Optimization with impact: my journey in public sector operations research

Optimization with impact:
my journey in public sector
operations research
Professor Laura Albert
Industrial & Systems Engineering
University of Wisconsin-Madison
laura@engr.wisc.edu
punkrockOR.com
@lauraalbertphd

Buena noches 
Es un placer hablar con todos ustedes hoy.
22 September 2020 Laura Albert 2

A brief bio
Professor of Industrial &
Systems Engineering
University of Wisconsin-
Madison.
From Chicago
Punk Rock Operations
Research blogger.
Former INFORMS VP of
Marketing, Communication,
and Outreach (2016 – 19)

Advances in operations research:
My research journey
Public safety:
Fire and emergency
medical services
Aviation security
policy
Infrastructure
protection /
resilience
Policing and illicit
opioid network
interdiction
Cybersecurity

Ingredients for a fulfilling academic
career in applied optimization
A good
problem
Collaborative
partners & real
data
Courage
Real world
impact!

Impact Achievement Highlights
• National Association of Counties
Achievement Award Winner, 2010
• Interfaces article
• INFORMS Impact Prize 2018
• Ragnar E. Onstad Service to Society
Award, College of Engineering, University
of Wisconsin-Madison, 2019.
• Boards and Commissions
• Wisconsin Emergency Medical Service Board
• Middleton Police Commission

Optimizers want to allocate scarce resources
to design and operate systems.
A system is a set of things—people, ambulances, computer
networks, vehicles, or whatever—interconnected in such a way
that they produce their own pattern of behavior over time.
Systems span people, resources, and processes.
Decisions are interconnected and cannot be made in isolation

Ingredient #1: good problems
A good
problem
Collaborative
partners & real
data
Courage
Real world
impact!

Aviation security

Passenger screening in the US
• 1996
• Checked baggage for high-
risk/selectee passengers
screened for explosives (run
by airlines)
• Goal: use limited
baggage screening devices

Early checked baggage security models
Baggage screening performance measures developed in
conjunction with the Federal Aviation Administration:
Policy implications?
Cover the most
flights
Cover the most
passengers on
covered flights
Cover flights by
fully utilizing the
devices

Passenger screening in the US
• November 2001 – Aviation
Transportation & Security Act
• Required all checked baggage
to be screened for explosives
• December 2001
• Remove shoes
• August 2006
• Liquids bans
• 2009 - 2010
• Explosive trace portals
• September 2012
• Less screening for seniors
(75+) and children (<12)
• December 2013
• TSA PreCheck for reduced
security

Moving beyond baggage screening:
How to screen passengers and baggage with
the right mix of screening equipment?

Risk-based Screening Framework
• How do match limited screening resources to passengers?
• Know everyone’s risk before they enter security screening;
allocate security resources to match risk.
• Risk-based security: Captured in the Dynamic Aviation Risk
Management System (DARMS) paradigm.
Assumptions:
• Most passengers are low-risk.
• Security resources are limited.
• Screening procedures make errors
(False alarms, False clears)

How can passengers be assigned?
Ahead of time (static models):
Integer programming models
Select the security classes to
use using Integer Programming
In real-time (dynamic models
for allocation decisions):
Markov Decision Processes &
Control Theory models

Key policy insights
(1) Better security is achieved by targeting scarce screening
resources at the “riskiest” passengers and doing less screening
on most passengers.
(2) Risk based screening more effective than random or uniform
screening in a resource-constrained environment.
(3) TSA Precheck implicitly does this,
which is why it reduces risk in low risk,
cost-constrained environments.

INFORMS Impact Prize (2018)
“For their pivotal role in the creation and widespread
adoption of risk-based aviation security strategies”

Ingredient #2: collaborative partners and
real data
A good
problem
Collaborative
partners & real
data
Courage
Real world
impact!

Fire and Emergency Medical Services

Collaborative partners

Anatomy of a 911 call
(110 call in Guatemala?)
Goal: faster response times
Service provider:
Emergency Call placed
Vehicle(s)
dispatched
Vehicle(s) en
route
Vehicle(s)
arrive at
scene
Service/care
provided
Vehicle(s)
leave scene
Vehicle(s)
arrive at
hospital
Patient
transferred
21
Response time from the patient’s point of view
Ambulance unavailable for other patients
Ambulance planning must consider tradeoffs across time and space
22 September 2020 Laura Albert
Call center:
Triage

Research questions
Which response time goals are best for patient health outcomes?
Lossfunction

Research questions
Which response time goals are best for patient health outcomes?
How to dispatch ambulances when triage is imperfect?
How to dispatch vehicles in equitable ways?
How can we consider location and dispatch decisions together?
How many ambulances should we keep in reserve for high-
priority patients when there is congestion?

More research questions
How to locate and dispatch ambulances in tiered systems?
When should we send multiple types of ambulances?
How to locate and dispatch non-transport vehicles?

