Presentation is loading. Please wait.

Presentation is loading. Please wait.

Mutual support in agent networks

Similar presentations


Presentation on theme: "Mutual support in agent networks"— Presentation transcript:

1 Mutual support in agent networks
Wan Ahmad Tajuddin Wan Abdullah Dept of Physics, Universiti Malaya, Kuala Lumpur A .K. M. Azhar Graduate School of Management, Universiti Putra Malaysia, Serdang

2 Can we model social/economic systems using (extended) statistical physics?

3 Agent models Agents: in free space/ on lattices/ on networks
Interactions: choices – game theoretic – strategies, memories [internal state], information [pertaining to internal state of engaged agent], learning

4 Some games people play... Prisoner's dilemma on lattice
Minority game and modifications Deviant's dilemma etc

5 This model Agents i, i = 1,...,N, with measure of influence (i).
Random meetings of i with j; i decides to support j or otherwise, depending on a strategy matrix (i (i,j) = 1 (i decides to support j) with probability (0.5+ |(i,)|)/(+ |(i,)|) (i,j) = 0 otherwise is noise level, is situational 6-digit binary number b5b4b3b2b1b0 current value of (i,j) (j,i) if |(i)-(j)| < (i) previously if (i) > (j) previously if |(i)-(j)| < (i) now if (i) > (j) now

6 This model (cont'd) j then has increased influence,
(j) → (j) + (i,j)(i) where  is a constant Agents are then ranked according to (i), and carries out reinforcement learning quantified by rank change: (i,) → (i,) + Ri' - Ri

7 Computational studies
Monte Carlo simulations N = 100 200 iterations (interactions per pair) normalise  each time learn for move 1 timestep ago

8 a = 0.3, n = 10.0

9 strategies

10 strategies

11 strategies

12 strategies – top 10%

13 strategies – top 10%

14 strategies – top 10%

15 influence scatter evolution

16 influence scatter evolution

17 support evolution

18 Learn for immediate move
Learn for move 2 timesteps ago

19 a = 0.1 a = 0.5

20 n = 5.0 n = 20.0

21 conclusions Emergence of structures/heterogeneities
Emergence of successful (?) strategies - understand rationale for strategies Phase space so far homogeneous Relate to real social dynamics?


Download ppt "Mutual support in agent networks"

Similar presentations


Ads by Google