1 / 34

Diffusion and Cascading Behavior in Random Networks

Diffusion and Cascading Behavior in Random Networks. Marc Lelarge (INRIA-ENS) Columbia University Joint CS/EE Networking Seminar June 2, 2011. (1) Diffusion Model. inspired from game theory and statistical physics. (2) Results. from a mathematical analysis. (3) Heuristic.

solana
Télécharger la présentation

Diffusion and Cascading Behavior in Random Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Diffusion and Cascading Behavior in Random Networks Marc Lelarge (INRIA-ENS) Columbia University Joint CS/EE Networking Seminar June 2, 2011.

  2. (1) Diffusion Model inspired from game theory and statistical physics. (2) Results from a mathematical analysis. (3) Heuristic

  3. (0) Context Crossing the Chasm (Moore 1991)

  4. (1) Diffusion Model (2) Results (3) Heuristic

  5. (1) Coordination game… • Both receive payoff q. • Both receive payoff • 1-q>q. • Both receive nothing.

  6. (1)…on a network. • Everybody start with ICQ. • Total payoff = sum of the payoffs with each neighbor. • A seed of nodes switches to (Morris 2000)

  7. (1) Threshold Model • State of agent i is represented by • Switch from to if:

  8. (1) Model for the network? Statistical physics: bootstrap percolation.

  9. (1) Model for the network?

  10. (1) Random Graphs • Random graphs with given degree sequence introduced by Molloy and Reed (1995). • Examples: • Erdös-Réyni graphs, G(n,λ/n). • Graphs with power law degree distribution. • We are interested in large population asymptotics. • Average degree is λ.

  11. (1) Diffusion Model q = relative threshold λ= average degree (2) Results (3) Heuristic

  12. (1) Diffusion Model q = relative threshold λ= average degree (2) Results (3) Heuristic

  13. (2) Contagion (Morris 2000) • Doesthereexist a finite groupe of playerssuchthattheir action underbest responsedynamicsspreadscontagiouslyeverywhere? • Contagion threshold: = largest q for whichcontagiousdynamics are possible. • Example: interaction on the line

  14. (2)Anotherexample: d-regulartrees

  15. (2) Someexperiments Seed = one node, λ=3 and q=0.24 (source: the Technoverse blog)

  16. (2) Someexperiments Seed = one node, λ=3 and 1/q>4 (source: the Technoverse blog)

  17. (2) Someexperiments Seed = one node, λ=3 and q=0.24 (or 1/q>4) (source: the Technoverse blog)

  18. (2) Contagion threshold No cascade Global cascades In accordance with (Watts 2002)

  19. (2) A new Phase Transition

  20. (2) Pivotal players • Giant component of playersrequiringonly one neighbor to switch. Tipping point: Diffusion like standard epidemic Chasm : Pivotal players = Early adopters

  21. (2) q above contagion threshold • New parameter: size of the seed as a fraction of the total population 0 <α< 1. • Monotone dynamic → only one final state.

  22. (2)Minimal size of the seed, q>1/4 Tipping point: Connectivity helps Chasm : Connectivity hurts

  23. (2) q>1/4, lowconnectivity Connectivity helps the diffusion.

  24. (2) q>1/4, highconnectivity Connectivity inhibits the global cascade, but once it occurs, it facilitates its diffusion.

  25. (2) Equilibria for q<qc • Trivial equilibria: all A / all B • Initial seedapplies best-response, hencecanswitches back. If the dynamic converges, itis an equilibrium. • Robustness of all A equilibrium? • Initial seed = 2 pivotalneighbors • > pivotalequilibrium

  26. (2) Strength of Equilibria for q<qc Mean number of trials to switch from all A to pivotal equilibrium

  27. (2) Coexistence for q<qc Players A Players B Coexistence

  28. (1) Diffusion Model (2) Results (3) Heuristic

  29. (3) Locally tree-like Independent computations on trees

  30. (3) Branching Process Approximation • Local structure of G = random tree • Recursive Distributional Equation (RDE) or:

  31. (3) Solving the RDE

  32. (3) Phase transition in one picture

  33. Conclusion • Simple tractable model: • Threshold rule introduces local dependencies • Random network : heterogeneity of population • 2 regimes: • Low connectivity: tipping point • High connectivity: chasm • More results in the paper: • heterogeneity of thresholds, active/inactive links, rigorous proof.

  34. Thank you! • Diffusion and Cascading Behavior in Random Networks. • Available at http://www.di.ens.fr/~lelarge

More Related