In this talk, Tong will start with the current landscape and typical use cases of Artificial Intelligence applications in the Telco domain. Then, she will introduce Intel’s strategy and products for Network AI, including our focus areas, our hardware portfolio, software stacks, roadmaps and some case studies. Speaker: Tong Zhang, Principal Engineer and Chief Architect for AI and Analytics of the Network Platforms Group, Intel

1. Tong Zhang, Principal Engineer
April 25, 2018

2. Consumer Health Finance Retail Comms Government Energy Transport Industrial Other
Smart Assistants
Chatbots
Search
Personalization
Augmented
Reality
Robots
Enhanced
Diagnostics
Drug
Discovery
Patient Care
Research
Sensory
Aids
Algorithmic Trading
Fraud Detection
Research
Personal Finance
Risk Mitigation
Support
Experience
Marketing
Merchandise
Loyalty
Supply Chain
Security
Cyber Security
Predictive
Management
Self-Optimized
Network
Location-based
Marketing
Network Edge
Analytics
Defense
Data
Insights
Safety &
Security
Resident
Engagement
Smarter
Cities
Oil & Gas
Exploration
Smart
Grid
Operational
Improvement
Conservation
Autonomous Cars
Automated
Trucking
Aerospace
Shipping
Search & Rescue
Factory
Automation
Predictive
Maintenance
Precision
Agriculture
Field Automation
Advertising
Education
Gaming
Professional & IT
Services
Media
Sports
Source: Intel forecastIdentifying Networking Workload Synergies with Other Verticals
Networking/Security
Relevant areas of Synergy
with CoSP/TelcoAI is transformative across industries
Earlyadoption

3. Many use cases – just opportunities ahead of us
Network Core services
 Revenue assurance (OSS/BSS), fraud detection
 Self Optimizing Network, Self Healing Network
 MANO integration e.g. Service simulation, Auto provisioning
 Predictive maintenance, Alarm Correlation, Root Cause Analysis
 Churn prediction, QoE analysis, Service assurance
Hybrid Network Core & Edge
 NLP (Natural Language Processing ), NLU, Chatbot (Virtual agent)
 Cyber Security e.g. DDOS, anomaly detection
 IoT – Edge Analytics, Video surveillance, Video summarization
 Autonomies systems (Cars, Drones, Robots, Factory automation)

4. 5G and EDGE services – AI will play a key role
Smart Cities & Buildings
Autonomous Driving
Retail
VR Gaming
1 Source: iDATA/Digiworld, 2013
2 Source: Infonetics Research, October 2014
3 Source: Network World, October 2014
NFV REVENUE2$8.1B
SDN Market
Growth3
65%
NEW MOBILE
SERVICES4$600B
NEW CLOUD
SERVICES5
$200B
NEW VIDEO
SERVICES6$200B
CDN/Edge Caching
Industrial Control

5. Machinelearning DeepLearning
Example
Features
 Detect similarities &
anomalies in sea of data
 Large, diverse dataset
 Fully-explainable
 Real-time updates
 Practical to
‘reverse engineer’
 Tabular/limited dataset
 Good enough accuracy
 Fully-explainable
 Difficult problem to
‘reverse engineer’
 Large, uniform dataset
 Highest accuracy
Other
examples
Credit fraud detection
Issue and defect triage
Predictive maintenance
 Regression
 Anomaly detection
 Feature extraction
 Image/speech recognition
 Natural language
processing (NLP)
 Pattern detection
MULTIPLE approaches to AI
Anti-Money
Laundering
Facial
recognition
Recommendation
engine
Cognitive
Reasoning
ANDMore…

6. Source: Forrester Research – Artificial Intelligence: Fact, Fiction. How Enterprises Can Crush It; What’s Possible for Enterprises in 2017
AI adoption is just beginning
58%
of business and technology
professionals said they're
researching AI, but only…
12%said they are currently
using AI systems.
In a recent Forrester Research survey…

