[第23回 Machine Learning 15minutes!] 15分で説明する「Microsoft AI Platform」
•
1 gefällt mir•658 views
[第23回 Machine Learning 15minutes!] 15分で説明する「Microsoft AI Platform」 https://satonaoki.wordpress.com/2018/04/28/ml15min-microsoft-ai/ 第23回 Machine Learning 15minutes! (2018/04/28) https://machine-learning15minutes.connpass.com/event/81760/
Empfohlen
Empfohlen
Weitere ähnliche Inhalte
Mehr von Naoki (Neo) SATO
Mehr von Naoki (Neo) SATO (20)
Kürzlich hochgeladen
Kürzlich hochgeladen (20)
[第23回 Machine Learning 15minutes!] 15分で説明する「Microsoft AI Platform」
- 2. 1950 1960 1970 1980 1990 2000 2010
- 5. ImageNet classification challenge Human parity 5.1% 28.2% 25.8% 16.4% 11.7% 7.3% 6.7% shallow 8 layers 19 layers 22 layers 2010 2011 2012 AlexNet 2013 2014 VGG 2014 GoogleNet 2015 3.57% Microsoft ResNet 152 layers
- 6. The revolution of ResNet 11x11 conv, 96, /4, pool/2 5x5 conv, 256, pool/2 3x3 conv, 384 3x3 conv, 384 3x3 conv, 256, pool/2 fc, 4096 fc, 4096 fc, 1000 AlexNet, 8 layers 3x3 conv, 64 3x3 conv, 64, pool/2 3x3 conv, 128 3x3 conv, 128, pool/2 3x3 conv, 256 3x3 conv, 256 3x3 conv, 256 3x3 conv, 256, pool/2 3x3 conv, 512 3x3 conv, 512 3x3 conv, 512 3x3 conv, 512, pool/2 3x3 conv, 512 3x3 conv, 512 3x3 conv, 512 3x3 conv, 512, pool/2 fc, 4096 fc, 4096 fc, 1000 VGG, 19 layers input Conv 7x7+ 2(S) MaxPool 3x3+ 2(S) LocalRespNorm Conv 1x1+ 1(V) Conv 3x3+ 1(S) LocalRespNorm MaxPool 3x3+ 2(S) Conv Conv Conv Conv 1x1+ 1(S) 3x3+ 1(S) 5x5+ 1(S) 1x1+ 1(S) Conv Conv MaxPool 1x1+ 1(S) 1x1+ 1(S) 3x3+ 1(S) Dept hConcat Conv Conv Conv Conv 1x1+ 1(S) 3x3+ 1(S) 5x5+ 1(S) 1x1+ 1(S) Conv Conv MaxPool 1x1+ 1(S) 1x1+ 1(S) 3x3+ 1(S) Dept hConcat MaxPool 3x3+ 2(S) Conv Conv Conv Conv 1x1+ 1(S) 3x3+ 1(S) 5x5+ 1(S) 1x1+ 1(S) Conv Conv MaxPool 1x1+ 1(S) 1x1+ 1(S) 3x3+ 1(S) Dept hConcat Conv Conv Conv Conv 1x1+ 1(S) 3x3+ 1(S) 5x5+ 1(S) 1x1+ 1(S) Conv Conv MaxPool 1x1+ 1(S) 1x1+ 1(S) 3x3+ 1(S) AveragePool 5x5+ 3(V) Dept hConcat Conv Conv Conv Conv 1x1+ 1(S) 3x3+ 1(S) 5x5+ 1(S) 1x1+ 1(S) Conv Conv MaxPool 1x1+ 1(S) 1x1+ 1(S) 3x3+ 1(S) Dept hConcat Conv Conv Conv Conv 1x1+ 1(S) 3x3+ 1(S) 5x5+ 1(S) 1x1+ 1(S) Conv Conv MaxPool 1x1+ 1(S) 1x1+ 1(S) 3x3+ 1(S) Dept hConcat Conv Conv Conv Conv 1x1+ 1(S) 3x3+ 1(S) 5x5+ 1(S) 1x1+ 1(S) Conv Conv MaxPool 1x1+ 1(S) 1x1+ 1(S) 3x3+ 1(S) AveragePool 5x5+ 3(V) Dept hConcat MaxPool 3x3+ 2(S) Conv Conv Conv Conv 1x1+ 1(S) 3x3+ 1(S) 5x5+ 1(S) 1x1+ 1(S) Conv Conv MaxPool 1x1+ 1(S) 1x1+ 1(S) 3x3+ 1(S) Dept hConcat Conv Conv Conv Conv 1x1+ 1(S) 3x3+ 1(S) 5x5+ 1(S) 1x1+ 1(S) Conv Conv MaxPool 1x1+ 1(S) 1x1+ 1(S) 3x3+ 1(S) Dept hConcat AveragePool 7x7+ 1(V) FC Conv 1x1+ 1(S) FC FC Soft maxAct ivat ion soft max0 Conv 1x1+ 1(S) FC FC Soft maxAct ivat ion soft max1 Soft maxAct ivat ion soft max2 GoogleNet, 22 layers
- 7. The revolution of ResNet 1x1 conv, 64 3x3 conv, 64 1x1 conv, 256 1x1 conv, 64 3x3 conv, 64 1x1 conv, 256 1x1 conv, 64 3x3 conv, 64 1x1 conv, 256 1x2 conv, 128, /2 3x3 conv, 128 1x1 conv, 512 1x1 conv, 128 3x3 conv, 128 1x1 conv, 512 1x1 conv, 128 3x3 conv, 128 1x1 conv, 512 1x1 conv, 128 3x3 conv, 128 1x1 conv, 512 1x1 conv, 128 3x3 conv, 128 1x1 conv, 512 1x1 conv, 128 3x3 conv, 128 1x1 conv, 512 1x1 conv, 128 7x7 conv, 64, /2, pool/2 Yellow lady's slipper Komondor Coucal
