The next evolution in cloud computing is a smarter application not in the cloud. As the cloud has continued to evolve, the applications that utilize it have had more and more capabilities of the cloud. This presentation will show how to push logic and machine learning from the cloud to an edge application. Afterward, creating edge applications which utilize the intelligence of the cloud should become effortless.
5. 1990s -Work on
machine learning
shifts from a
knowledge-
driven approach
to a data-driven
approach.
1997 — IBM’s
Deep Blue beats
the world
champion at
chess
2006 – Geoffrey
Hinton et al.
coins the term
“deep learning”
to explain new
algorithms that
let computers
“see”.
2010 — The Microsoft
Kinect can track 20
human features at a
rate of 30 times per
second
6. 2012 – Google’s X Lab
develops a machine
learning algorithm that is
able to autonomously
browseYouTube videos to
identify the videos that
contain cats.
2014 – Facebook
develops DeepFace, a
software algorithm that is
able to recognize or verify
individuals on photos to
the same level as humans
can.
2016 – Google’s artificial
intelligence algorithm
beats a professional
player at the Chinese
board game Go
7. AI Market Growth
In 2018, the global AI market is expected to be worth approximately 7.35 billion U.S. dollars.
9. Perfect Silicon Opportunity
GPUs
• NVIDIA
• TeslaV100 16GB GPU ($ 19,000 from Dell)
• NVIDIA 900-2G402-0020-010Video Card 900-
2G402-0020-010Tesla M60 16GB GDDR5 PCI
Express 3 Active Cooling Bulk (4,587.81 Newegg)
• AMD
• AMD Radeon Pro SSG 100-506014 16GB + 2TB
2048-bit HBM2 + SSG CrossFire SupportedVideo
Card (4,399.99 Newegg)
FPGAs
FPGAs contain an array of programmable logic
blocks, and a hierarchy of reconfigurable
interconnects that allow the blocks to be "wired
together", like many logic gates that can be inter-
wired in different configurations. Logic blocks can be
configured to perform complex combinational
functions, or merely simple logic gates like AND and
XOR. In most FPGAs, logic blocks also include
memory elements, which may be simple flip-flops or
more complete blocks of memory.
10. CAAS (Cognition as a Service)
• Cognitive Services
• Amazon (Amazon Rekognition, Amazon Lex)
• Google (Video Analytics, Speech Recognition)
• IBMWatson
• Many more…
11. Ease of Use
The explosion of Machine Learning technology has
made libraries that are easy to use and can make once
impossible seeming tasks possible.
Published: 24 September 2014
13. What is Deep Learning?
Deep learning is a class of machine learning
algorithms that:
• use a cascade of multiple layers of nonlinear
processing units for feature extraction and
transformation. Each successive layer uses the
output from the previous layer as input.
• learn in supervised (e.g., classification) and/or
unsupervised (e.g., pattern analysis) manners.
• learn multiple levels of representations that
correspond to different levels of abstraction;
the levels form a hierarchy of concepts.
16. Activation Function
• Sigmoid or Logistic
• Tanh or hyperbolic tangent
• ReLU (Rectified Linear Unit)
• Leaky ReLU
• Exponential Linear Unit
There are two major classification
function types:
• Linear Activation
• Non-Linear Activation
Visualization by Erik Reppel
19. Encoder and Decoder
An encoder is a network (FC, CNN, RNN, etc) that takes
the input, and output a feature map/vector/tensor.These
feature vector hold the information, the features, that
represents the input.The decoder is again a network
(usually the same network structure as encoder but in
opposite orientation) that takes the feature vector from
the encoder, and gives the best closest match to the actual
input or intended output.
20.
21. Deconvolutional Network
After the network is trained to understand the “style”, new artwork can be used as the input and create new art.
24. Partially Connected Layers
Discern features by looking at “parts” of
the image and finding which of “parts”
are important for what the data.
Visualization by Erik Reppel
26. Types of RNN
• Fully recurrent
• Independently recurrent
(IndRNN)
• Recursive
• Hopfield
• Bidirectional associative
memory
• Elman networks and
Jordan networks
• Echo state
• Neural history
compressor
• Long short-term memory
• Second order RNNs
There are many types of recurrent
neural networks.This presentation will
only go over it basically.