7. Intel portfolio for Telco AI/Analytics
7
Mllib BigDLOptimizedframeworks
Intel® DL Studio
Intel® Deep Leaning
Deployment Toolkit
Movidius
Fathom
Intel® Math Kernel Library
(MKL, MKL-DNN)
Intel® Data Analytics
Acceleration Library
(DAAL)
Intel Python
DistributionAccelerationlibraries
Intelml/dlplatforms
ComputingHardware GNAVision
Cospexperience

8. Real-time
Inference
Mainstream
Training
Intensive
Training
Mainstream
Inference
Higher Inference
Throughput
Vision
1-20W
Speech/Audio
1-100+mW
Mainstream
Inference
Autonomous
driving
Custom
Inference
Intel AI compute
Intel
GNA
(IP)
Mainstream AI
Flexible
Acceleration
General
AI
DeepLearning
traininginference
DataCenter/
Workstation
DataCenter/
Workstation
EDGE/Gateway
‡ Future
All products, computer systems, dates, and figures are preliminary based on current expectations, and are subject to
change without notice.
‡

9. End-to-End AI compute
Cloud/appliance gateway EdgeMany-to-many hyperscale for stream
and massive batch data processing
1-to-many with majority
streaming data from devices
1-to-1 devices with lower power and
often UX requirements
Ethernet
& Wireless
Wireless and non-IP
wired protocols
 Secure
 High throughput
 Real-time
Intel® Xeon® Processors
Intel® Core™ & Intel Atom® Processors
Intel® FPGA
Intel® Xeon Phi™ Processors*
Intel® Processor Graphics
Intel® Movidius™ VPUVision
(DC=Datacenter. WS = workstation
*Formerly codenamed as the Crest Family
All products, computer systems, dates, and figures specified are preliminary based on current expectations, and are subject to change without notice.
CPU+
Intel® Nervana™ Neural Network Processor (NNP)✝
Intel® GNA (IP)*Speech

10. INFERENCE THROUGHPUT
Up to
198x
Intel® Xeon® Platinum 8180 Processor
higher Intel optimized Caffe GoogleNet v1 with Intel® MKL
inference throughput compared to
Intel® Xeon® Processor E5-2699 v3 with BVLC-Caffe
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer
systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating
your contemplated purchases, including the performance of that product when combined with other products. For more complete information visit: http://www.intel.com/performance Source: Intel measured as of June
2017. Configurations: See the last slide in this presentation. *Other names and brands may be claimed as the property of others.
TRAINING THROUGHPUT
Up to
127x
Intel® Xeon® Platinum 8180 Processor
higher Intel Optimized Caffe AlexNet with Intel® MKL
training throughput compared to
Intel® Xeon® Processor E5-2699 v3 with BVLC-Caffe
Deliver significant AI performance with hardware and software optimizations on Intel® Xeon® Scalable Family
Inference and training throughput uses FP32 instructions
Up to 191X Intel® Xeon® Platinum 8180 Processor higher Intel optimized Caffe Resnet50 with Intel® MKL inference throughput compared to Intel® Xeon® Processor E5-2699 v3 with BVLC-Caffe
Up to 93X Intel® Xeon® Platinum 8180 Processor higher Intel optimized Caffe Resnet50 with Intel® MKL training throughput compared to Intel® Xeon® Processor E5-2699 v3 with BVLC-Caffe
Performance estimates were obtained prior to implementation of recent software patches and firmware updates intended to address exploits referred to as "Spectre" and "Meltdown." Implementation of these updates may make these results
inapplicable to your device or system.
Hardware plus optimized software
Intel® Xeon® processor platform performance
Optimized
Frameworks
Optimized Intel®
MKL Libraries

11. Directoptimization Intel®ngraph™library
Translates participating deep learning
framework compute graphs into
hardware-optimized executables for
many different targets
(CPU, GPU, NNP, FPGA, VPU, etc.)
Intel AI libraries
MKL-DNN
Open-source optimized deep
neural network functions for
new frameworks
clDNN
Open-source optimized deep
neural network functions for
Intel GPUs
DAAL
Data Analytics Acceleration
Library for analytics and
machine learning
Intel Python
Distribution
Optimized distribution of most
popular & fastest growing
language for machine learning
Framework Framework Framework Framework Framework

