The document discusses unlocking unstructured data through artificial intelligence and machine learning. It outlines stages of AI from enhanced to bespoke and pre-trained models with transfer learning. It also discusses cognitive services, intelligent APIs, data for inference, and developer tools and frameworks. Finally, it outlines the machine learning process from preparing and registering data to training, testing, building, and deploying models and monitoring performance.
9. Cognitive Services
Intelligent APIs
Data for inference
Developer Tools
Frameworks
Extensive training data
Pre-Trained Models with
Transfer Learning
Some training data
Enhanced AIPre-Trained AI Bespoke AI
Stages of AI
10.
11.
12.
13.
14. Sophisticated pretrained models
To simplify solution development
Azure
Databricks
Machine Learning
VMs
Popular frameworks
To build advanced deep learning solutions TensorFlow KerasPytorch Onnx
Azure
Machine Learning
LanguageSpeech
…
Azure
Search
Vision
On-premises Cloud Edge
Productive services
To empower data science and development teams
Powerful infrastructure
To accelerate deep learning
Flexible deployment
To deploy and manage models on intelligent cloud and edge
Cognitive Services
45. Azure Text Analytics
Azure Language Understanding Service
Azure Machine Learning Workspace
Azure Machine Learning Visual Interface
Azure Machine Learning AutoML
Azure Machine Learning Notebooks
Azure Machine Learning Service API
Azure Machine Learning Deployment
Azure Machine Learning BERT Demo
Beyond Word Embeddings NLP Tutorial
Notes de l'éditeur
There are three main solutions you can build with Azure AI.
1. Build apps that have the ability to interact with users naturally, with pre-trained AI models.
2. Build your own machine learning and deep learning solutions
3. Unlock insights lying latent in various types of documents.
There are three main solutions you can build with Azure AI.
1. Build apps that have the ability to interact with users naturally, with pre-trained AI models.
2. Build your own machine learning and deep learning solutions
3. Unlock insights lying latent in various types of documents.
There are three main solutions you can build with Azure AI.
1. Build apps that have the ability to interact with users naturally, with pre-trained AI models.
2. Build your own machine learning and deep learning solutions
3. Unlock insights lying latent in various types of documents.
Microsoft Envision 2016
We’ll mainly be using the Azure Machine Learning service in this talk.
Experiments – this is where you try stuff / will have a history / track code through snapshots for every run / keep sanity (metadata)
Data Stores – where does your data live – MNIST is fine until it doesn’t fit on your hard drive, you also need to share with others, and you spilt your coffee once on your machine and lost everything
Compute – the metal where the experiments run (physical run on compute)
Models – output (just the good ones – you specify what to keep and how to version)
Images – package model with inference code and everything you need for it to run properly
Deployments – actual running models with ip addresses and everything
Compute – the actual machines where these things run (I only want these things to run when I have jobs!!)
Storage – literal azure storage account where we store: data use it like a scratch space or local drive (you can add more!! Using datastores from before – this is basically the default datastore)
Registry – where we store the images we talked about in the last slide (but also for running all of the experiments! – you can add your own – and run your experiments from there)
Key vault – storing secrets that we need to attach a compute storage, dsvm, we don’t want people to have your passwords – how would they want us to word this?
App Insights – this is for deployed models and their telemetry (you can also customize what you collect – but we won’t talk about that here)
TODO: Need to discuss issue of cost
TODO: Discuss that this is the baseline and does not include any compute resources that might be added (or external data sources)
Our approach to ML frameworks is simple.
We give customers the flexibility to choose their deep learning framework, without getting locked one framework.
To help with this we’ve created a community project in partnership with Facebook that allows customers to train in one framework and use another one for inference
Now, let me move to the ML services on Azure
DANIEL slide
Explain the flow step by step
Use following slides as a recap of the demo you just did.
TODO: maybe have YOLO output tensor diagram
TODO: Use a screenshot from the YOLO deployment, not MNIST.