Inference platform

Transportation industry insights

Crowd density prediction

Traffic status diagnosis

Data Lake

+

Edge computing

TrafficGo

3G/4G/5G/Ethernet

Congestion cause analysis

Traffic signal optimization

Geomagnetic sensors

Checkpoints

Induction coils

Cameras

Traffic control optimization

Radars

Carrier data

Floating car data

Weather data

Notes:

https://www.huaweicloud.com/product/trafficgo.html

Advantages:

Comprehensive and in-depth data mining: Fully integrates Internet and transportation big data to explore the value of big data.

District-wide coordination: Maximizes traffic volume, minimizes vehicle wait time, and coordinates travel requirements of vehicles and pedestrians for smooth traffic.

Real-time traffic signal scheduling: Formulates the first security communication interface standards for intelligence-infused traffic management and signal control systems.

Precise tracking and planning: Accurately predicts trajectories and plans routes in advance.

Full time: 24/7 traffic incident monitoring in all areas

Intelligent: regional coordination and optimization for traffic lights

Complete: identification of key congested points and paths, and impact analysis of traffic congestion

Predictive: Crowd density prediction makes you master crowd migration patterns.

Accurate: 24/7 master of comprehensive and accurate traffic conditions

Convenient: real-time traffic light scheduling for on-demand release

Visualized: Live traffic situations are displayed on big screens.

Refined: refined management of key vehicles

<!-- Slide number: 10 -->

Industrial Intelligent Twins

HUAWEI CLOUD Industrial Intelligent Twins builds an industrial intelligent platform that deeply integrates industrial knowledge and AI to adapt to frequent changes of working conditions and break the application limit of the industrial mechanism. In addition, Huawei works with industry-leading know-how customers and partners to build an open and win-win industrial intelligent ecosystem to make enterprises more intelligent and promote industrial upgrade.

Solution architecture

A simple and efficient platform for industrial AI development

Industrial knowledge graph

Knowledge market

Machine & AI model

Factory digital twin

Policy configuration

Industrial intelligent data lake

Multimodal data

Mechanism models

AI models

Multi-scenario, lightweight operating environment

Dynamic update

Industrial knowledge

One-stop, data-governance development environment

Multimodal, low-code development environment

Microservice and graph-based mechanism model

Multimodal data

Data modeling

Mechanism integration

Model development

Operating platform

Development platform

Notes:

https://www.huaweicloud.com/product/ei_industrial.html

Three changes:

From manual experience to data intelligence: New experience in improving efficiency and product quality can be obtained from data using data mining and analysis.

From digital to intelligent: Intelligent analysis has become a new driving force for digital transformation of enterprises.

From product production to product innovation: Data collaboration from product design to sales in an enterprise and upstream and downstream data collaboration in the industry chain create new competitive advantages.

Application cases:

Product quality improvement: Sort and analyze customer feedback, Internet comments, competitors, repair records, and historical sales data to discover critical product issues, optimize product design, and improve product quality.

Intelligent device maintenance: Use forecasting and inference methods such as time series forecast, neural network forecast, and regression analysis to predict whether a fault occurs, the fault occurrence time, and fault type based on system status. This design improves service O&M efficiencies, reduces the equipment downtime, and lowers the onsite service cost.

Production material estimation: Accurately analyze and estimate materials required for production based on historical material data, reducing the warehousing period and improving efficiencies.

In-depth algorithm optimization: Implement in-depth optimization based on the industry's time series algorithm model and Huawei's supply chain.

<!-- Slide number: 11 -->

EIHealth

HUAWEI CLOUD EIHealth covers genome, clinical research, and drug discovery. It focuses on people's health and promotes the combination of AI and healthcare to improve the service capability and coverage of healthcare. With Huawei's powerful computing, storage, and AI algorithms, EIHealth enables the genome, disease diagnosis, and pharmaceutical fields.

