Open Innovation for a Fully Connected and Intelligent World · 2018-11-20 · A Fully Connected and...
Transcript of Open Innovation for a Fully Connected and Intelligent World · 2018-11-20 · A Fully Connected and...
Open Innovation
for a Fully Connected and
Intelligent World
A Fully Connected and Intelligent World is Coming
Mobile broadband and APPs Better than the Von Neumann’s Architecture Core Network / Inter Data Center
9 years, 180X traffic estimated by GSMA
Global mobile networks Global cloud computing
Spine
Leaf
Tor
Super Spine/Core
Consumer
Devices
Networks
Enterprise
and Cloud
Wireless
Optical
IP
OTT
OTC
Verticals
Huawei R&D in Europe: Open and Joint Value Creation
research
solution
European technical experts
European Academics
technical collaboration
projects
63M €
investment from Huawei in
technical partnerships with
over 100 of European top
universities, research institutes
and enterprises in 2017
230
200+
180+
1,600industry
partners
TO DEVELOP JOINT
Innovative
Solutions
Huawei R&D in Europe
Close to industrial and academic partners, accelerate ICT-based innovations of future life and future work.
Telecom
IT
ITS / Mobility
Industry 4.0
IoT
E-/M-Health
Wireless Networks
Key Technical InnovationIndustry Solution Development Fundamental Research
Cloud and Big Data
Optical System
Terminals and IoTings
+ +New Materials
New Theories
New Components
New Devices
Better Perception
COMPUTER
VISION
NLPLOCAL
DECISION
ROBOT
TECH
AIR SENSOR
TASTE
SENSOR
More Information, Better Service & Experience
Better Connection
Autonomous /
Flying
Transportation
Digital Industry and
Robotics
XR and Edge
Intelligence
Better Intelligence
Spectral-
Methods
(FFT)
Graph-
Algorithms
Map-
Reduce
Dense /
Sparse-Linear-
Algebra
C-A GEMM/ SPMV
Graphical-
Models
(Tree)
C-A BFS/DFS
Finite-State-
Machines
NFA/DFA
Circuits Monte-Carlo Structured/
Unstructured-
Grids
N-Body
Methods
Molecular Model
Dynamic
Programmin
g
Needleman Wunsch
(Sequence alignment)
Backtracking
and Branch
& Bound
N-Queens
(Artificial Intelligence)
Network
Packet Data
Forward
Network
Protocol/
Data Aware
Software
Radio
Network
Data
Security
Data
Storage
Data
Analysis/Searching
Machine
Learning
Multi-
media
Analysis
Computer
Vision
Cancer
Genomics
Signal
Processing
Recommen
dation
? ?
To be explored.
Collaborations for New Materials, New Devices
ABEL
MOBILFUNK
AMPLEON
ARM
BOLLORE
CEVA
DHL
EPCOS
FPC
INFINEON
NXP
PANALPINA
RF360
SALCOMP
ST
WPP
The world’s first
smartphone with
independent de
pth computing c
hipset
Connection: High Performance & Autonomous Networks
Performance optimization based on air interface big data analysis
Trace tA Trace tB Trace tC
Span t3, t3’Span t2, t2’Span t1, t1’
Span tx, tx’
La
ten
cy (
mse
c)
Date/time
250 msec
Span t1
Span t3
Span t3
t1 t2t3
Trace tA
Trace tB
Trace tC
430 msec
Sliding window
General idea
• For i in time_series indexes:
• For sliding window w[i, i+wsize]
• Calculate μ1 of w[i, i+wsize/2]
• Calculate μ2 of w[i+wsize/2,
i+wsize]
• Anomaly if abs(μ2/μ1) >= k
Anomaly and Pattern Detection
Time-Series Analysis
1 2
4
5
INPUT – LSTM – SAMPLING – LSTM – OUTPUT
Stability Analysis Deep Learning
Trace Analysis 3
AI assisted cloud operation
Better Performance & Less TCO
Data analytics and machine learning tools are applied
to predict and further optimize:
• Link performance including the radio link and
backhaul.
• The traffic requirements (and QoS) of machine-
type communication traffic (including vehicle
communication).
AI assisted cloud operation: improving the efficiency
and stability. Bring intelligence into Vehicular
Networks.
Computing: Full Stack and Full Scenarios AI Solution
CANN(Compute Architecture for Neural Networks)
ModelArts
General
APIs
Advanced
APIsPre-integrated
Solutions
Ascend
All Scenarios
MindSpore TensorFlow PyTorch PaddlePaddle …
Application
Enablement
Framework
Chip Enablement
IP & Chip
HiAI Engine
HiAI Service
Ascend-MaxAscend-MiniAscend-Tiny Ascend-LiteAscend-Nano
Fu
ll S
tack
AI Applications
Consumer DevicePublic
Cloud
Private
Cloud
Industrial
IoT
Device
Edge \
Computing
Compute
Power budget
Model size
Latency
Inference ?
Training ?
