longtailsnavigation10

longtailsnavigation10 - Long Tails and Navigation Networked...

Info iconThis preview shows pages 1–7. Sign up to view the full content.

View Full Document Right Arrow Icon
Long Tails and Navigation Networked Life CIS 112 Spring 2010 Prof. Michael Kearns
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
One More ( Structural ) Property… A pro pe rly tune d  α -m o de l c a n  simultaneously  explain small diameter high clustering coefficient other models can, too (e.g. cycle+random rewirings) But what about connectors and heavy-tailed degree distributions?   α -model and simple variants will  not  explain this intuitively, no “bias” towards large degree evolves “all vertices are created equal” As usual, we want a “natural” model to explain this
Background image of page 2
Quantifying Connectors: Heavy-Tailed Distributions
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Heavy-tailed Distributions Pareto  or  power law  distributions:  for random variables assuming integer values > 0 probability of value x ~ 1/x^ α typically 0 <  α  < 2; smaller  α gives heavier tail here are some  examples sometimes also referred to as being  scale-free For binomial, normal, and Poisson distributions the tail probabilities approach 0  exponentially  fast  Inverse  polynomial  decay vs.  inverse  exponential decay What kind of phenomena does this distribution model? What kind of process would  generate  it?
Background image of page 4
Distributions vs. Data All these distributions are  idealized models In practice, we do not see distributions, but  data Thus, there will be some  largest  value we observe Also, can be difficult to “eyeball” data and choose model So how do we distinguish between Poisson, power law, etc? Typical procedure: might restrict our attention to a  range  of values of interest accumulate  counts  of observed data into equal-sized bins look at counts on a  log-log plot note that power law:  log(Pr[value = x]) = log(1/x^ α ) = - α  log(x)  linear, slope – α Normal/Gaussian: log(Pr[value = x]) = log(a exp(-x^2/b)) = log(a) – x^2/b non-linear, concave near mean  Poisson:  log(Pr[value = x]) = log(exp(- λ λ ^x/x!)  also non-linear Let’s look at the paper on  dollar bill migration
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Heavy Tails Recap We plot the distribution or histogram of some “resource” on the x-axis, we put the amount or quantity of this resource (e.g. degrees)
Background image of page 6
Image of page 7
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 02/19/2010 for the course CIS 112 taught by Professor Kearns during the Spring '06 term at UPenn.

Page1 / 16

longtailsnavigation10 - Long Tails and Navigation Networked...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online