1 Lecture 16 Epidemics University of Nevada – Reno Computer Science & Engineering Department Fall 2015 CS 791 Special Topics: Network Architectures and Economics
Outline Epidemics Branching process SIR model SIS model SIRS model Synchronization 2
Epidemiology The study of populations and the transfer of disease between subsets of those populations. Models the rate at which a healthy population becomes infected and recovers 3 What if we have a network of contact between individuals?
Worm/Virus Spreading in the Internet How to control the spread of the threat by reinforcing local rules? E.g.: What is the maximum probability of contagion between two neighbors so that the overall network is protected against a complete invasion by the threat? How to model? Resembles a lot to the epidemiology 4
Model Basics and Goals 5 A contagious threat/idea Key goals: Will the epidemic invade the whole network? How long will the epidemic take to invade? Contagion One person “infects” another p – probability that an infectious person transmits the threat to a neighbor he/she meets k – # of new people an infected node meets during a time slot Reproductive Number: R 0 – expected # of new nodes getting infected at a time slot
Branching Process 6 k = 3 A 3-ary tree
Branching Process high p 7
Branching Process low p 8
Branching Process If R 0 < 1, the epidemic ends in a finite amount of time How long time? If R 0 > 1, the epidemic runs forever with a positive probability How large is the probability? Birth-Death Processes from queuing theory 9
SIR Model Susceptible, Infected, Removed Model The process Some in I and some in S state initially Each infectious node stay in the I state for t I slots During each step an I node has p probability of infecting a susceptible neighbor After t I slots, an I node “recovers” and goes to R state 10
SIR Model Susceptible: 11
SIR Model Infected 12
SIR Model Removed 13
SIR Model Recovered 14
SIR Model Recovered 15
SIR Model and Network Shape Network shape can change dynamics significantly.. p = 2/3, k = 2 R 0 = 4/3 > 1 A highly infectious threat But, (1/3) 4 = 1/81 probability that the epidemic will stop at each step! With probability 1, the epidemic will stop in a finite number of steps! 16
SIR Model and Percolation Another way to look at the SIR model Run the probability of contagion for each link Then see if there is a path from the root(s) of the threat to a node Helps to figure out if a particular node is more/less likely to be infected Percolation Theory: Also used for understanding the end-to-end connectivity patterns 17
SIS Model Only two states: Susceptible and Infected The Removed state is eliminated The process Some in I and some in S state initially Each infectious node stays in the I state for t I slots During each step an I node has p probability of infecting a susceptible neighbor After t I slots, an I node goes back to S state 18
SIS Model More general and maybe more realistic Can model larger set of scenarios A threat disappearing and appearing over cycles An inbounded supply of nodes 19
SIS and SIR Models 20
SIRS Model Three states: Susceptible, Infected, Removed The process Some in I and some in S state initially Each infectious node stays in the I state for t I slots During each step an I node has p probability of infecting a susceptible neighbor After t I slots, an I node goes back to R state and stays there for t R slots, and then returns to the S state 21
Synchronization Small-world can change things.. c – probability that a node has long-range links Again, network characteristics in action.. 22
Optimal Marketing – Revisited Which k people should we “infect” so that the spread of the “disease” is as large as possible? We can greedily pick these k people: first, pick the single best individual x1, then the best individual paired with x1, … 23
Optimal Marketing – Revisited The problem is in general NP-Hard, even for 0/1 transmission probabilities. The greedy algorithm picks an initial set of k people so that the spread is at least 63% as big as the largest possible spread over all k node starting sets. 24
Lecture 16: Summary Reproductive number w.r.t. 1 Network shape has serious effects Synchronization effects due to small- world 25
Lecture 16: Reading Easley & Kleinberg, Chapter 21 Chen, Gao, and Kwiat, Modeling the Spread of Active Worms, IEEE INFOCOM,