- 8. Language to image synthesis ”A bird with wings that are blue and a red belly” “this bird is red with white and has a very short beak” “A herd of sheep grazing on a lush green field”
- 9. Generative Adversarial Networks G Generator Real samples D Discriminator Real Fake Error
- 10. Attentional GAN “this bird has a green crown black primaries and a white belly” bird this has belly white black green white this bird
- 11. Speech recognition human parity Human parity 5.9%
- 12. Speech recognition human parity ResNet VGG B-LSTM Combinator at word level “the cat sat” Word hypotheses Posterior probabilities … Example 1 Example 2 Example 3 Example 4
- 13. Machine reading human parity 1. Microsoft – MSR 82.650% 2. HIT and iFLYTEK Research 82.482% 3. Alibaba iDST NLP 82.440% 4. Microsoft – MSR 82.136% 5. Tencent DPDAC NLP 81.790% … 11. Microsoft – MSR 79.901% 13. Microsoft – Business AI 79.608% 14. Alibaba iDST NLP 79.199% 14. HIT and iFLYTEK Research 79.083% 15. Microsoft – Business AI 78.978% … Human Parity Exact Match % SQuAD (Stanford Question Answering Dataset) 500+ articles 100,000+ question-answers pairs
- 14. Reward Action Ms. Pacman high score Agent Environment Reinforcement learning
- 15. Form 10-K 2016
- 16. Azure AI Services Azure Infrastructure Tools
- 25. Local machine Scale up to DSVM Scale out with Spark on HDInsight Azure Batch AI (Coming Soon) ML Server Azure Machine Learning - Experimentation A ZURE ML EXPERIMENTATION Command line tools IDEs Notebooks in Workbench VS Code Tools for AI
- 26. DOCKER Single node deployment (cloud/on-prem) Azure Container Service Azure IoT Edge Microsoft ML Server Spark clusters SQL Server Azure Machine Learning – Model Management A ZURE ML MODEL MANAGEMENT
- 27. Azure Machine Learning Workbench Windows and Mac based companion for AI development Full environment set up (Python, Jupyter, etc) Embedded notebooks Run History and Comparison experience New data wrangling tools
- 28. Visual Studio Tools for AI Visual Studio extension with deep integration to Azure ML End to end development environment, from new project through training Support for remote training Job management On top of all of the goodness of Visual Studio (Python, Jupyter, Git, etc)
- 30. Azure Machine Learning Studio Platform for emerging data scientists to graphically build and deploy experiments • Rapid experiment composition • > 100 easily configured modules for data prep, training, evaluation • Extensibility through R & Python • Serverless training and deployment Some numbers: • 100’s of thousands of deployed models serving billions of requests
- 31. Machine Learning & AI Portfolio When to use what? What engine(s) do you want to use? Deployment target Which experience do you want? Build your own or consume pre- trained models? Microsoft ML & AI products Build your own Azure Machine Learning Code first (On-prem) ML Server On- prem Hadoop SQL Server (cloud) AML services (Preview) SQL Server Spark Hadoop Azure Batch DSVM Azure Container Service Visual tooling (cloud) AML Studio Consume Cognitive services, bots