Transcript ppt - Samer Hassan

Stepping on Earth:
A Roadmap for Data-driven
Agent-Based Modelling
Samer Hassan
University of Surrey
Luis Antunes
Universidad Complutense de Madrid
Universidade de Lisboa
Juan Pavón
Nigel Gilbert
Contents

The Classical View

Breaking the Rules

Dealing with Data

Data-driven Flow

Difficulties
Samer Hassan
ESSA 2008
2
Classical Logic of Simulation
Samer Hassan
ESSA 2008
3
Classical Paradigm

Axelrod’s KISS


Occam’s razor
Simplicity is helpful:





Samer Hassan
Transmitting the model
Promoting understanding
Promoting extensibility
Abstract models -> more general?
Easier to design, analyse and check
ESSA 2008
4
Contents

The Classical View

Breaking the Rules

Dealing with Data

Data-driven Flow

Difficulties
Samer Hassan
ESSA 2008
5
Data-driven Modelling

Use of empirical data


Attempt to be more realistic


Getting closer to the target
Increasing complexity of the models


Instead of standard distributions
Against the KISS paradigm
Strongly linked to the intense use of the
data available
Samer Hassan
ESSA 2008
6
A Well-known Alternative

Edmonds & Moss’ KIDS:
1.
Design the most similar model to the target
2.
Analyse the model to see which parts could
be simplified while preserving the behaviour
3.
With further simplifications, it could be used
in several contexts with assured good
foundations
Samer Hassan
ESSA 2008
7
A New Perspective

According to KISS, complexity should only
increases when difficulties are met

‘Deepening’ KISS:
1.
Begin with KISS
2.
Through the use of evidence and especially data, a
collection of models can be developed and explored
3.
With the design space (of agents, societies and
experiments) explored, the best features of each
model can be used to design a stronger model
Samer Hassan
ESSA 2008
8
Contents

The Classical View

Breaking the Rules

Handling Data

Data-driven Flow

Difficulties
Samer Hassan
ESSA 2008
9
Statistical Distributions

‘Let’s assume that salaries follow a
Gamma distribution’

Real data closer than standard
distributions

Distributions are static

Not necessary more general
Samer Hassan
ESSA 2008
10
Handling data

Representative sample from target
population? Surveys
1.
Survey’s questions not phrased in the right
way for the researcher’s interests
 Compromises
2.
Unlikely including interconnections and
interactions between sample members
 Networks?
3.
Samer Hassan
Inherently qualitative data?
ESSA 2008
11
Handling data

How to deal with dynamical processes?



Panel studies
Ethnography
Usual activities recording
• Official documents
• Internet records

Before extra efforts


Samer Hassan
Check what’s out there!
National Data Archives
ESSA 2008
12
Contents

The Classical View

Breaking the Rules

Dealing with Data

Data-driven Flow

Difficulties
Samer Hassan
ESSA 2008
13
Proposal for Data-Driven ABM
Samer Hassan
ESSA 2008
14
The Data-driven Flow
Samer Hassan
ESSA 2008
15
Technologies can help

Typical smooth behaviour ->Soft computing




Search for patterns and clusters in the input and output


Neural Networks for adaptive learning behaviours
Fuzzy Logic for modelling social processes
Evolutionary algorithms can optimise agent behaviour
Classifiers and Data Mining
Representation of concepts


Samer Hassan
Ontologies represent an easy-to-handle interface with
experts, and a formal view that can be inserted in the ABM
Natural Language Processing for a better representation of
the simulation output
ESSA 2008
16
Contents

The Classical View

Breaking the Rules

Dealing with Data

Data-driven Flow

Difficulties
Samer Hassan
ESSA 2008
17
Difficulties

Sometimes, a KISS model is enough

Ease of understanding & communication of KISS is
partially lost


Modularity
‘Deepening’ stages lead to understanding

Demands special effort in gathering data

Addressed difficulties related to the procedures:


surveys not providing the required data; lack of information;
qualitative or subjective data
Technologies should only be used in their certain context
Samer Hassan
ESSA 2008
18
Thanks for your attention!
Samer Hassan
[email protected]
University of Surrey
Universidad Complutense de Madrid
Universidade de Lisboa
Samer Hassan
ESSA 2008
19
Contents License

This presentation is licensed under a
Creative Commons Attribution 3.0
http://creativecommons.org/licenses/by/3.0/
Para ver esta película, debe
disponer de QuickTime™ y de
un descompresor TIFF (sin comprimir).

You are free to copy, modify and distribute it as long as
the original work and author are cited
Samer Hassan
ESSA 2008
20