2 day Deep Learning Workshop at Karunya - Session 2
Download as PPTX, PDF0 likes154 views
The document discusses strategies for succeeding in an AI journey using Keras deep learning on cloud. It recommends starting with TensorFlow 2.0 and Keras in Google Colab for easy prototyping without setup. The strategy includes iterating on models, performing deep distributed training using cloud GPUs/TPUs, and serving models online/offline. Keras supports multiple backends like TensorFlow, CNTK, and MXNet to avoid lock-in to a single ecosystem.
2 day Deep Learning Workshop at Karunya - Session 2
- 1. Artificial Intelligence for everyone Strategy for Success How to succeed in AI journey? https://sites.google.com/view/AIforEveryone AI for Good Workshop March 2019 Session 2
- 2. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone Google A.I. Experiments: Envisioning to deploy Deep Learning? What is your strategy to lead the AI wave ? Image credits: Gartner
- 3. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone What is you AI strategy? Do you have a Strategy to jump start your AI journey? friendly framework ? instant setup? Do you have a Strategy to develop successful AI Products? deep distributed training? production grade inference? How to succeed in Applied AI ?
- 4. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone How to succeed in AI? Credits: Jeff Dean at AI Frontiers: Trends and Developments in Deep Learning Research, Google Brain
- 5. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone How to develop successful AI products ? A friendly library Cloud Compute Zero setup Free/ Pay per second Successful AI products Cloud platform TensorFlow 2.0 Colab
- 6. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone Need the fastest path to success in AI ? Experimenting with Deep Learning Prioritize developer experience: API designed for human beings. Follows best practices ! Rapidly prototype: An API focus on enabling fast experimentation Get State of art results in Deep Learning: Winners at Kaggle competitions often use this API Easily turn idea into products: Deploy on Google Cloud, AWS, Azure, iOS, Android, browser Raspberry Avoid lock you into one ecosystem: Keras development is backed by Google, AWS, Microsoft, Nvidia “Making deep learning accessible to everyone” http://keras.io Deep Learning with Python by François Chollet ISBN 9781617294433
- 7. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone AWS Keras on Cloud Start your Deep Learning projects Easy prototyping Zero Setup Access on web browser Turn ideas to production Deploy on cloud on Multi-GPU Avoid lock-in to one ecosystem tensorflow mxnet CNTK A friendly Deep Learning library Cloud platform Deep Learning for everyone! Easily turn idea into products Need a Scaleable AI Product Strategy ? Keras on Cloud
- 8. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone Avoid lock-in into one ecosystem Avoid lock-in into one ecosystem Keras supports multiple backend engines The TensorFlow backend (from Google) The CNTK backend (from Microsoft) The Theano backend MXNet (from Amazon)
- 9. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone Google A.I. Experiments: Summary: Success Strategy for AI journey Keras Deep Learning on Cloud for success in AI 2. Iterate models 3. Deep Distributed Training 4. Serving 1. Jump start Add layers / pretrained models API for humans Optimized defaults Use Cloud Scale TPU/GPU Use saved models to Predict online/offline Tensorflow, CNTK, mxnet
- 10. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone Want to explore if Keras is a good choice? • TensorFlow Dev Summit 2019 Keynote • https://youtu.be/b5Rs1ToD9aI?t=450
- 11. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone Summary : Jump start • Updated March 2019 • Strategy to Jump Start Learn Keras with TensorFlow 2.0 Learn how to use tf.Keras Use tf.Data with TensorFlow 2.0 Learn how to use Dataset API Use Google Co-Lab Just need web browser Free hardware If your goal is research especially on advanced models that require dynamic graphs (such as varying text or video with unpredictable length), it is better to explore pytorch with fast.ai. If your goal is to deploy immediately as products especially different target devices such smartphones today, TensorFlow Lite offers great path.
- 12. Acknowledgments & Credits are mentioned to inspirational resources presented in the end of Slide. For more like this, https://sites.google.com/view/AIforEveryone Other good Frameworks to learn www.Fast.ai
- 13. Acknowledgments 1) KERAS.io François Chollet’s Book on “Deep Learning with Python” 2) TensorFlow Dev Summit 3) Gartner 4) www.fast.ai https://www.tensorflow.org/dev-summit www.gartner.com