• Gated recurrent unit
• Bi-directional
• Continuous-time
• Hierarchical
• Recurrent multilayer
perceptron network
• Multiple timescales
model
• NeuralTuring machines
• Differentiable neural
computer
• Neural network
pushdown automata
33. Popularity
TensorFlow and Keras are also favorites
among deep learning researchers,
coming in #1 and #2 in terms of
mentions in scientific papers uploaded
to the preprint server arXiv.org
34. Perfect Cloud Opportunity
InstanceType Price Region
CloudTPU $6.50 USD perTPU per hour. US
PreemptibleTPU $1.95 USD perTPU per hour. US
CloudTPU $7.15 USD perTPU per hour. Europe
PreemptibleTPU $2.15 USD perTPU per hour. Europe
CloudTPU $7.54 USD perTPU per hour. Asia Pacific
PreemptibleTPU $2.26 USD perTPU per hour. Asia Pacific
39. Content Moderator Custom Vision Service Video Indexer
Process and extract smart
insights from videos
Customizable web service
that learns to recognize
specific content in imagery
Machine-assisted moderation
of text and images, augmented
with human review tools
Computer Vision API
Distill actionable
information from images
Face API
Detect, identify, analyze,
organize, and tag faces in photos
Emotion API
Personalize experiences
with emotion recognition
Video API
Analyze, edit, and process
videos within your app
40. Edge and Mobile Ready
CustomVision service has ready made project types for
mobile and edge.
44. Notebooks
IDEs
Azure Machine Learning
Workbench
Project Brainwave
VS Code Tools for AI
C A PA B I L I T I E S
Experimentation and
Model Management
Services
AZURE MACHINE
LEARNING SERVICES
Spark
SQL Server
Virtual
machines
GPUs
Container
services
SQL Server
Machine Learning
Server
ON-PREMISES
EDGE
Azure IoT Edge
TRAIN & DEPLOY
OPTIONS
AZURE
45. Manage project dependencies
Manage training jobs locally, scaled-up
or scaled-out
Git based checkpointing and version
control
Service side capture of run metrics,
output logs and models
Use your favorite IDE, and any
framework
Experimentation service
U S E T H E M O S T P O P U L A R I N N O V A T I O N S
U S E A N Y T O O L
U S E A N Y F R A M E W O R K O R L I B R A R Y
46.
47.
48. Run Anywhere
• Central processing units (CPU) are general-purpose
processors. CPU performance is not ideal for graphics
and video processing.
• Graphics processing units (GPU) offer parallel
processing and are a popular choice for AI
computations.The parallel processing with GPUs result
in faster image rendering than CPUs.
• Application-specific integrated circuits (ASIC), such as
Google’sTensorFlow Processor Units, are customized
circuits. While these chips provide the highest efficiency,
ASICs are inflexible.
• FPGAs, such as those available on Azure, provide the
performance close to ASIC, but offer the flexibility to be
reconfigured later.
51. IoT in the Cloud and on the Edge
IoT in the Cloud
Remote monitoring and management
Merging remote data from multiple IoT
devices
Infinite compute and storage to train
machine learning and other advanced AI
tools
IoT on the Edge
Low latency tight control loops require near real-
time response
Protocol translation & data normalization
Privacy of data and protection of IP
Symmetry
53. Concept – Module
• A module image is a package containing the software that defines a module.
• A module instance is the specific unit of computation running the module image on an
IoT Edge device.The module instance is started by the IoT Edge runtime.
• A module identity is a piece of information (including security credentials) stored in IoT
Hub, that is associated to each module instance.
• A module twin is a JSON document stored in IoT Hub, that contains state information
for a module instance, including metadata, configurations, and conditions.
54. • Installs and updates workloads on the device.
• MaintainsAzure IoT Edge security standards on the device.
• Ensures that IoT Edge modules are always running.
• Reports module health to the cloud for remote monitoring.
• Facilitates communication between downstream leaf devices and the IoT Edge device.
• Facilitates communication between modules on the IoT Edge device.
• Facilitates communication between the IoT Edge device and the cloud
57. Learning Resources
TensorFlow and Keras
• TensorFlow Programmer’s Guide andTutorials
• Keras getting started
• TensorFlowTutorials
• TensorFlowTutorials
• MoreTensorFlowTutorials
Azure ML and IoT Edge
• Azure ML Services
• Azure ML ServicesTutorial
• Azure MLWorkbench on Code Academy
• Microsoft IoT School
• Azure IoT Edge Docs
58. • Jared Rhodes
• Email – jrhodes@qimata.com
• Blog – Qimata.com
• GitHub – Qimata
• Twitter - @Qimata
Extending the power of the
cloud to the edge
Hinweis der Redaktion
The statistic shows the size of the artificial intelligence market worldwide, from 2016 to 2025. In 2018, the global AI market is expected to be worth approximately 7,35 billion U.S. dollars. Some current major uses of artificial intelligence include image recognition, object identification, detection, and classification, as well as automated geophysical feature detection. The largest proportion of revenues come from the AI for enterprise applications market.
