Causal Emergence of Soft Events - PowerPoint PPT Presentation

1 / 17
About This Presentation
Title:

Causal Emergence of Soft Events

Description:

Unlike newspaper articles, their goal is not to report facts but just to report opinions. ... Axelrod mentioned the structural uncertainty of 'soft' political events. ... – PowerPoint PPT presentation

Number of Views:12
Avg rating:3.0/5.0
Slides: 18
Provided by: myriama
Category:

less

Transcript and Presenter's Notes

Title: Causal Emergence of Soft Events


1
Causal Emergence of Soft Events
AAAI 2007 Symposium Crystal City, Arlington, VA
  • Myriam Abramson
  • Naval Research Laboratory

2
Overview
  • What are soft events?
  • Cognitive Maps
  • Opinion Rule Mining
  • Learning Cognitive Maps
  • Fuzzy Neural Net (Kosko)
  • Dynamic Programming
  • A counter-insurgency cognitive map
  • Conclusions

3
What are soft events?
  • Some events and their factors are hard to observe
    and characterize but we can form an opinion about
    them.
  • Political events
  • Whether a marriage will last
  • Whether somebody will be hired
  • The value of a bridge in a time of war
  • .

4
Cognitive MapsBackground
  • Axelrod et al, The Structure of Decision The
    Cognitive Maps of Political Elites, 1976
  • Concept variables
  • Policy variables
  • Utility variables
  • Causal relationships
  • Bart Kosko, Fuzzy Thinking, 1994
  • Bart Kosko, Neural Networks and Fuzzy Systems,
    1992

5
Cognitive mapsEarly Example


Ability of Persian Gov To Maintain Order
Amount of Security in Persia
British Utility
-
Policy of Withdrawal

Strength of Persian Gov

Present Policy of Intervention In Persia
6
Cognitive MapsExample

Terrorism
Privacy
-
Simple cognitive map describing the opinion that
terrorism dilutes privacy (by indirectly
increasing surveillance) but that increased
privacy facilitates terrorism.
7
Opinion Rule MiningMotivation
  • Blogs are proliferating and are becoming an
    important source of information.
  • Unlike newspaper articles, their goal is not to
    report facts but just to report opinions.
  • Distinction between opinions and opinion rules
  • Opinion rules reflect a mental model of how we
    think the world around us works.
  • Opinion rule extraction in the blogosphere is
    unobtrusive and can reflect macro trends.

8
Opinion Rule MiningMethodology
  • Opinion rules can be represented by an
    association rule
  • IF A happens THEN B happens
  • with confidence 60 and support 30
  • Event A can be associated positively or
    negatively to Event B.

9
Opinion Rule MiningExamples
  • The amount of security augments the ability to
    maintain order.
  • International pressure inhibits the threat of
    war.
  • If the President has a long-term of office then
    impeachment is needed (that is promotes utility),
    but if the President has a short-term, then
    impeachability would be bad (that is would have
    lower utility).

10
Learning Cognitive MapsFuzzy Cognitive Maps
(Kosko)
  • Axelrod mentioned the structural uncertainty of
    soft political events.
  • Koskos insight was to apply fuzziness and
    Bidirectional Association Memory (BAM) to the
    quantification of cognitive maps.
  • Learning a cognitive map means to find the
    hidden patterns where the system either stays in
    equilibrium or settles in a cycle.

11
Fuzzy Neural Nets(Kosko)
  • Starting with initialized concept variables
  • A concept is clamped with value 1 to test its
    effects
  • Iteratively, multiply the vector of concept
    variables with the connection matrix
  • The effects are detected by the change in value
    of the concept values.

C1 (1,0,0)
0 1 0 1 0 -1 0 -1 1
E
C2 C1 E
12
Learning Cognitive MapsDynamic Programming (DP)
  • DP can apply to both deterministic and stochastic
    problems
  • Only utility variables
  • The utilities of the nodes are defined
    recursively from an initial value.
  • The confidence of an association rule quantifies
    the links of a cognitive map.
  • Monte Carlo simulation DP to learn a cognitive
    map.

C
A
B
Xt1 0.2Xt 0.8Yt
t 0.3 0.5 0.1
t1 0.14 0.34 0.41
t2 0.36 0.18 0.35
Node A Node B Node C
13
Counter-insurgency Cognitive Map
Strength of Government


Cooperation With Localities

-
Population Well-being
-
Intimidation
Police

-
Population Hardship


Cultural Understanding

-
-

Military Strikes
Infrastructure

-
-
-
Insurgency
Force Protection
-
Fick, N., To Defeat the Taliban Fight Less, Win
More. The Washington Post, August 12, 2007.
14
Counter-Insurgency Doctrine To Defeat the
Taliban Fight Less, Win More
  • When U.N. teams began building new stretches of
    road in volative Afghan provinces such as Zabul
    and Kandahar, insurgents inevitably attack the
    workers. But as the projects progress and
    villagers begin to see the benefits of having
    paved access to market and health care, The
    Taliban attacks become less frequent.
  • The more you protect your forces, the less safe
    you may be. To be effective, troops, diplomats
    and civilian workers need to get out among the
    people. Of course, mingling with the population
    means exposing ourselves to attacks, and
    commanders have an obligation to safeguard their
    troops.

15
Conclusion
  • Opinion rules can help predict the behavior of
    opponents and understand a culture.
  • Feedback loops in cognitive maps make them a
    dynamical system.
  • Cognitive maps of collectivities can be
    synthesized from the mining of opinion rules in
    open-source data in a non-obtrusive way.
  • Dynamic programming Monte Carlo simulation
    based on a cognitive map model can predict
    emergent patterns and soft events.

16
Opinions?
17
Background Slides
Maruyama (1963)
Write a Comment
User Comments (0)
About PowerShow.com