Discover, manage, deploy, monitor – rinse and repeat. In this session we show how Azure Machine Learning can be used to create the right AI model for your challenge and then easily customize it using your development tools while relying on Azure ML to optimize them to run in hardware accelerated environments for the cloud and the edge using FPGAs and Neural Network accelerators. We then show you how to deploy the model to highly scalable web services and nimble edge applications that Azure can manage and monitor for you. Finally, we illustrate how you can leverage the model telemetry to retrain and improve your content.
Vector Databases 101 - An introduction to the world of Vector Databases
AI for an intelligent cloud and intelligent edge: Discover, deploy, and manage with Azure ML services
1. AI for Intelligent Cloud and
Intelligent Edge:
Discover, deploy, and manage
with Azure ML Services
James Serra
Microsoft
Technical Architect, Data & AI
Blog: JamesSerra.com
2. About Me
Microsoft, Big Data Evangelist
In IT for 30 years, worked on many BI and DW projects
Worked as desktop/web/database developer, DBA, BI and DW architect and developer, MDM
architect, PDW/APS developer
Been perm employee, contractor, consultant, business owner
Presenter at PASS Business Analytics Conference, PASS Summit, Enterprise Data World conference
Certifications: MCSE: Data Platform, Business Intelligence; MS: Architecting Microsoft Azure
Solutions, Design and Implement Big Data Analytics Solutions, Design and Implement Cloud Data
Platform Solutions
Blog at JamesSerra.com
Former SQL Server MVP
Author of book “Reporting with Microsoft SQL Server 2012”
3. I tried to understand AI products on my own…
And felt like I was body slammed by Randy
Savage:
Let’s prevent that from happening…
4. Intro to Azure Machine Learning
Model Management
Hardware Acceleration
Edge Integration
AICloud+Edge
5. Machine learning is a data science technique that allows computers to
use existing data to forecast future behaviors, outcomes, and trends.
6. Prepare Data Build & Train Deploy
Custom AI
Building your own AI models for Transforming Data into Intelligence
12. Cognitive Services capabilities
Infuse your apps, websites, and bots with human-like intelligence
https://azure.microsoft.com/en-us/services/cognitive-services
14. Data Science
Lifecycle
TEAM DATA SCIENCE PROCESS (TDSP) IS MICROSOFT’S AGILE,
ITERATIVE METHODOLOGY TO DELIVER PREDICTIVE ANALYTICS
SOLUTIONS AND INTELLIGENT APPLICATIONS EFFICIENTLY
https://docs.microsoft.com/en-us/azure/machine-learning/team-data-science-process/overview
15. WHAT IS AZURE MACHINE LEARNING SERVICE?
Set of Azure Cloud
Services
Python
SDK
Prepare Data
Build Models
Train Models
Manage Models
Track Experiments
Deploy Models
That enables
you to:
16. What is Azure Machine Learning service?
Start training on your local machine and then scale out to the cloud
17. Azure Machine Learning Services
• Deprecate Azure Machine Learning Workbench
• Unified SDK, CLI and UX for training and deploying models
• Full Integration with Visual Studio Code and Azure DevOps
• Improved support for multiple compute targets
• Four new models for FPGA
• Vision AI Dev Kit available to order on Oct 1
21. Understanding the Edge: Heavy Edge vs Light Edge
Cloud: Azure Heavy Edge Light Edge
Descriptio
n
An Azure host that
spans from CPU to
GPU and FPGA VMs
A server with slots to insert CPUs, GPUs, and FPGAs or a X64 or ARM system that
needs to be plugged in to work
A Sensor with a SOC (ARM CPU, NNA, MCU) and memory
that can operate on batteries
Example
DSVM / ACI / AKS /
Batch AI
- DataBox Edge
- HPE
- Azure Stack
- DataBox Edge - Industrial PC
-Video Gateway
-DVR
-Mobile Phones
-VAIDK
-Mobile Phones
-IP Cameras
-Azure Sphere
- Appliances
What runs
model
CPU,GPU or FPGA
CPU,GPU or
FPGA
CPU, GPU x64 CPU Multi-ARM CPU
Hw accelerated
NNA
CPU/GPU MCU
Why Edge? latency, less data sent, filter, aggregate, work offline
25. 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
26. Comparable Table
Azure Machine Learning Studio Machine Learning Services
