Download Ch12GIA - University of Denver

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Personal knowledge base wikipedia , lookup

Knowledge representation and reasoning wikipedia , lookup

Agent-based model in biology wikipedia , lookup

Ecological interface design wikipedia , lookup

History of artificial intelligence wikipedia , lookup

Embodied cognitive science wikipedia , lookup

Agent-based model wikipedia , lookup

Mathematical model wikipedia , lookup

Neural modeling fields wikipedia , lookup

Cognitive model wikipedia , lookup

Time series wikipedia , lookup

Transcript
New Approaches to Spatial
Analysis—Chapter 12
•
Theoretical chapter— biological and ecological
•
Recent Changes in GIS-- Technical and Theoretical
•
Geocomputation-– developing field
processes and models basis for spatial models
Spatial Modeling Developments— recent advances
and linking of models to GIS
•
Chapter points:
•
•
•
•
•
Describe computer power’s impact on GIS
field.
Outline briefly the implications of
complexity with this
Describe emerging geographic analysis
techniques
Describe cellular automation and agent
based models—applications
Outline ways of coupling spatial models to
GIS
Advent of GIS and Spatial Analysis
•
•
•
Spatial Analysis existed long before1911
Journal of GIS emerged in 1987
Contemporary GIS materialized
and became popular during the
information age, huge expansion of
computers, while Spatial Analysis
came about much earlier.
Complexity of Computers
•
•
•
New scientific non-linear view of the world
Complex systems analysis— biology and
thermodynamics
Limit to power of prediction in nonlinear
systems -- Why are weather forecasts always wrong—
especially in Denver?
•
Computers are critical to describe
relationships in complex analysis–- crucial to
development of theories of complexity.
Huge array of new computer tools
•
Automated tools—arrays of data—
discovery of new ways of analyzing
relationships and processes–methods
have no mathematical assumptions about underlying
causes of patterns—can be used for investigation of nonlinear phenomena.
•
Computer modeling and
Simulation– important in Geography and distinct
from statistical processes—represent world as its, actual
causal mechanisms
Geocomputation
•
Definition — ”the use of computers to tackle
•
Vague —
•
Stan Openshaw--
•
Leads into idea of Artificial Intelligence...
geographical problems that are too complex for manual
techniques”.
what would be a better definition?
Computational complexity?
Centre for computational Geography at the
University of Leeds. Asks “can we use cheap computer power in place
of brain power to help us discover patterns in geospatial data?
Geocomputation—
Artificial Intelligence
•
Artificial Intelligence-- attempt to endow a
•
First things first— need a intelligent
computer with some of the intelligence capabilities
of intelligent life forms… without imitating exactly
the same information processing steps of humans or
biological systems…
approach—humans versus computers: GAM
discussed in Chapter 5..
•
Adaptability and effective use of
information are key to a human investigation
approach.
•
AI being applied to Geographical
problems— discussed next….
Geocomputation– AI Applications
Expert Systems
Expert Systems–earliest approach
(knowledge + reasoning = intelligence)
•
•
•
•
construct a formal representation of the human-expert knowledge
in some field of data that is of interest, knowledge base is stored
in a set of production rules: if then conditional statements. May
be more complex…weights or probabilities before final action
Inference system—guides the expert system through its
knowledge base—rules to apply and order to apply them
Knowledge acquisition system and output device—acquiring the
knowledge and storage of rules on why the particular conclusion
was obtained.
Limited applications in geography, suited best for narrowly
defined, well-understood fields of application
Geocomputation– AI Applications
Artificial Neural Networks
Artificial Neural Networks (ANNs)
brain-like structure = intelligence
•
•
•
•
•
•
simple model of brain—interconnected set of neurons, a neuron
being a simple element with an input and output
Value of output signal = to weighted sum of input signals, signal
values are usually 0 or 1
Hidden layers connecting neurons
Supervised (known data)—adjusted connection weights to
activity, classify input data by learning the subtle patterns in data
set; example signal levels in remote sensing data
unsupervised mode (traditional)--similar to clustering analysis
solution.
Similar to multivariate statistical methods—maps combinations of
input X onto combinations of Y—may take any form, not limited
to logistic regression.
ANN Examples