Solution architecture

1 + 3 + X: 1 EIHealth cloud platform + 3 sub-domain platforms + X enterprise/organization customized platforms (delivered with ISVs)

Platform applications

Gene database

AI drug discovery

AI-assisted patient screening

Multi-omics analysis

Medical image modeling

Computer-aided drug discovery

Drug discovery

Genome

Clinical research

| Genome AutoML | DeepVariant | GROMACS Molecular dynamics | AI model for patient screening | Virtual screening of drugs | Drug combination prediction |

| --- | --- | --- | --- | --- | --- |

| Data management | Process management | Development environment | Service management | Knowledge graph | AI Gallery |

EIHealth assets

EIHealth components

ModelArts

AI

OCR

GES

KGS

DIS

Big data

DLF

MRS

CSS

DWS

DLI

AOM

PaaS

SWR

DCS

RDS

APIG

CCE

Kunpeng

x86

IaaS

SFS

OBS

LB

Atlas

Notes:

Process: DeepVariant, single-cell sequencing, expression spectrum analysis, drug molecular docking, and virus genome analysis

Knowledge graphs: gene knowledge graph, medical knowledge graph, and drug knowledge graph

AI-powered drug discovery: target gene prediction, drug sensitivity prediction, drug combination prediction, and drug toxicity prediction

AI auto learning: genome auto learning (AutoGenome) and graph deep learning

Imaging models: ECG model, lung nodule model, lung CT model, stroke model, pathological section model, and federated learning model

Success stories: HUAWEI CLOUD works with Huazhong University of Science and Technology and BlueNet to launch AI-assisted diagnosis of COVID-19 to deliver AI medical image analysis capabilities in epidemic areas, improving the quality and efficiency of auxiliary diagnosis.

The Shennong project online platform is jointly developed by HUAWEI CLOUD and four scientific research institutions. It can assist doctors and R&D personnel in evaluating possible drugs against COVID-19 and can be used as a website for introducing antiviral drug discovery and popularizing drug discovery knowledge. The joint research team performed drug screening on the Mpro protein and S protein/ACE2 receptor of COVID-19. Five drugs that have good binding with the Mpro protein, Beclabuvir, Saquinavir, Bictegravir, Lopinavir, and Dolutegravir, were screened out.

<!-- Slide number: 12 -->

HeatingGo

Based on the existing auto control facilities and technologies, HeatingGo uses technologies such as AI and big data to implement intelligent heat supply supervision, operations, and control for heat sources, networks, stations, and users.

It provides a heating system that is inclusive, inexpensive, and controllable.

Cloud

Solution architecture

Residential building

NB-IoT

Data training

Data inference

Data preprocessing

Indoor temperature awareness

Heating policy

Data aggregation

Temperature awareness

Huawei AI

20℃

Policy promotion

Data reporting

Management and control apps of heating enterprises

Command and dispatch center of a heating enterprise

Data reporting

Data reporting

Heat exchanger

X

18℃

Heat source

Heat source sensing

User apps

User: temperature control based on temperature control panels and mobile apps

Heat source: heat demand forecast and complete closed-loop

Heat exchanger: AI-based smart policies for balancing heat supply

Building: AI-based smart policies

Primary circuit

Secondary circuit

Notes:

24/7 service: AI can avoid the impact of factors like the level of experience, sense of responsibility, and mood, and implement 24/7 auto adjustment.

Refined adjustment: Integrates multiple types of information, such as the indoor temperature, outdoor temperature, wind speed, and humidity, and intelligently formulates policies based on historical operating data, implementing hour-level adjustment and control policies without manual intervention.

Source network linkage: Implements source network linkage through heat demand prediction and optimal heat network adjustment to resolve supply-demand contradictions and achieve supply-demand balance.

Network-wide adjustment and control: Considering the mutual impact between heat exchangers, buildings, and households, adjusts the policy inference based on the entire network.

Continuous learning and optimization: Continuous learning and optimization based on real-time data, and continuous improvement of the inference algorithm make the algorithm more intelligent.

Comprehensive heating supervision: A heating information supervision platform is built using new technologies such as AI, big data, cloud, and 5G to implement dynamic collection, data governance, intelligent diagnosis, and scientific monitoring of heating operation data, improve the service quality of heating enterprises, and improve the city heating governance level.