Ascend-SKU
Device
Earphone Always-on Smartphone Laptop
20 MOPS 100 GOPS 1-10 TOPS 10-20 TOPS
1 mW 10 mW 1-2 W 3-10 W
10 KB 100 KB 10 MB 10-100 MB
< 10 ms ~10 ms 10-100 ms 10-500 ms
Y Y Y Y
N N Y Y
Nano Tiny Lite Mini
Edge
IPC Edge Server
10-20
TOPS
10-100
TOPS
3-10 W 10-100 W
10-100 MB 100+ MB
10-500 ms ms ~ s
Y Y
Y Y
Mini Multi-Mini
Cloud
Data Center
200+ TOPS
200+ W
300+ MB
ms ~ s
Y
Y
Mini
Unified architecture for all chips with high performance and efficiency.
Unified training and inference framework with consistent development experience and greater development efficiency.
Autonomous, Electrified, Connected Vehicle: Connection
ICT is the key enabler for safety, efficiency,
on-demand service and experience.
Platooning for Logistics
20% gas saving!
Tele-Operated Driving
5G Enabled
Sensor sharing
uplink
Cloud
downlink
direct link
Direct LinkCellular Network
+• Support relative speed up to 500km/h
• Low latency Communication
• GNSS-based side-link synchronization
• eNB-scheduled side-link for better performance
• Uplink semi-persistent transmission
• Shorter multicast/broadcast period
• QoS for ITS applications
Online
Navigation
Congestion
warning
Multi-media
download
Speed limit
alarm
Speed
Advisory
OTT/APPs 5GNR V2X Uu 5GNR V2X Sidelink E2E latency
3~10ms
Reliability
99.999%
Positioning
0.1~0.5m
Electrical
Connected
Intelligent
Transportation
New VehicleAuto driving, connected
New InfrastructureIntelligent infrastructure based on new network
New NetworkConverged communication
Sharing
Infotainment
Autonomous, Electrified, Connected Vehicle: Computing
14 x Cameras | 6 mmWave radars +
12 ultrasonic radars + 8 lidars
High Performance
352 TOPS
Real-time OS kernel latency <10 us
ROS latency < 1 ms
Guaranteed Low Latency
< 200 ms
ASIL D
Safe & Highly Reliable
ISO 26262 ASIL D
Active-Active | -40°C to +85°C
1 TOPS/W
System Efficiency
AI Core efficiency 3 TOPS/W
Chipset efficiency 2 TOPS/W
ASIL: Automotive Safety Integrity Level (a top security standard)
Connected Industry Automation
Traditional Automation Pyramid and 5G
Public cloud
Enterprise cloud
/edge
Latency
~50 ms
<1 ms
Mostly broad band traffic (eMBB)
Less requirements to low latency and
extreme high reliably
LTE-evolution
5G (3GPP Rel 15+)
Less broad band traffic
Strong requirements to low latency
and extreme high reliably, and
deterministic traffic (URLLC)
• 5G (3GPP Rel 16+)
• 5G mmWave
• 5G as TSN switch
Reliability99.9%
>99.9999%
Storage
Server
Gateway
Gateway
Gateway
Enterprise cloud
/edge
Storage
Server
Option A Option B
Higher productivity, lower cost, more flexible and on-demand service.
Modernization of Operation Room
Illumination
OR-Lamps telefonepager
Endoscopy
Electrocautery
X-ray
Ultrasound
TeleconsultationPACS
Data-management
Suction
OR-table
Server
Better quality, higher efficiency and lower cost.
Sensor + Connection + AI
Global Consensus
Standard
Organization
Industrial
Organization
Industrial
Association
Policy and
Regulation
Technical
Innovation Organization
Global
Organization
European
Organization
Country
Organization
3GPP
IETF
ITU-T(Swiss)BBF
TMF
IEEE
ISO(Swiss)IEC(Swiss)
GSMA(UK)NGMN(GER)ISG NFV(FR)ISG MEC(FR)OneM2M
WIoT
mmWTA
5GAA
5GACIA
5GPPP(BEL)WWRF
H2020(BEL)FP7(BEL)
ITU-R(SWISS)
OFCOM(UK)UK Spectrum Policy Forum
(]UK)BNetzA(GER)PTS(SWEDEN)
AIOTI(BEL)Industry 4.0(GER)ECP(BEL)Digital Europe(BEL)ETNO(BEL)
ETSI(FR)CEN(BEL)CENELEC(BEL)ECMA(SWISS)
EC(BEL)CEPT(DAN)EFTA(SWISS)
UK 5G IC(UK)
German 5G IC(GER)Tech UK(UK)ICT KTN(UK)HyperCat(UK)
BSI(UK)AFNOR(FR )DKE(GER)DIN(GER)BITKOM(GER)
TCCA(UK)
Enabling a Fully Connected and Intelligent World
Win – Win
Collaborations
Research,
Development,
Standardization,
Industrialization,
Products and
Solutions
KUKA
5GPPP & IMT-2020
ETSI 3GPP AIOTI GSMAProducts and Solutions
Cloud and Enterprise Solutions
5G, IoT and Fixed Networks
Consumer Devices
Aalto University
Thank you.