Pros • Rapid development (Drag and Drop)
• Works well with relatively simple
datasets
• Pre-built ML algorithms
• Cheap
• Fast (VMs with GPUs)
• Different optimization methods,
CI/CD pipeline
• Full control during training
• Manage computing resources
(choose VM size)
• Use open source ML libraries
Cons • Can be slow
• Limited optimization methods,
operationalized architecture
• Less control during training
• Fixed computing resources
• More elaborate to build, require
deeper knowledge of machine
learning
• Deeper models need much more
data with much more memory
• Higher costs for VM with GPU
31. Breakthroughs in deep
learning demand real-time AI
Convolutional Neural Networks (CNN)
ht-1 ht ht+1
xt-1 xt xt+1
ht-1 ht ht+1
yt-1 yt yt+1
Recurrent Neural Networks (RNN)
Deep neural networks (DNN) have enabled major
advances in machine learning and AI
Computer vision
Language translation
Speech recognition
Question answering
And more…
Problem
DNNs are challenging to serve and deploy
in large-scale online services
Heavily constrained by latency, cost, and power
Size and complexity outpacing growth of commodity CPUs
33. hardware architecture designed to accelerate real-time AI
calculations
Project Brainwave unique
advantage (DNN models and
FPGA)
No batching required
Brainwave delivers the ideal combination:
High hardware utilization
Low latency
Low batch sizes
In short, it is a hardware architecture and learning platform
designed to accelerate real-time AI calculations
Batch Size
Performance
Brainwave
NPU
1256
34. Credit: Henk Monster. Licensed under the Creative Commons Attribution 3.0 Unported license.
36. Project Brainwave use cases
For accelerated real-time image processing:
- Identify manufacturing defects
- Detect spills or open freezer doors in a retail store
- Conduct real-time medical image analysis
- Tract endangered species
- Detect cars parked in a fire lane
37.
38. The power of deep learning on FPGA
Performance Flexibility Scale
Rapidly adapt to evolving ML
Inference-optimized numerical precision
Exploit sparsity, deep compression
Excellent inference at low batch sizes
Ultra-low latency | 10x < CPU/GPU
World’s largest cloud investment in FPGAs
Multiple Exa-Ops of aggregate AI capacity
Runs on Microsoft’s scale infrastructure
Low cost
$0.21/million images on Azure FPGA (inferencing)
39. Project BrainWave
A Scalable FPGA-Powered DNN Serving Platform
Fast:
Flexible
Friendly:
F F F
L0
L1
F F F
L0
Pretrained DNN Model
in TensorFlow, CNTK, etc.
Scalable DNN Hardware
Microservice
BrainWave
Soft DPU
Instr Decoder
& Control
Neural FU
Network switches
FPGAs
40. Azure ML and Project Brainwave
• New DNN models
• ResNet 152, DenseNet-121, VGG-16, SSD-VGG
• Customizable weights
http://aka.ms/aml-real-time-ai
Easily deploy models to FPGAs for ultra-low latency with
Azure Machine Learning powered by Project Brainwave
43. Why Intelligent Edge?
High-speed data processing,
analytics and shorter response
times are more essential than ever.
Intelligent Cloud
• Business agility and scalability: unlimited computing
power available on demand.
Intelligent Edge
• Can handle priority-one tasks locally
even without cloud connection.
• Can handle generated data that is too
large to pull rapidly from the cloud.
• Enables real-time processing through
intelligence in or near to local devices.
• Flexibility to accommodate data privacy related
requirements.
44. The components of a ML application
Vision
AI dev
kit
Vision
AI dev
kit
46. Vision AI Developer Kit
Hardware Specification
Tutorial: Develop a C# IoT Edge module and deploy to your simulated device
https://docs.microsoft.com/en-us/azure/iot-edge/tutorial-csharp-module
48. Vision AI Developer Kit
A connected camera reference solution
Altek version available to order soon at https://visionaidevkit.com
49. Q & A ?
James Serra, Big Data Evangelist
Email me at: JamesSerra3@gmail.com
Follow me at: @JamesSerra
Link to me at: www.linkedin.com/in/JamesSerra
Visit my blog at: JamesSerra.com (where this slide deck is posted via the “Presentations” link on the top menu)