Slides used during the webinar session on Machine Learning.Net and Windows Machine Learning on 2019 02 21 for the GLUGnet User Group for .NET, Web, Mobile, Database
Modular Monolith - a Practical Alternative to Microservices @ Devoxx UK 2024
Â
Webinar GLUGNet - Machine Learning.Net and Windows Machine Learning
1. Getting Started with
Machine Learning .Net and
Windows Machine Learning
[ ML.Net & WinML ]
Bruno Capuano
Innovation Lead @Avanade
@elbruno | http://elbruno.com
2. why should I care about AI and ML?
As a developer,
3. Some problems are difficult to solve using traditional algorithms and
procedural programming.
4. IBM slaps patent on coffee-delivering drones that can read
your MIND (link)
5. IBM slaps patent on coffee-delivering drones that can read
your MIND (link)
7. âIt has exquisite buttons âŚ
with long sleeves âŚworks for
casual as well as business
settingsâ{f(x) {f(x)
Machine Learning: âProgramming the Unprogrammableâ
9. Is this A or B? How much? How many? How is this organized?
Regression ClusteringClassification
Machine Learning Tasks
10. Get started with Machine Learning
Prepare Data Build & Train Evaluate
Azure Databricks Azure Machine Learning
Quickly launch and scale Spark on demand
Rich interactive workspace and notebooks
Seamless integration with all Azure data
services
Broad frameworks and tools support:
TensorFlow, Cognitive Toolkit, Caffe2, Keras,
MxNET, PyTorch
In the cloud â on the edge
Docker containers
Windows Machine Learning
13. Azure Machine Learning Services
gives you an end-to-end
solution to prepare data and
train your model in the Cloud.
WinMLTools converts existing
models from CoreML, scikit-
learn, LIBSVM, and XGBoost
Azure Custom Vision makes it
easy to create your own image
models - https://customvision.ai/
Azure AI Gallery curates models
for use with Windows ML -
https://gallery.azure.ai/models
How do I get ONNX models to use in my
application?
14. 1. Developers can focus on their data and
their scenarios, using Windows ML for
model evaluation
2. Enables using ML models trained with a
diverse set of toolkits
3. Hardware acceleration gets fast evaluation
results across the diversity of the entire
Windows device ecosystem.
Windows ML solves three problems for you
Direct3D
GPU
CPU
DirectML
Model Inference Engine
WinML Win32 API
WinML UWP API
Win32 App
WinML Runtime
UWP App
17. Easy / Less Control Full Control / Harder
Vision Speech Language
Knowledge SearchLabs
TextAnalyticsAPI client = new TextAnalyticsAPI();
client.AzureRegion = AzureRegions.Westus;
client.SubscriptionKey = "1bf33391DeadFish";
client.Sentiment(
new MultiLanguageBatchInput(
new List<MultiLanguageInput>()
{
new MultiLanguageInput("en","0",
"This vacuum cleaner sucks so much dirt")
}));
e.g. Sentiment Analysis using Azure Cognitive Services
9% positive
Pre-built ML Models (Azure Cognitive Services)
18. ML.NET is for building custom models
Custom models
Easier / Less Control Harder / Full Control
Pre-built models
TensorFlow
ML.NETVisionSpeech LanguageKnowledge Search
19. Prepare Your Data Build & Train Run
Build your own custom machine learning models
22. Is this A or B? Kid or Baby
Based on the age:
Kid or Baby
Age classes explained
23. And more! Samples @ https://github.com/dotnet/machinelearning-samples
Customer segmentation
Recommendations
Predictive maintenance
Forecasting
Issue Classification
Image classification
Object detection
Sentiment Analysis
A few things you can do with ML.NET âŚ
24. Proven & Extensible Open Source
https://github.com/dotnet/machinelearning
Build your own
Supported on Windows, Linux, and macOS
Developer Focused
ML.NET 0.10.0 (Preview)
Machine Learning framework made for .NET developers
25. Windows 10 (Windows Defender)
Power Point (Design Ideas)
Excel (Chart Recommendations)
Bing Ads (Ad Predictions)
+ moreAzure Stream Analytics (Anomaly Detection)
ML.NET is Proven at scale, enterprise ready
26. ML.NET is a framework for building custom ML Models
31. Load Data Extract Features Train Model Evaluate Model
Model
consumption
labels + plain text labels + feature vectors model
End to End ML Workflow
32. Load Data Extract Features Train Model Evaluate Model
Model
consumption
labels + plain text labels + feature vectors
Enter...
in ML.NETLearningPipelines!
model
End to End ML Workflow
33. Load Data Extract Features Train Model Evaluate Model
Model
consumption
Machine Learning is Iterative
36. ML.Net, working with
TensorFlow frozen models
MakeMagicHappen();
https://www.microsoft.com/net/learn/apps/machi
ne-learning-and-ai
37. ⢠API improvements
⢠Additional ML Tasks and Scenarios
⢠Improved Deep Learning with TensorFlow
⢠Scale-out on Azure
⢠Better GUI to simplify ML tasks
⢠Improved tooling in Visual Studio
⢠Improvements for F#
⢠Language Innovation for .NET
Road Ahead for ML.NET