Top Banner
43

AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Jan 24, 2020

Download

Documents

dariahiddleston
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence
Page 2: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

AI for Earth:

AI for Protecting Wildlife, Forests, Fish

MILIND TAMBE

Founding Co-director, Center for Artificial Intelligence in Society (CAIS)

University of Southern California

[email protected]

Co-Founder, Avata Intelligence

Page 3: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

USC Center for Artificial Intelligence in Society (CAIS.USC.EDU)

7/18/2017 3

Page 4: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Mission Statement: Advancing AI research driven by…

Grand Challenges of Social Work

Ensure healthy development for all youth

Close the health gap

Stop family violence

Advance long and productive lives

End homelessness

Achieve equal opportunity and justice

7/18/2017 4

Page 7: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Overview of CAIS Project Areas

AI for Low Resource Communities

Social networks: Spread HIV information, influence maximization

Real-world pilot tests: Big improvements

7/18/2017 7

Page 8: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Partnerships

7/18/2017 8

Low Resource Communities

Public Safety and Security

Wildlife Conservation

Page 9: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 9

What Might We Lose?

Murchison Falls National Park, Uganda

Page 10: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Protecting Wildlife in Uganda

7/18/2017 10

Page 11: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 11

PAWS: Protection Assistant for Wildlife Security

Massive forests (1000 sq miles) to protect, limited security resources:

Generate “intelligently” randomized patrols

Learn adversary models

Patrol boat in Bangladesh

at Global Tiger Conference, 2014

Patrol with Rangers, Indonesia

Trip with WWF, 2015

Fang

Page 12: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 12

ARMOR: Assigning Limited Security Resources [2007]

AI for Public Safety and Security

2007

Airports

Canine patrol at

LAX (ARMOR)

PitaJain

Page 13: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Game Theory

2007

Airports

-1, 1 0, 0 1, -1

1, -1 -1, 1 0, 0

Player B

Player A

Paper Rock Scissors

Paper

Rock

Scissors

0, 0 1,-1 -1,1

AI-based DECISION AIDS TO ASSIST IN SECURITY

Page 14: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 14

Terminal #1 Terminal #2

Terminal #1 4, -3 -1, 1

Terminal #2 -5, 5 2, -1

Adversary

Model: Stackelberg Security Games

Defender

Set of targets, payoffs based on targets covered or not…

Challenges faced: Massive scale games; difficult for a human planner

Kiekintveld

Stackelberg: Defender commits to randomized strategy, adversary responds

Page 15: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 15

AI-based DECISION AIDS TO ASSIST IN SECURITY

Page 16: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 16

Security Game Deployments

Security Games

2007 2011

Ports

2009

Airports Air Marshals

Fang

Page 17: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 17

PROTECT: Ferry Protection Deployed

Page 18: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 18

Global Applications of Security using Game Theory

Page 19: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

19/

• Game theory vs Previous Method

Ticketless Travelers Caught

Arrests at LAX checkpoints

0

5

10

15

20

#Captures per30 min

#Warnings per30 min

#Violations per30 min

Game Theory

Previous Method

0102030405060708090

100

Miscellaneous

Drugs

Firearm Violations

SOME RESULTS OF GAME THEORY for SECURITY

Page 20: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 20

Learn from crime data

Game Theory calculate

randomized patrols

Patrollers execute patrols

Poachers attack targets

PAWS: Applying AI for protecting wildlife

Game Theory +Fang

Poacher Behavior Prediction

Page 21: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 21

Uganda Andrew Lemieux PantheraMalaysia

PAWS Patrols in the Field

Early Trials in Uganda and Malaysia

Important Lesson: Geography!

Fang

Page 22: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 22

PAWS: Protection Assistant for Wildlife Security [2016]

Game Theory + Poacher Behavior Prediction + Forest Street Map

Page 23: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 23

PAWS: Preliminary Evaluation

0.57

0.86

Human Activity Sign/km

Previous Patrol PAWS Patrol

Fang

Page 24: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 24

PAWS: Protection Assistant for Wildlife Security

Massive forests (1000 sq miles) to protect, limited security resources:

Generate “intelligently” randomized patrols

Learn adversary models

Patrol boat in Bangladesh

at Global Tiger Conference, 2014

Patrol with Rangers, Indonesia

Trip with WWF, 2015

Page 25: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 25

Learn from crime data

Game Theory calculate

randomized patrols

Patrollers execute patrols

Poachers attack targets

Predicting Poaching from Past Crime Data

PAWS: Applying AI for protecting wildlife

Game Theory + Poacher Behavior Prediction

Nguyen

Page 26: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Predicting Poaching from Past Crime Data