12. software.intel.com/intel-distribution-for-python
Easy, Out-of-the-box Access
to High Performance Python
 Prebuilt, optimized for numerical
computing, data analytics, HPC
 Drop in replacement for your
existing Python (no code changes
required)
Drive Performance with
Multiple Optimization
Techniques
 Accelerated NumPy/SciPy/Scikit-
Learn with Intel® MKL
 Data analytics with pyDAAL,
enhanced thread scheduling with
TBB, Jupyter* Notebook interface,
Numba, Cython
 Scale easily with optimized MPI4Py
and Jupyter notebooks
Faster Access to Latest
Optimizations for Intel
Architecture
 Distribution and individual
optimized packages available
through conda and Anaconda
Cloud
 Optimizations upstreamed back to
main Python trunk
For developers using the most popular and fastest growing programming language for AI
Intel distribution for Python*
Advancing Python Performance Closer to Native Speeds

13. Other names and brands may be claimed as the property of others.
ON
**
Growing list of open-source frameworks now optimized for CPU
Today’s most
used deep
learning
framework,
includes ML
functionality,
led by Google
Established
deep learning
framework for
computer
vision/CNN,
led by
Berkeley
(BVLC)
Cognitive
toolkit / deep
learning
framework, led
by Microsoft
Robust deep
learning
framework, led
by Amazon &
DMLC
Flexible DL
framework for
mobile & large
scale
deployment,
led by
Facebook
Framework for
deep learning
at scale in
Apache Spark
workflow,
supported by
Intel
Intel’s
reference
deep learning
framework,
originally from
Nervana
Systems
More
frameworks
enabled are
always under
consideration
for POR
and/or
optimized via
Intel® nGraph™
Library
Andmore…
ONNX: Open Neural Network eXchange
Intel AI frameworks
Intel-led
Mostpopular
Most popular ‘machine learning’ library, now optimized via
Intel® Data Analytics Acceleration Library (DAAL)
Popular ‘machine learning’ library for
Apache Spark framework
ML
MlLib
ON
DL
Limited
availability today
Limited
availability today

14. Intel®deep
learningstudio
Intel®deep
learningdeploymenttoolkit
Intel©
computervisionsdk
Intel®
Movidius™Sdk
Convert
&
Optimize
Trained
Model
Facilitates optimized
inference deployment
models trained using the
following frameworks:
TensorFlow, Caffe, or
MXNet
Enterprise customer
tool to compress full DL
development cycle;
coming to the Intel®
Deep Learning System
DL deployment kit
-PLUS-
OpenCV* and OpenVX*
support for deep
learning-based
computer vision on
CPU, IPU, GPU & FPGA
Intel® Movidius™ Vision
Processing Units (VPU)
software development
kit for inference
deployment
Inference
on
HW Target
Intel AI tools
*Other names and brands may be claimed as the property of others.

15. Telco AI focus areas
• Network analytics
 Network operation: service quality assurance; security analytics; predicative analytics
and anomaly detection
 User data analysis: monetization of user data; churn prediction and fraud detection;
monitoring user behavior for public affairs
• Edge cloud (MEC)/5G
 Digital surveillance system; automated driving; connected industrial robots
• Media/Video processing
 Media analytics (video/image/speech, etc.)
 Smart call center: voice bot/chat bot; big data platform

16. Compute Network Storage
Existing
Management/
Analytics Systems
Intel® Run Sure Technologies
Resilient System Technologies Resilient Memory Technologies RAID
Open Standard
Presentation
MANO SDN
Emerging
Analytics
Systems
Resource
Telemetry
Interfaces
Open Collection
IPFIX
CLI sFlow Future
REST
APIs
Base Platform
SYSLOG
Standard Driver InterfacesStandard OS Telemetry Open Virtual Switching
Intel® Infrastructure Management Technologies
Platform service assurance system

