Вы находитесь на странице: 1из 31

Advancing

Machine Learning and AI


with Geography and GIS

Robert Kircher
rkircher@esri.com
Welcome
&

Thanks
GIS is expected to do more, faster.

find where
see where predict where

locate, connect WHERE route where

map the unmapped

machines can be “trained” to do more for us


Topics Today

Emerging Geo AI Business Cases

Geo-enabling Machine Learning Landscape and Factory

Notable Geo AI challenges and limitations


Emerging Geo AI Business Cases

Accidents, Anomaly Predictive Asset Predictive Predictive Advanced Feature


Prediction Allocation Routing, ETA, GeoMarketing Extraction
Traffic

Police
Autos

Landcover
Landuse Crops
Classification

Object Detection
Risk
Map the Unmapped Protection
Management

Augmented Reality
Pollution (AR)
Exposure
Industry Advancing Fast (with Geo)

Focused Computing
GPUs, Devices

Spatial Statistics
SaaS, Cloud-based

Organizations

Platforms, On-premises
"Gives computers the ability to learn without being explicitly programmed"
AI > ML > DL
Artificial Intelligence Machine Learning Deep Learning

Supervised Learning Deep Supervised Learning


1. Training features Labels
Reasoning Knowledge Perception
Representation

NLP Robotics Machine Learning

2. Predicting

Unsupervised Learning Dog


Reinforcement Learning
ML Algorithms
Data + ML Models Predictions
(inferences)
AI
AR
(algo’s)
(statistics) decisions
understanding
insights
classify
ML changes things in exciting ways for GIS

evidence based models … faster … automated … data driven … accurate … tunable … platforms … pervasive
Machine Learning Capabilities

Today Emerging

ArcGIS
Geo ML:
Spatial Analytics, Geostatistics,
Spatial Statistics, Surfaces
Geo-enabled ML or Geo-centric ML
ML Landscape, Factory and Pattern

Data Wrangling/Preparing Model Training/Tuning Share/Deploy Model(s) Use Model(s)

Classification Predict, Infer


5
Data Discovery/Create 4
Augmented Reality
• imagery Clustering 3
(AR)
• vector 2

• business 1
0
1 2 3 4 Computer Vision
Regression
Exploratory Data Analysis regression
(EDA)
Features/Labels training Deep Learning Visualize, Map It
(neural networks)

Data Gathering/”Wrangling”
Recommend

Data Preparation ML algorithms Find/Detect Features


neural networks

data sources Model Validation


apps, maps, services, devices
Geo ML Landscape, Factory and Pattern … geo-enabling and advancing
Data Wrangling/Preparing Model Training/Tuning Share/Deploy Model(s) Use Model(s)

Classification Predict, Infer


Spatial Statistics 5
Vast
DataCollections, Expertise
Discovery/Create Spatial Analysis 4
Geo Data Sources Augmented Reality
• imagery Clustering
GP Tools 3
(AR)
•Structured/Unstructured
vector Geo-certify 2
GeoAnalytics
• business Server 1
0
1 2 3 4 Computer Vision
Regression
Maps, Visualization
Exploratory Data Analysis regression Integrated GIS Platform
Jupyter/Python/R
(EDA) Mapping Platform
training Deep Learning Visualize, Map It
(neural networks)
Data Gathering/”Wrangling”
Geo-centric Datasets
Geo-referenced Datasets Recommend

Localized Datasets
Geo
Data CLT
Preparation
Geocoding ML algorithms Find/Detect Features
neural networks

data sources Model Validation


apps, maps, services, devices
Demo: CAFO Detection
Using TensorFlow CNNs to Detect CAFO sites from Satellite
Imagery + Consuming the model from ArcGIS Pro

Demo Code:
https://github.com/Qberto/ML_ObjectDetection_CAFO
Please notice process details
• Geo Data Preparation
• Model Training Duration
• Integrating Model Across GIS Platform
• Augmented Reality (AR)
• Recent Demo at Esri (ignore some references)

… imagine your business problem, your data, your model


Some
Challenges
&

Limitations
It’s complex. Let’s simplify it

technology, automation, platforms, computing, etc.


Look past hype … it’s distracting

prevent another “Machine Learning Winter”


Pred = W A + W A + … + Wloc(Location)
1 1 2 2

not fully integrated … yet


Behavior1
Feature1

Traditional Model (Geo-less)

Location
Limited expertise among
Geo Community in training
and tuning models …
math, statistics, spatial
statistics

Introduce Geo Data Science into industry


Limited exposure to Geo in
academia … Data Science,
Computer Science
Proving Model
Trustworthiness,
Authoritativeness
(Geo) Data Bias
Next Steps
Machine Learning needs human analysis to succeed.

GIS Community, People, Professionals, Academia.


Advancing Geo AI

Build on Emerging Business Cases and Geo-enabled ML


Patterns

Find your place in ML landscape and factory

Together, let’s continue to overcome Geo AI challenges


and limitations

Оценить