepathihalli · Women in ML Symposium 2023
• 01h 14mKeras (3) for the Curious and CreativeNgesa Marvin · PyData Global 2023
• 31mMaking machine learning accessible using KerasPavithra Vijay · WeAreDevelopers AI Congress Vienna 2018
• 33mWhat's new in Machine LearningKemal El Moujahid, Sarah Sirajuddin, Craig Wiley · Google I/O 2021
• 43mDeep learning beyond the learningJörg Schad · Codemotion Rome 2018
• 07mMachine learning tools from GoogleNadav Eiron · Google's Machine Learning Virtual Community Day 2021
• 24mKeynote: Next-generation frameworks for Large-scale AIAnima Anandkumar · Scale By the Bay 2020