17. Spot Landscape
DataSources
DNS
Infrastructure Logs
Proxy
Infrastructure
Logs
Routers with
Netflow Protocol
Enabled on Interfaces
SpotGUI
Security and
Context Use
Cases Share
Information Using
STIX TAXII
Data Integration Data Storage Machine Learning
Collectors
Filesystem
(HDFS)
Spot ML
Algorithms
Spark
Master Node (s) Cloudera
Manager/Navigator
Native Authentication
LDAP / Kerberos
Authentication
Cluster Administration
Cloudera Manager
Machine
learning
Generates
CSV Files
with the
Results
Operational Analytics Adding Context Using
Reputation Services for Public IP Address
(GTI) Facebook Threat Exchange
Defining the
Interface to Share
the Suspicious
Connections with
I-Sec Products
and Other Brands.
Spot
Visualization
Server/iPython
Server
Spot Data API
nfdump
tshark
parser
TLS Https 443
Streaming Data
MLlib
Machine Learning
Algorithms Output,
Apache Spot
Recommend Scala
VPC Flows AWS in Development
Batch
Processing
Spot Data API
Security
Orchestration
DXL and
the OSC
Apache Spot for cyber security

18. Acumos AI open source platform
Source: https://www.acumos.org/

19. ENI – Experiential Network Intelligence
19
Network Operations
Policy-driven IP managed networks
Radio coverage and capacity optimization
Intelligent software rollouts
Policy-based network slicing for IoT security
Intelligent fronthaul management & orchestration
Service Orchestration and Management
Context aware VoLTE service experience optimization
Intelligent network slicing management
Intelligent carrier-managed SD-WAN
Network Assurance
Network fault identification and prediction
Assurance of tight service requirements with network slicing
Infrastructure Management
Policy-driven IDC traffic steering
Handling of peak planned occurrences
Energy optimization using AI
Using Artificial Intelligence techniques in the network management system

20. Learn Develop Share
 Online tutorials
 Webinars
 Student kits
 Support forums
 Intel Optimized
Frameworks
 Exclusive access
to Intel® AI
DevCloud
 Project showcase
opportunities at
 Intel Developer
Mesh
 Industry &
Academic events
 Comprehensive
courseware
 Hands-on labs
 Cloud compute
 Technical Support
Teach
For developers, students, instructors and startups
Intel®AIacademy
software.intel.com/ai

21. More information at ai.intel.com
Findoutmore
learn
explore
engage
Use Intel’s performance-optimized
libraries & frameworks
Contact your Intel representative for
help and POC opportunities

22. Notices and Disclaimers
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such
as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of
those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your
contemplated purchases, including the performance of that product when combined with other products. For more information go
to www.intel.com/benchmarks.
Benchmark results were obtained prior to implementation of recent software patches and firmware updates intended to address exploits referred to
as "Spectre" and "Meltdown". Implementation of these updates may make these results inapplicable to your device or system.
Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software or service activation.
Performance varies depending on system configuration. No computer system can be absolutely secure. Check with your system manufacturer or
retailer or learn more at intel.com.
Details on claims for slide 10. Testing conducted by Intel. 198x inference throughput claim configuration: New case of Intel® Xeon® Platinum 8180
Processor higher Intel optimized Caffe GoogleNet v1 with Intel® MKL inference throughput compared to base case of Intel® Xeon® Processor E5-2699
v3 with BVLC-Caffe. 191x inference throughput claim configuration: New case of Intel Xeon Platinum 8180 Processor higher Intel optimized Caffe
Resnet50 with Intel® MKL inference throughput compared to base case of Intel Xeon Processor E5-2699 v3 with BVLC-Caffe. 127x training
throughput claim configuration: New case of Intel Xeon Platinum 8180 Processor higher Intel Optimized Caffe AlexNet with Intel® MKL training
throughput compared to base case of Intel Xeon Processor E5-2699 v3 with BVLC-Caffe. 93x training throughput claim configuration: New case of
Intel Xeon Platinum 8180 Processor higher Intel optimized Caffe Resnet50 with Intel® MKL training throughput compared to Intel Xeon Processor E5-
2699 v3 with BVLC-Caffe.
Intel, the Intel logo, Xeon, Intel Atom, Intel Core, Iris, Movidius, Stratix, Intel Nervana, and Intel Xeon Phi are trademarks of Intel Corporation or its
subsidiaries in the U.S. and/or other countries.
*Other names and brands may be claimed as the property of others.
© Intel Corporation.