Zongjie Diao, Director of Product Strategy and Management, Cisco
Oct 2019
Operationalize Machine Learning for Real Impact
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
CIO’s top agenda
Gartner 2019 CIO Survey vs 2018 CIO Survey
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
AI Initiatives Seen Across Industries
Video Captioning
Content Based Search
Nl, Vr and Ar
Media and
Entertainment
Cancer Cell Detection
Drug Discovery
Medical Research
Healthcare
Fraud Detection
Cryptocurrencies
Algorithmic Trading
Finance
Face Recognition
Crowd Analytics
Cyber Security
Security and
Defense
Theft Detection
Auto Checkout
Targeted Marketing
Retail
Reduce Product Defects
Increase Production Speed
Shorten Downtime
Manufacturing
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
useless… and a waste of money
Machine Learning is
So said many companies who have tried…..
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
35%
50%
15%
How often are trained model deployed?
Rarely Often AlwaysUp to 85% trained models may never
be used
https://www.iianalytics.com/2019-analytics-predictions-and-priorities
Low deployment
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Long deployment
Expected AI
deployment
timeline
How long it really takes
Start planning
Still planning
Piloting
Restarting Finally launch
3-5X
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Source: forrester.com/predictions (2018)
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential https://www.mckinsey.com/featured-insights/artificial-intelligence/notes-from-the-frontier-modeling-the-impact-of-ai-on-the-world-
economy?cid=other-eml-alt-mgi-mck-oth-1809&hlkid=5ebe957bb3594f96bedda5695e4664fd&hctky=10366723&hdpid=677435cb-04b0-
445e-afba-4588aa47d2fe
The success of deployment
matters.
The speed of deployment
matters.
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
• Major ML deployment challenges
• Cisco solutions to operationalize machine learning
Agenda
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential #DataSymposium2019© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
• Major ML deployment challenges
• Cisco solutions to operationalize machine learning
Agenda
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
1
2
3
"Designed by brgfx / Freepik"
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Which process(es) below is (are) the most critical
component(s) of ML deployment?
A. Data collection B. Data preparation
C. Training and evaluation D. Inference and
deployment
1
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Collect
Clean
Correlate
Train
Data Pipeline for Multiple Data Sources
Collect Clean Correlate Train Data
Model
Result
Collect Clean Correlate Train Model
Collect Clean Correlate Train Model
Social
Video
Model
More
Data
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Who are the key stakeholders to ensure a
successful ML deployment at scale? (select
multiple choices)
A. Data scientists B. Data engineers
C. CXO &Line of
business D. IT
2
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Framework and SW instruments need to be supported
ML/DL Framework /
Infrastructure
Data
Infrastructure
Inferencing &
Ingestion End
Point
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Data Scientists
Data Engineers
IT Team
• Rapidly evolving open source ML frameworks
• Start with workstation or cloud
• Hard to scale for production
• Want cloud like experience on-prem
• Lack of AI/ML expertise
• Need new infra architecture
• Need to solve silos and manageability
• Need enterprise readiness and security
Major challenge gaps exist between key stakeholders in realizing ML impact
CIOs Business
Leaders
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
How much percent of Machine Learning models are
developed on Cloud vs. on-premise?
A. 90% vs. 10% B. 70% vs. 30%
C. 40% vs. 60% D. 10% vs. 90%
3
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
The benefits of both – it’s a hybrid world when rubber meets the road
ML model
development
~60% ~40%
Source: Gartner, Market Guide for Machine Learning Compute Infrastructures, Sep 2108, ID: G00362287; Figure 3
Data gravity and integration
Performance
TCO
Governance
Remodel, retraining
At scale production
Fast deployment
Test-dev
Simplicity
APIs
One time
Cloud On-Prem
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Cloud Experience On-prem deployment
Gap 2:Consistent machine learning experience across cloud and on-prem
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
1
2
3
"Designed by brgfx / Freepik"
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
What can Cisco do?