PAWS: Applying AI for protecting wildlife

Poacher Behavior Prediction

7/18/2017 26

Nguyen

Page 27: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Data from Queen Elizabeth National Park, Uganda

Poacher behavior prediction

Number of poaching attacks over 12 years: ~1000

How likely is an

attack on

a grid Square

Ranger patrol

frequency

Animal density

Distance to

rivers / roads

Area habitat

Area slope

7/18/2017 27

Nguyen

Page 28: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Initial Attempt Behavioral Game Theory Models:Dynamic Bayes Net

Attacking probability

Detection probability

Ranger observation

Ranger patrol

Animal density

Distance to rivers /

roads / villages

Area habitat

Area slope

M

j

jxSEU

jxSEU

j adversary

adversary

e

eq

1'

)',(

),(Adversary’s

probability of

choosing target j

Lack of sufficient data over the entire park

7/18/2017 28

Nguyen

Page 29: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Poacher Behavior Prediction

Poacher Behavior Prediction

Classifier 1 Classifier 2 Classifier 3

0 1 1

Aggregation Rule

1

Majority

1

Ensemble of Decision Trees (with feature boosting)

7/18/2017 29

Kar Ford

Page 30: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 30

Boost in “heavily monitored” regions of the park:

Improve accuracy

Learn local poachers’ behavior; distinct parameters

Boost Decision Tree Ensembles with

with Graphical Models

Classifier 1 Classifier 2 Classifier 3

0 1 1

Aggregation Rule

1

Majority

1

Decision Tree

Decision Tree

+ Graphical modelGraphical model

NguyenGholami

Page 31: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

L&L

Uniform Random SVM CAPTURE Decision Tree Our Best Model

Poacher Behavior Prediction

Poacher Attack Prediction

Results from 2015

7/18/2017 31

Page 32: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Real-world Deployment (1 month)

Two 9-sq. km patrol areas

Where there were infrequent patrols

Where no previous hot spots

7/18/2017 32

Kar Ford

Page 33: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 33

Real-world Deployment: (1 month)

Page 34: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Real-world Deployment: Results

Two 9 sq KM patrol areas: Predicted hot spots with infrequent patrols

Trespassing: 19 signs of litter etc.

Snaring: 1 active snare

Poached Animals: Poached elephant

Snaring: 1 elephant snare roll

Snaring: 10 Antelope snares

Hit rates (per month)

Ours outperforms 91% of months

Historical Base Hit

RateOur Hit Rate

Average: 0.73 3

7/18/2017 34

Page 35: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 35

Real-world Deployment: Field Test 2 (6 months)

2 experiment groups (27 areas of 9 sq KM each)

1:HIGH: 5 Areas

2: LOW: 22 Areas

• Rangers trained; unaware of this classification

FordGholami

Page 36: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 36

0

0.02

0.04

0.06

0.08

0.1

0.12

High (1) Low (2)

Cat

ch p

er

Un

it E

ffo

rt

Experiment Group

Real-world Deployment: Field Test 2 (6 months)

Catch Per Unit Effort (CPUE)

Unit Effort = km walked

Our high CPUE: 0.11

Our low CPUE: 0.01

Historical CPUE: 0.04

Page 37: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 37

Field Test Side Effects:Queen Elizabeth National Park

Rangers followed poachers’ trail; ambushed camp

Arrested one (of 7) poachers

Confiscated 10 wire snares, cooking pot, hippo

meat, timber harvesting tools.

Pursuit of poachers

Signs of road building, fires, illegal fishing

Page 38: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Next Steps

Deployments in other parks

Murchison Falls National Park, Uganda (WCS); Cambodia (WWF)

Inclusion for general deployment worldwide

7/18/2017 38

Page 39: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

7/18/2017 39

Green Security Games:

Patrolling From the Sky

Credit: Arvind Iyer, AirShepherd

Bondi

Page 40: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Three Poachers Hiding

Page 41: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

Towards the Future: AI for Earth

Significant potential: AI for Earth

Not just applications; novel research challenges:

Fundamental computational challenges from use-inspired research

Designing AI for Earth:

• Interpretability

• Complementing human autonomy

Methodological challenges:

Encourage interdisciplinary research: measures impact in real world

7/18/2017 41

Page 42: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

AI for Social Good

7/18/2017 42

Page 43: AI for Earth: AI for Protecting Wildlife, Forests, Fish · AI for Earth: AI for Protecting Wildlife, Forests, Fish MILIND TAMBE Founding Co-director, Center for Artificial Intelligence

THANK YOU

CAIS.USC.EDU