Let’s illustrate some of the ideas with an
example using emojis
Research methods used:
• Markov decision processes
• Integer programming
• Stochastic programming
• Spatial queuing models

How to route vehicles to patients?

Better ways to route vehicles to patients?

Multiple response is sometimes best

It is sometimes best to send non-transport
vehicles

Routing multiple fire engines/ladders to calls

Ingredient #3: courage
A good
problem
Collaborative
partners & real
data
Courage Real world
impact!

To protect cyber-infrastructure, we must
protect their supply chains
“The operation of critical networks and information infrastructures
depends on the assured availability of trustworthy hardware and
software. Vulnerabilities in the supply chain can enable attacks
on the integrity, availability, or confidentiality of networks and the
data they contain.”
President Obama’s Strategy for Cyberspace, 2011
Enayaty-Ahangar, F., Albert, L.A., DuBois, E. 2020. A survey of optimization models and
methods for cyberinfrastructure security. To appear in IISE Transactions.
https://doi.org/10.1080/24725854.2020.178130622 September 2020 Laura Albert 33

Protecting critical information technology (IT)
infrastructure
• IT infrastructure relies on a globalized supply chain that is
vulnerable to numerous risks.
• Goal: reduce risk to critical infrastructure by identifying a mix of
security mitigations that enhance supply chain security.
Laura Albert
Supply chain
layer
Physical
infrastructure
layer
Designer
SupplierSupplier
Manufacturer Assembly
Distribution
Steal@A
Change@C Inject@
DESteal@B
Entry
Insertion
Points
Attack layer
Mitigations
Steal@B
Entry
Insertion
Points
22 September 2020 34

Attack paths
1 2 6
1 2 7
1 3 9 14 15 16
1 3 10 17
1 3 10 18
1 4
1 5 13
1 5 18
1 5 8 11 12 18
1 5 8 11 12 19
1 5 8 11 12 20
Laura Albert22 September 2020 35

Attack paths with mitigations
1 2 6
1 2 7
1 3 9 14 15 16
1 3 10 17
1 3 10 18
1 4
1 5 13
1 5 18
1 5 8 11 12 18
1 5 8 11 12 19
1 5 8 11 12 20

Models with adaptive adversaries
Attack paths
Expected coverage models
(Deterministic, stochastic, worst-case)
Attack graphs
Project management / interdiction
With adaptive adversaries

Stackelberg game approach
• One shot game representing cyber-security planning
• Defender/Leader acts first by interdicting
components of the network (i.e., selects a set of
mitigations that maximally delay the weighted
completion of individual exploits/arcs)
• Attackers/Follower(s) acts second by performing
recourse (i.e., completing project in response to the
mitigations as quickly as possible through project
management / critical path).
Attack graph: graph representation of an attack project
• Nodes: intermediate project objectives
• Arcs: attack exploits
• Longest path: represents the attacker's critical path

Example
1
2
3
Attacker 1 Attacker 2
5
2
2
4
5
7
2
3
1
6
2
2
Arc lengths
Completion time = 5 Completion time = 6 Total time = 11

Example
1
2
3
5
2
2
4
5
7
2
3
1
6
2
2
Arc lengths with critical paths shown in boldface

Example
1
2
3
5 + 2 (𝑚𝑚3)
2 + 2 (𝑚𝑚2)
2 + 2 (𝑚𝑚1)
4
5
7
2 + 1 (𝑚𝑚3)
3 + 2 (𝑚𝑚2)
1 + 2 (𝑚𝑚1, 𝑚𝑚2)
6
2
2
Arc lengths with delay times caused by three mitigations 𝑚𝑚1, 𝑚𝑚2, 𝑚𝑚3

Example
1
2
3
5
2
2 + 2
4
5
7
2
3
1 + 2
6
2
2
Arc lengths when mitigation 𝑚𝑚1 selected:
The critical path in attacker 1’s graph changes. The critical path in attacker 2’s
graph is unchanged and the total project is not lengthened

Example
1
2
3
5
2 + 2
2
4
5
7
2
3 + 2
1 + 2
6
2
2
Completion time = 6 Completion time = 7
The critical paths in both graphs change and both completion times are longer.
Total time = 13

Example
1
2
3
5 + 2
2
2
4
5
7
2 + 1
3
1
6
2
2
Completion time = 7 Completion time = 7
The critical paths in both graphs are the same and both completion times are
longer.
Total time = 14

Example
1
2
3
5
2 + 2
2 + 2
4
5
7
2
3 + 2
1 + 2
6
2
2
Arc lengths when mitigations 𝑚𝑚1 and 𝑚𝑚2 selected:
The critical paths in both graphs change and both completion times are longer.22 September 2020 Laura Albert 45

The resulting bi-level stochastic interdiction
problem is hard to solve

Real world impact
A good
problem
Collaborative
partners & real
data
Courage
Real
world
impact!