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Data scientists / engineers IT
Cloud experience On-prem deployment
Dedicated/Silo’d
infrastructure
Shared/Integrated
infrastructure
“Operationalizing Machine Learning” is key to realize the real impact of AI/ML - Cisco is here to bridge the gap
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Data
Engineer
s
Data
Scientist
s
IT Team
+
HX Turnkey Starter Kit
for dev and
deployment
UCSD Workflow
˝
UCS C480 ML
UCS/HX C220 Servers
UCS C240 Servers
HyperFlex GPU Nodes
Inference Test/Dev Deep Learning
Cisco Solutions
IT needs to be part of the design team for AI/ML from the beginning
NGC
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
UCS C480 ML Rack Server A No-Compromise Purpose Built Server for Deep Learning
8 SXM2 Nvidia V100 GPUs
NVlink GPU Interconnect
Latest Intel Processors
UCSM & Intersight managed
Up to 24 drives with 6 NVMe
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
HyperFlex 4.0 for Inferencing on the Edge
Video Customers Customer
Sentiment
Video Inferencing
HyperFlex 4.0
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Cisco Hybrid Cloud
Cisco is #2 contributor to
Kubeflow with 2.5M lines of code
HyperFlex
Data Collect Clean Correlate Train
UCS + Cloud On Prem
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Cisco: Unified Architecture
App Data Source Storage ML Process Inference
Common Ethernet Fabric
Collect data
Prepare Data
Train models
Evaluate models
Deploy & Improve
Cisco CCP Nvidia vGPU
+
Dedicated Shared
Data Scientists Experience
• Dedicated workstation vs. Share ML infrastructure
Data Center Architecture
• Silo ML infra vs. Integrated Data Center Architecture
Networking Architecture
• InfiniBand vs. Scalable Ethernet
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Cisco UCS Cisco HyperFlex Cisco ACI Cisco Multicloud Portfolio
Cisco AI/ML- A Holistic Approach: Edge to Core
Accelerated Computing
for Inference at the Edge
Accelerated Computing
for AI/ML in the Core
Big Data/Analytics
and ERP
AI/ML Stack
Partnerships
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Let’s see how Cisco is helping keep the world safe, clean
and convenient through our ML portfolio and eco-system.
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Let’s see how Cisco is helping keep the world safe, clean
and convenient through our ML portfolio and eco-system.
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential © 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Smart City Solution Customer Use Case Example
Sm
art
Cit
y i
n
a B
ox
Ze
ro t
ou
ch
Pro
vis
ion
AI/
ML
Dee
p L
ea
rnin
g
Citizens
Digitization
Smart Parking
Security & Safety
Cities
Solid Waste Management
Intelligent Traffic, Water Mgmt.
Inte
llig
en
t D
ata
Mg
mt.
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Cisco DC Solutions for Smart City
Smart City in a box
Governance
City Automation
Safety & Security
Secure Delivery of Apps
Visual Analytics
Cisco Hyperflex Anywhere,
Cloud to DC to Edge
Cisco VSM Surveillance,
S3260, CVLT, Cohesity
Cisco UCSM,
UCSD, CWOM
C480 ML Deep Learning
C240 M5 Inference
Single Console
Cisco Cloud Center
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Cisco Security Architecture for Smart Cities
Awiros
Cross Vertical(parking , Lighting)
Kinetic for City - HX
Edge
Camera
Analytics
Camera
Video Surveillance - 3260
Video Analytics ( AI/ML) – 480 ML
Software
Predictive Analytics
Data Aggregation
and Normalization
Analytics
Sensor Management
Sensor / Camera
Alerts
and Events
Alerts
and Events
Alerts and Events
Alerts and Events
1
2 2 1
2
4
5 Dashboard ( GeoShield)
Camera Camera
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
https://www.cisco.com/c/en/us/solutions/data-center/artificial-
intelligence-machine-learning/index.html
Cisco DevNet, Cisco DMZ Lab
Operationalizing ML is
the key to success!
Cisco has solutions and is ready to
help!
© 2018 Cisco and/or its affiliates. All rights reserved. Cisco Confidential
Thank you!