<!-- Slide number: 13 -->

WaterGo

With technologies including AI and edge computing, Huawei is transitioning cameras into universal sensors capable of sensing all elements related to water, such as its volume and quality, making them "eyes" for river and lake monitoring. The cameras are widely distributed, intelligent, and responsive. Leveraging AI technologies, they improve the existing prediction and scheduling system, making water resource management more efficient and intelligent.

Chief


Drainage & flood control



Volume



Pollution handling




VAS


Quality



Management



Geo

Genius


API

Supervision





Cabinet


Schedule


KG


SDC


NLP


HUAWEI CLOUD


Solution architecture

Oriented to digital government and smart city sectors

Suitable for drainage and flood control, water pollution control, and water resource management

The universal AI platform provides services such as Video Analysis Service (VAS), ModelArts, Knowledge Graph, Natural Language Processing (NLP), FusionInsight, and GeoGenius.

Capable of monitoring and scheduling water volume and quality

Various edge devices and terminals can access the network.

Notes:

All-scenario video AI capability: The vision solution features low upfront investment, easy maintenance, controllable operational loss, the ability to interpret multiple messages from a single image, and auto-learning.

Device-edge-cloud synergy: Device-edge-cloud synergy with support for GPUs and Huawei's in-house D series chips significantly reduces communication and computing costs.

AI foundation for industries: Training and inference with massive amounts of data and knowledge, with both Huawei and third-party algorithms embedded

Effective problem disclosure: Summarize best practices, and help customers quickly close problems while formalizing the problem solving procedures.

Success stories: Suzhou Water built a video capture system for the Pingjiang River as a pilot project. This system automatically captures images of garbage dumping and sewage discharges in the Pingjiang River area. The aim is to build a smart, modernized river monitoring platform to alleviate the pressure of the river management team called "river chiefs"; improve work efficiency; reduce the operating expenditures; and ensure sustainable river management. The project was named an "Excellent Case of Smart Water Conservancy" by the Ministry of Water Resources of China, setting up an example for promoting the River Chief System (RCS) in other regions in China.

<!-- Slide number: 14 -->

GeoGenius

Powered by HUAWEI CLOUD's accumulation in cutting-edge technologies such as AI and big data, GeoGenius builds a one-stop, full-process intelligent development cloud platform for remote sensing. This cloud platform functions as both a data platform and an intelligent computing platform to help you focus on mining core values of data and developing application algorithms. It enables rapid application innovation in the industry and provides technical support for key tasks such as investigation, monitoring, evaluation, supervision, and law enforcement of natural resources, ecology, meteorology, environmental protection, and oceans.

Agriculture and forestry monitoring

Ecological environment monitoring

Natural resource survey

Emergency response and disaster prevention

Weather forecast

Marine conservation

Notes:

Natural resource survey: Use remote sensing image data and AI technologies to extract speckles of identified objects in a region, implementing quick survey of natural resources on a large scale.

Ecological environment monitoring: Use remote sensing images to dynamically monitor land use/coverage changes, ecological environment quality, ecological red lines, environmental impact of human activities, and natural resource development.

Agriculture and forestry monitoring: Use remote sensing to invert related vegetation indexes to monitor the growth trend of crops in the planting cycle, build crop growth models and yield prediction models, and predict the total crop yield and single crop yield.

Weather forecast: The measurement and reverse performance can accurately reflect the physical and ecological parameters of the atmosphere, land, and ocean and can be used for climate, atmospheric, and disaster monitoring.

Marine conservation: Use remote sensing technology to perform quantitative inversion of seawater quality, implement large-scale marine water quality monitoring, predict seawater pollution trends, and reduce sewage monitoring costs.

Emergency response and disaster prevention: Use remote sensing technology throughout the whole process of geological disaster investigation, monitoring, warning, and evaluation. It can meet the timeliness requirements of sudden geological disaster relief, and is an important information acquisition method for the prediction, investigation, analysis, and detection of debris flow disasters.