Operations research with impact
Good
research
Media
outreach
Industrial
practices
Legisla-
tion
Influence
policy

My research has had impact
Adoption of the use of
non-transport vehicles
In software to route fire
engines
Wisconsin EMS Board
TSA PreCheck
Cyber-security
consortium to
develop industrial
standards
Expanding diversion programs
Middleton Police Commissioner

INFORMS Advocacy Efforts
2019 Government & Analytics Summit

Media
outreach

Our research and stORies sometimes
capture the attention of the general public
• Small window of opportunity to disseminate the real-world
impact of our research
• But we must package our messages for public consumption
• Viewers/readers want to learn something new.
What will they learn from us?

The general public’s reaction to seeing math,
modeling, and analytics in the news

Applied operations research should not
be a well kept secret

Final messages
A good
problem
Collabora-
tive partners Real data Courage
Real world
impact!

I’ve talked about research impact in
different applications

The COVID-19 pandemic has affected all
of these application areas
Public safety:
Fire and emergency
medical services
Aviation security
policy
Infrastructure
protection /
resilience
Policing and illicit
opioid network
interdiction
Cybersecurity

Making a difference with operations
research and analytics
OR and analytics can lead to tools.
OR and analytics can introduce ideas.
OR and analytics can influence policy.
OR and analytics are optimistic.

Questions?
Thank you!
Laura Albert
University of Wisconsin-Madison
laura@engr.wisc.edu
punkrockOR.com
bracketology.engr.wisc.edu
@lauraalbertphd

References
Cyber-security
Zheng, K., and Albert, L.A. 2019. Interdiction models for delaying
adversarial attacks against critical information technology
infrastructure. Naval Research Logistics 66(5), 411 – 429.
Enayaty-Ahangar, F., Albert, L.A., DuBois, E. 2020. A survey of
optimization models and methods for cyberinfrastructure security. To
appear in IISE Transactions.
https://doi.org/10.1080/24725854.2020.1781306
Aviation security
McLay, L. A., S. H. Jacobson, and J. E. Kobza, 2006. A Multilevel
Passenger Prescreening Problem for Aviation Security, Naval
Research Logistics 53 (3), 183 – 197.
Lee, A.J., L.A. McLay, and S.H. Jacobson, 2009. Designing Aviation
Security Passenger Screening Systems using Nonlinear Control.
SIAM Journal on Control and Optimization 48(4), 2085 – 2105.
McLay, L. A., S. H. Jacobson, and A. G. Nikolaev, 2009. A
Sequential Stochastic Passenger Screening Problem for Aviation
Security, IIE Transactions 41(6), 575 – 591.
McLay, L.A., S.H. Jacobson, A.J. Lee, 2010. Risk-Based Policies for
Aviation Security Checkpoint Screening. Transportation Science
44(3), 333-349.
Albert, L.A., Nikolaev, A., Lee, A.J., Fletcher, K., and Jacobson,
S.H., 2020. A Review of Risk-Based Security and Its Impact on TSA
PreCheck, To appear in IISE Transactions.
Media Engagement
L. A. Albert. 2020. Engaging the media: Telling our operations
research stories to the public. SN Operations Research Forum 1
(14) https://doi.org/10.1007/s43069-020-00017-0
Emergency Medical Services
McLay, L.A., 2009. A Maximum Expected Covering Location Model
with Two Types of Servers, IIE Transactions 41(8), 730 – 741.
McLay, L.A. and M.E. Mayorga, 2010. Evaluating Emergency
Medical Service Performance Measures. Health Care Management
Science 13(2), 124 – 136.
Ansari, S., McLay, L.A., Mayorga, M.E., 2015. A maximum expected
covering problem for locating and dispatching servers. To appear in
Transportation Science.
McLay, L.A., Moore, H. 2012. Hanover County Improves Its
Response to Emergency Medical 911 Calls. Interfaces 42(4), 380-
394.
Ansari, S., Yoon, S., Albert, L. A., 2017. An approximate Hypercube
model for public service systems with co-located servers and
multiple response. Transportation Research Part E: Logistics and
Transportation Review. 103, 143 – 157
Yoon, S., and Albert, L.A. 2020. A dynamic ambulance routing
model with multiple response. To appear in Transportation
Research Part E: Logistics at Transportation Science.
Yoon, S., Albert, L.A., and V.M. White 2020. A Scenario-Based
Ambulance Location Model for Emergency Response with Two
Types of Vehicles. To appear in Transportation Science.

Optimization with impact: my journey in public sector operations research

More Related Content

Similar to Optimization with impact: my journey in public sector operations research (20)

More from Laura Albert (20)

Recently uploaded (20)

Optimization with impact: my journey in public sector operations research