Learnings from TensorFlow Dev Summit 2019 - GDG Cambridge

At the TensorFlow Dev Summit Viewing Party organized by GDG Cambridge, I gave a talk with the key takeaways from the TensorFlow Dev Summit 2019.

Transcript

1. Learnings from TensorFlow Dev Summit 2019 TensorFlow Dev Summit Viewing Party, GDG Cambridge Margaret Maynard-Reid, 3/18/2019 @margaretmz
2. @margaretmz | #MachineLearning #GDE | #TFDevSummit TensorFlow Dev Summit 2019 pre-summit dinner with ML GDEs Margaret, Kshitiz, Josh Margaret, François check-in Margaret, Vikram, Wolff, Tim ML GDEs group photo Paige & Margaret Margaret, Hoi, Attila, Sandeep Margaret, Sergii, Laurence 2
3. @margaretmz | #MachineLearning #GDE | #TFDevSummit TensorFlow 2.0 Alpha pip install -U --pre tensorflow Great for both beginners & experts; for both researchers & developers: • tf.Keras as the high level API and eager execution as default • declutters the duplicated APIs for consistency • still offers flexibility for those who need low level APIs Summit video | Documentation | My notes on TensorFlow 2.0 3
4. @margaretmz | #MachineLearning #GDE | #TFDevSummit TensorFlow Roadmap 4
5. @margaretmz | #MachineLearning #GDE | #TFDevSummit Learn TensorFlow 2.0 • Redesigned tensorflow.org • Udacity Intro to TensorFlow for Deep Learning (link) • Coursera Intro to TensorFlow for AI, ML & Deep Learning (link) • Fast.ai embracing Swift for Deep Learning (link) • MIT Deep Learning (link) 5
6. @margaretmz | #MachineLearning #GDE | #TFDevSummit #PoweredbyTF 2.0 • TensorFlow World by O’Reilly, 4-day conference, CFP open till 4/23 • Powered by TensorFlow 2.0 Challenge, hosted on DevPost • Google Summer of Code, students get to work on TensorFlow 6
7. @margaretmz | #MachineLearning #GDE | #TFDevSummit TensorFlow Lite talks There were two sessions on TensorFlow Lite: • Day 1 - TensorFlow Lite Overview ◦ Demo of Corel edge TPU ◦ Sparkfun microcontroller demo by Pete Warden • Day 2 - TensorFlow On-Device: Compressing, Quantizing, and Distributing ◦ Talk 1 ◦ Talk 2 ◦ Q&A with TensorFlow Lite team 7
8. @margaretmz | #MachineLearning #GDE | #TFDevSummit TensorFlow Lite Roadmap Focus on these areas: • Usability — get it working end to end. • Performance — fast execution: edge TPU delegate, GPU delegate. • Optimization — smaller/faster models with quantization & pruning et. • Documentation — better docs, tutorials and examples. 8
9. @margaretmz | #MachineLearning #GDE | #TFDevSummit TensorFlow Lite Roadmap https://www.tensorflow.org/lite/guide/roadmap 9
10. @margaretmz | #MachineLearning #GDE | #TFDevSummit Swift for TensorFlow Next generation ML framework!!! • Deploying dynamic models • Application integrated neural nets • Flexible and extensible autodiff • Improved developer workflows Seamless python interoperability & easier transition to Swift for TensorFlow… link to video | link to repo 10
11. @margaretmz | #MachineLearning #GDE | #TFDevSummit TensorFlow Extended End to end ML platform: Now integrates with AirFlow and Kubeflow (Link to videos: TFX overview & Pre-training Workflow | Post-training Workflow) Data ingestion Data validation Transform Estimator Keras Model Analysis Model Analysis Logging 11
12. @margaretmz | #MachineLearning #GDE | #TFDevSummit TensorFlow.js 1.0 A library for training and deploying ML models in the browser • Pretrained models - Image, Audio and Text classifications (NPM or hosted scripts) • Convert existing python models ◦ with a command line tool ◦ Saved Model, TFHub, Keras • Train in the browser and Node.js - with Layers API • Deploy on Platform link to video 12
13. @margaretmz | #MachineLearning #GDE | #TFDevSummit TensorFlow Community & SIGs Special interest groups: • Addon - tf.contrib is going away and moved into Core or Addon • Build - build, test & packaging • I/O - connecting TensorFlow to other systems • Networking - supporting more networking • Rust - a language binding for Rust • TensorBoard - inclusive roadmap link to video | tensorflow.org/community 13
14. @margaretmz | #MachineLearning #GDE | #TFDevSummit Other interesting talks • TensorBoard, now works in Colab! (link to video) • TensorFlow hub - reusable ML and easier transfer learning (link to video) • TF agents - reinforcement learning (link to video) Research & Future talks • Federated Learning - training on decentralized data (link to video | link to doc) • Mesh TensorFlow  - for large-scaled models (link to video) • Sonnet (by DeepMind)  - for complex models by researchers (link to video) 14
15. Read my blog post on Medium - bit.ly/learnings-TFDevSummit-2019 15
16. @margaretmz | #MachineLearning #GDE | #TFDevSummit Thank you & keep in touch! Follow me on Twitter, Medium or GitHub to learn more about Deep learning, TensorFlow and Android @margaretmz @margaretmz margaretmz 16