Success stories:

Based on HUAWEI CLOUD GeoGenius, the Key Laboratory of Remote Sensing and Geographic Information System (GIS) of Zhejiang University has built a set of global integrated earth observation system. The observation results of the system are mainly based on satellite remote sensing. Combined with observation means such as low- and medium-altitude ground remote sensing and IoT terminals, the earth system provides high-coverage, long-time sequence, and multi-frequency observation, delivering an innovative mode for remote sensing industry services.

<!-- Slide number: 15 -->

Smart Logistics Solution

The smart logistics solution provides AI-enabled 3D packing services to improve container loading rate. In addition, it provides the vehicle route optimization service to reduce customers' transportation costs.

Customers

Send a packing request.

(RESTful API)

Customers

In the response message of the packing service, obtain the packing plan to facilitate packing.

HUAWEI CLOUD

Work out a 3D packing plan with a high loading rate.

Customer API requests

1. Container information

2. Goods information

Customers

Packing plan

HUAWEI CLOUD

3D packing service

| | | | |

| --- | --- | --- | --- |

| | | | |

| | | | |

| | | | |

| | | | |

Improve container space utilization, reduce the number of boxes

Reduce the operational cost

Improve the overall logistics operation efficiency

Notes:

Success stories: According to the actual test dataset provided by a customer, the average packing rate of Huawei's packing service in full-container load (FCL) delivery is 92.74%, 5.84% higher than the customer's manual packing rate (86.90%). The average packing rate of Huawei's packing service in less than-container load (LCL) delivery is 85.78%, 2.86% higher than the customer's manual packing rate (82.92%), requiring 3.6% fewer containers.

<!-- Slide number: 16 -->

Intelligent Stand Allocation Solution

Based on AI algorithms and scenario understanding, the intelligent stand allocation system provides the optimal stand allocation solution powered by Huawei big data platform and ModelArts, improving the docking rate, passenger experience, and airport operation efficiency.

Allocation solution evaluation

Metric statistics

View display

Intelligent algorithm engine

Allocation solution generation

Rule engines

Intelligent stand allocation system

Cloud enablement application platform

Big data

Stand information library

Stand allocation library

Model information library

Flight information library

…

Airport system

Operation and command information platformAOMDP

Operation resource management system

Flight information management system

Air traffic control authorities

Notes:

Fewer ground conflicts

Minimized taxiing time

Less ground support time

Controlled docking rate

Visible allocation, simple operation interface, and easy management and maintenance

Stable and reliable system

Core indicators guaranteed

Flexible and configurable allocation rules

Success stories: Shenzhen Airport

The docking rate, a key metric, is improved by 5%, and the conflict rate is reduced by 10% compared with manual allocation. For an airport with an annual passenger volume of 50 million, 2.5 million passengers do not need to take a ferry bus every year. The 10-second rolling speed improves the commander's work efficiency and airport operation efficiency.

<!-- Slide number: 17 -->

HUAWEI CLOUD EI Overview

EI Intelligent Twins

AI Services

Case Studies of HUAWEI CLOUD EI

Notes:

<!-- Slide number: 18 -->

Essential AI Platforms

Huawei HiLens

Multimodal AI development application platform featuring device-cloud synergy

Graph Engine Service (GES)

China's first commercial distributed native graph engine with independent intellectual property rights

ModelArts (AI development)

One-stop AI development platform

Notes:

<!-- Slide number: 19 -->

ModelArts

ModelArts is a one-stop AI development platform. It provides data preprocessing, semi-automated data labeling, distributed training, automated model building, and model deployment on the device, edge, and cloud to help AI developers build models quickly and manage the AI development lifecycle.

Cloud

AI Gallery


Model training

AI model

Batch

Device

Edge

HiLens

Edge server

Video analysis

IoT gateway

Mobile phone

Camera

Edge station

Notes:

ExeML: ExeML automates model design, parameter tuning and training, and model compression and deployment with the labeled data. The process is free of coding and does not require expertise in model development.

Device-edge-cloud: indicates devices, Huawei intelligent edge devices, and HUAWEI CLOUD, respectively.

Online inference: a web service that synchronously provides the inference result for each inference request.

Batch inference: a job that processes batch data for inference.

Ascend chips: a series of Huawei-designed AI chips with high computing performance but low power consumption.