Linear classifier
can only draw straight lines though the
cases as boundaries between the two
classes—numerous wrong
classifications
clear on knowledge base and how one
arrived at solution

Neural Network
ANN has potential to draw any line
shape through the cloud of
observations—produces a much more
accurate classification—no way of
knowing this is going to perform better,
but results show that ANN handles
larger, more complex problems (“scale
up better”)
problem of overtraining—matched too
well to training data set, learned
idiosyncrasies too well
black-box solutions—only see the
solution not what’s on the inside. OK
for land cover maps, not so good for
fire risk maps.
Geocomputation– AI Applications
Genetic Algorithms
•
•
•
•
•
•
Another AI technique—generate answers without the
how or why.
Loosely modeled on Evolution—genetic adaptation
and mutations that have evolved because they are
successful
Coding scheme devised to represent candidate
solutions—simplest level, string of binary digits
1001001010001
Potential solution scored on fitness criteria—
successful solutions allowed to “breed”
Crossover or mutation—random exchanges between
strings to produce new strings or randomly flipping
bits on current pop, slight changes are better than
huge randomization.
Now rare in spatial analysis and GIS literature.
Geocomputation– AI Applications
Agent-Based Systems
•
Also called Agent Technology—an agent is a
•
Autonomy—has the capacity for independent
•
Reactivity—can react in various ways to its current
•
Goal Direction—makes use of its capabilities to
•
Intelligent/communicate with other agents
•
Openshaw and MacGill’s space-time
attribute creature
computer program with various properties:
action
environment
pursue the current tasks at hand
solve problems in multiagent systems, example internet
search engines
Spatial Models
•
•
Instead of random models, develop process models that
explicitly represent the real processes and mechanisms
that operate to produce the observable geographical world
action
Possible to use in 3 different ways:
- as a basis for pattern measurement and hypothesis testing
- for prediction
- to enable exploration and understanding of how processes
works in the real world
•
Judgment about plausibility becomes as important as
results, especially crucial for prediction and exploration—
example using closed models to describe open systems of the
real world…sometimes this is necessary because to use an open
system would be impractical.
Spatial Models—
cellular Automata
•
•
•
Applicable to Raster GIS—a grid
consisting of nominal variable, a
finite number of discrete states.
Cell states changes/evolves
according to model time step, current
state of the cell and neighbors in the
lattice—again biology.
Classic CA, John Conway’s “Game of
Life”, can be used to represent a
geographical process
Cellular Automata
Two-dimensional automata
Spatial Models— Agent Models
•
Agents represent humans in a real simulated environment
•
Model Building Tools and Programs–
•
Cellular Models versus Agent Models…prediction of
permanent landscape features versus movement of people
across a landscape—important for future research
•
Predicting the past versus predicting the future—how well
does a model that predicts historical records be used to
predict future occurrences?
•
Equifinality problem-Open and closed model problem
again, what is the theoretical plausibility—needs to be
tested again and again.
Star Logo (MIT media lab);
Ascape (Brookings Institute); Swarm (Santa Fe Institute).
Finally—Coupling Models and GIS
•
•
•
•
Important—how the different spatial
models can be connected to the vast range
of geospatial data?
Models used in GIS for geographical data
types are different than those used in
spatial modeling.
Most significant—GIS data are static
whereas in spatial models are dynamic…
Software design problem of how to make
spatio- temporal data rapidly accessible....
Three Approaches to GIS Modeling
•
•
•
•
Loose Coupling—files transferred between a GIS and the model—
dynamics are calculated in model and displayed in GIS
Tight Coupling—each system write files that can be read by the
other…still difficult to view moving images in a GIS, each image
requires a new files and still a slow process.
Integrated Model and GIS systems do exist
- putting the required GIS functions in the model
- putting model functions in a GIS—harder
- develop a generic language for building models in a GIS
environment.
Latter two approaches being explored by researchers
- Magical
- GRASS GIS
(Geographic Resources Analysis Support System)
- PC Raster