Transcript No Slide Title

What is Agent-Based
Modeling?
Virginia A. Folcik Nivar Ph.D.
“Essentially, all models are wrong,
but some are useful.”
“Remember that all models are wrong; the practical
question is how wrong do they have to be not to be useful.”
George Box, Statistician and Prof. Emeritus,
Univ. of Wisconsin-Madison,
Industrial and Systems Engineering
Model VS. Simulation
MODEL
•less detail
•generalizable
•involves learning
process
•an art form
•answers “what if” questions
SIMULATION
•more detail
•not generalizable to
all situations
•answers specific questions
What can be modeled?
Sub-atomic particles
Atoms
Molecules
SYSTEMS BIOLOGY Macromolecules (DNA, RNA, Proteins)
Organelles
Cells (Organisms)
Tissues
Organs
ALL CAN BE STUDIED
Organ systems (Immune System)
WITH
Organisms
MATHEMATICAL
Populations [Societies, Economies,
OR
Ecologies (90%), Businesses]
AGENT-BASED MODELING
{
“Swarm Intelligence”
•Social insects are extremely successful creatures
•They maintain their existence following simple rules
based on local information
•They self-organize, have no central control
•They exhibit flexibility (adjust task allocation)
•Colonies are robust
•They use indirect communication
Ex. Southwest Airlines Cargo Operations
Bonabeau, Dorigo and Theraulaz, 1999
Complex Adaptive Systems
•Large collections of autonomous members
•Members react to their local environment according
to a set of internal rules
•No centralized control
•Exhibit emergent behavior
•Exhibit stigmergy
Examples: insect colonies,
societies, ecologies, economies,
biologies, business firms, cities,
schools of fish, flocking birds, drivers
in traffic, terrorist networks…
Emergence
Complex systems arise from the simple behavior
of the individuals that constitute them.
The whole is greater than the sum of the parts.
Examples of emergent phenomena:
•Braess’s Paradox (1968) Adding more lanes to a highway often
makes traffic jams worse.
•Employee bonuses result in reduced productivity.
•An increase in the number of shoppers in a supermarket decreases
sales of certain products.
•A deficiency in innate immunity (dendritic cells) results in the
increased incidence of hypersensitivity reactions.
Stigmergy
The indirect communications that take place between
individuals in a complex system.
(French entomologist Pierre-Paul Grassé in 1950’s)
•Cytokines and chemokines
•Pheromones
Ex. France Telecom, British Telecom, and MCI Worldcom,
phone call routing.
Agent-Based Modeling
a.k.a. Individual-Based Modeling, Bottom-Up
Modeling or Pattern-Oriented Modeling
Perceptions
Rules
Agent
Actions
Environment
State Diagrams
A
Start
State
B
C
Final
State
In the immune simulation, contact with cells or signals
(or their absence) trigger the transitions from one state to the next.
State Diagram: T cells
Zone 2
Activated T1*
Inactive T0
State 0
DC1
†
†
State 3
1 T1
CK1
DURATION_CK1_Zone 2
Initialization
DURATION_CK2_Zone 2
*Moves randomly seeking
No MK1 or CK1
CK2
NumT1_ToSend
2 T1**
progeny
to Zone 2
2 T2**
State 9
Apoptosis
NumT2_ToSend
an Ag-matched B1 or B2.
1 T2
Contact with B adds time to life.
DC2 †
**First Ag- and type-matched DC
State 2
State 4
contact with a T-cell causes proliferNo MK1 or CK1
†
ation of T’s into Zone 2.
Activated
†DC contact extends life.
Activated T2*
Memory T2*
‡Time is up, apoptosis.
All T cells moving in Zone 2
State 1
Memory T1*
Agent-Based Modeling:
Currently the best tool for studying
behavior within a complex system
•ABM’s capture emergent phenomena.
•Provide a natural (logical) description of a complex system. Experts
in a field can relate to the model.
•Agent-based modeling is flexible.
•Explain phenomena by “growing” them.
•Abstraction: the process of removing detail from a representation.
Choosing the appropriate level of abstraction
Grimm et al. (2005) Science 310:987-991.
How do you create an ABM?
1. Gather relevant information
about interactive entities (agents).
3. Implement theories in a
computer simulation.
2. Formulate theories about
agent behavior.
4. Observe the behavior of the system,
looking for emergent behavior patterns.
How do you know
if it works?
Verification: To ensure that the program
is doing what you intend it to do.
Ex. State diagrams
Validation: To ensure that the model
emulates the intended behavior.
Ex. Apply a perturbation with a known
consequence in the “real world”.
Some Universities Teaching Agent-Based Modeling
Argonne National Laboratories
Univ. of Mich., Center for the Study of Complex
Systems
RePast
Swarm
Humbolt State Univ. and Lang, Railsback and Assoc.
“Pattern-Oriented Modeling”
Swarm
Northwestern Univ.
Santa Fe Institute, Computer Sciences
Univ. of Notre Dame, Interdisciplinary Center for the
Study of Biocomplexity
Univ. of Torino, Italy
NetLogo
complexity
complexity
Univ. of Washington
complexity
Univ. of California, Berkeley
Auburn Univ.
complexity
complexity
complexity
The End