The built-in AI data framework combines auto pre-labeling and hard example labeling to improve the data preparation efficiency by over 100 folds.

The Huawei-developed MoXing high-performance distributed framework uses core technologies such as hybrid parallel cascade, gradient compression, and convolution acceleration, greatly reducing the model training time.

Models can be deployed on devices, edges, and clouds in different scenarios with one click to meet the requirements of high concurrency and lightweight deployment.

ModelArts allows visualized management of the AI development lifecycle, including data preparation, training, modeling, and inference. It also supports resumed training, training result comparison, and model traceback.

AI Gallery supports data and model sharing, helping enterprises improve AI development efficiency and allowing developers to convert knowledge to value.

<!-- Slide number: 20 -->

ModelArts Functions

Online learning


Development and management: Notebook + Pycharm SDKs + Immersive development

Training management

Model management

Data processing

Algorithm development

| Deployment management | | |

| --- | --- | --- |

| | | |

| | | |

Data

Intelligent data filtering

Intelligent data labeling

Intelligent data analysis

Labeling scoring and evaluation

Data augmentation (data refinery)

Interactive intelligent labeling

AI apps

Model repository

Precision tracking

Model evaluation

Model conversion

Notebook 2.0

Jupyterlab

ML Studio

MindStudio

Code samples

MindSpore samples

ModelZoo samples

Built-in algorithms (industry algorithms)

Auto hyperparameter tuning

Pangu large model

Visualized workflow

Reinforcement learning

Federated learning

Elastic training

Cloud real-time service

Cloud batch service

Device-edge-cloud synergy

Data collection

Models

APIs

Engineering pipelines

Datasets

Model sales

API sales

AI Gallery

AI Gallery

Notes:

Data processing: Supports data filtering and labeling, and manages dataset versions, especially big datasets for deep learning, so that the training results can be reproduced.

Fast and simplified model training: Provides the Huawei-developed MoXing deep learning framework, enabling high-performance distributed training.

Model deployment across the cloud, edges, and devices: Supports one-click deployment of trained models as real-time and batch inference services to clouds, edges, and devices in a wide array of production environments.

ExeML: Enables model building without coding and supports a wide array of scenarios, including image classification, object detection, and predictive analytics.

Visualized workflow: Uses GES to manage and visualize the lifecycle of AI development workflows, implementing data and model source tracing.

AI Gallery: Supports frequently used models and datasets, and internal or public sharing of enterprise models in AI Gallery.

<!-- Slide number: 21 -->

ExeML

ExeML is an entry-level service provided by ModelArts for beginners. Powered by the ExeML engine, ExeML enables even beginners to build, train, verify, and publish models.


Zero coding

Step 2:

Train the model.

Step 1:

Upload data and label it.

Step 3:

Check and publish the model.

No AI experience required

<!-- Slide number: 22 -->

Data Management

A wide range of data formats

Intelligent data filtering

Automatic image clustering

Recognition rate of invalid images > 80%

Five types of data (image, audio, video, text, and table)

Custom data formats

Data management

Team labeling

Dataset

Automatic pre-labeling

Team labeling

Iterative semi-auto data pre-labeling

Labeling efficiency up by 5 times

Great for ultra-large-scale labeling

Team labeling

Data access

Intelligent pre-labeling

Dataset

Iterative intelligent labeling framework

Single labeling

Auto feature mining

In-depth data optimization suggestions in 30+ dimensions

General models for various scenarios

Adaptive to data and algorithm changes

Intelligent data filtering and auto pre-labeling

Notes:

<!-- Slide number: 23 -->

Training Platform

Flexible, high-efficiency and cost-effective

Training platform (machine learning/deep learning)

Multiple built-in algorithms, import of partner algorithms, custom training logic and images, and quick switchover between heterogeneous resources

Linearly improved parallel training capability, auto parameter tuning, and a wide array of development modes

Elastic training, economical mode, self-developed chips and software, and ultimate cost-effectiveness

Notebook/Workflow

Configurable training

Model management

Model training

Data management

Built-in training models, accelerating AI implementation

Algorithm development

Model management

100+ algorithms, including image, text, time series, and reinforcement learning algorithms

One-click training requiring only data source configuration

Auto parameter tuning

Model training

Model visualization

Dataset

Model evaluation

Multiple development modes, meeting different requirements

In-cloud development (Notebook+SDK)

On-premises development (IDE+PyCharm ToolKit)

Elastic training, improving training efficiency

Turbo mode: Resources are automatically expanded during off-peak hours, accelerating the training speed by 10 times (from 2 hours to 10 minutes).

