Swift for TensorFlow: machine learning with no boundaries.

on YouTube ↗24m

" summary

Swift for TensorFlow claims automatic differentiation built into the language, C++ interop with no wrappers, and cross‑platform deployment to Linux, macOS, Windows, Android and iOS. Engineering work shows parity with C++ on AlphaGo Zero re‑implementations with far fewer lines of code, Jeremy Howard ports fastai to Swift for performance, and TPU runs cut training from days or weeks to four hours or minutes. Roadmap items include mobile/embedded support, concurrency and MLIR integration for on-device training.

AI-written summary · about

" chapters (9)

1. 0:00Intro
2. 6:54Accuracy Improvements
3. 9:33Swift for TensorFlow
4. 12:02Crossplatform
5. 14:48Performance
6. 16:31Collaborations
7. 18:03Use Cases
8. 21:08Developer Tools
9. 22:49Outro

" related talks

• 39mSwift for TensorFlow: Machine Learning with No BoundariesPaige Bailey · Scale By The Bay 2019
• 28mSwift for TensorFlowJames Bradbury, Richard Wei · Google I/O 2019
• 40mIntroduction to Swift for TensorflowBrett Koonce · AltConf 2019
• 47mSwifty TensorFlowSiamak Ashrafi · 360|iDev 2019
• 26mMachine Learning with TensorFlowRajat Monga · MLconf SF 2016
• 27mMachine Learning and MobileBrett Koonce · Scale By The Bay 2019
• 32mAI for Mobile and IoT Devices: TensorFlow LiteSarah Sirajuddin, Tim Davis · Google I/O 2019
• 31mTensorflow for JanitorsDániel Molnár · code.talks 2017