Economic mode: Optimized scheduling reduces the training cost by 30%.

Notes:

<!-- Slide number: 24 -->

Inference Platform

Unified management

Unified management of models of different vendors, frameworks, and functions

High-concurrency model deployment, low-latency access, auto scaling, grayscale release, and rolling upgrade

Inference platform

Resource management

Model management

Model deployment

Online inference

Edge inference

Batch inference

Flexible deployment

Models can be deployed as real-time and batch inference services on the cloud, edge, and devices.

Deployment control and management

Model import management

Resource management

Real-time data

Result output

Model deployment

Data access

Online inference

Inference process

Model compression and conversion

Batch data

O&M and monitoring

Batch inference

Iterative model update

Model evaluation and diagnosis

Hard example mining, automatic identification of hard examples, and quick adaptation to data changes

Edge inference

(IEF)

Notes:

Unified management: Supports unified management of models of different vendors, frameworks, and functions, high-concurrency model deployment, low-latency access, auto scaling, grayscale release, and rolling upgrade.

Flexible deployment: Supports a rich array of deployment scenarios, including real-time inference services and batch inference jobs on clouds, edges, and devices.

Iterative model update: Supports auto hard example mining to quickly adapt to data changes.

Dynamic scaling: On-demand allocation of inference resources ensures high concurrency and low latency

Seamless upgrade: Grayscale release and rolling upgrade enable O&M and upgrade to be performed without affecting user experience.

Automatic discovery of hard examples: Significantly reduces the manual labeling workload and accelerates AI implementation.

End-to-end closed-loop iteration: Iterative model and data development ensures real-time adaptation to changes.

<!-- Slide number: 25 -->

AI Gallery

AI Gallery

Dataset

Model

Industry workflow

D chipset operator

Algorithm

Subscription

Release

ModelArts

Enterprise-class suite customized for AI

General AI platform

Optimization

Deployment

Training

Edge

Edge site

Cloud

Built-in algorithm

Cloud

Real-time service

Device

Batch service

Parallel training

On-demand subscription and quick deployment

Publishing at zero cost

No fees for sellers to publish a model because no resources are consumed

On-demand subscription, deployment, and charging for buyers

Quick service deployment and low-cost O&M because of a full set of deployment capabilities

Seller

Buyer

Development/Import

Model finetuning

Encrypted deployment

Portfolios of cloud models and algorithms for buyers to optimize models based on differentiated data

Confidentiality of container deployment protects intellectual property

Notes:

<!-- Slide number: 26 -->

ModelArts Pro

ModelArts Pro is a professional development suite for enterprise-class AI applications. AI development is simplified with the advanced algorithms, quick training, and built-in workflows and models provided by HUAWEI CLOUD. In addition, customers can quickly develop, share, and launch applications through custom workflow orchestration. With these, HUAWEI CLOUD aims to create an open AI ecosystem for Inclusive AI. The ModelArts Pro suite consists of NLP Suite, OCR Suite, and Visual Suite, which allows it to meet AI implementation requirements in different scenarios.

ModelArts Pro

| Application asset management Models, services, and templates | Application iteration Models, services, and templates | Application implementation Online deployment, HiLens deployment, and appliances |

| --- | --- | --- |

Industry workflow package

Applications

Step-by-step workflow execution -> Application generation

Creating industry applications using workflows

| Custom AI components AI algorithm, data processing, and model evaluation | Pre-installed Huawei AI components Data processing, domain model, and deployment | Workflow sharing Shared with specified users and published to AI Gallery |

| --- | --- | --- |

ModelArts

Workflow

Workflow

Workflow orchestration and commissioning

Accumulating Huawei AI atom capabilities