Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Simulation of Contagion on Very Large Social Networks
Keith Bisset, Jae-Seung Yeom Network Dynamics and Simulation Science Lab
Virginia Tech
§ Compute the spread of contagion across large contact networks (3x10^9 -‐ 8x10^10 nodes) § Contagious Disease (Influenza, Smallpox, Pertussis) § InformaDon Flow (Rumors, Fads) § Social Contagion (Smoking, Obesity, IncarceraDon, Social Unrest) § Computer Malware (Viruses)
§ Goal: Evaluate intervenDons to reduce or increase spread
The Problem – Contagion Diffusion
Need for ComputaDonal Epidemiology
§ InfecDous diseases account for more than 13 million deaths a year.
§ Controlled experiments oWen impossible in epidemiology due to ethical and pracDcal reasons.
§ ComputaDonal models help in understanding the space-‐Dme dynamics of epidemics.
§ Pervasive computaDonal environments can provide real-‐Dme access to models, data, and expert opinion to analysts as an epidemic unfolds.
§ Course of AcDon Analysis § Cost-‐benefit analysis § PrioriDzaDon of intervenDons § Planning Exercises
§ SituaDonal Awareness § EsDmaDon of number infected § EsDmaDon of disease
characterisDcs
§ Preparedness and Training
§ SyntheDc InformaDon Provides § InteracDons with other systems § Rapid response to emerging crises § Flexibility to adapt to different areas § IntegraDon of disparate data sources
§ ForecasDng § Surveillance § Social media data § Blogs § Survey data
ComputaDonal Epidemiology Driven by SyntheDc InformaDon
ConstrucDng syntheDc social contact networks
DISAGGREGATED POPULATION GENERATOR
DISAGGREGATED SYNTHETIC POPULATION
ACTIVITY, LOCATIONS, & ROUTE ASSIGNMENT
SYNTHETIC SOCIAL CONTACT NETWORK
WORK
SHOP
OTHERHOME
OTHER
WORKLUNCH
WORK
DOCTOR
SHOP
LOCATIONS
ROUTES
LOCATION ASSIGNMENT
HOMEWORK
GYM
DAYCARE
SHOP
WORK
LUNCH
LUNCH
SYNTHETIC NETWORK
SOCIAL NETWORKS
PEOPLE - age- household size- gender- income
LOCATION (x,y,z) -
land use -business type -
EDGE LABELS- activity type: shop, work, school- start time 1, end time 1- start time 2, end time 2
AGE
INCOME
STATUS
AUTO
26
$57K
Worker
Yes
26
$46K
Worker
Yes
7
$0
Student
No
12
$0
Student
No
Washington, DC
LOCATION
CENSUS POPULATION
POPULATION INFORMATION
SOCIAL NETWORKS
SYNTHETIC POPULATION
ANNAJOHN ALEXJOHN MATT
HOUSEHOLD
PERSON 1
AGE
INCOME
STATUS
4 PEOPLE
JOHN
26
57K
WORKER
SimulaDng contagion over dynamic networks
transition by interaction
S
I
Local transition
P = 1-exp(t· log(1-I·S·T)) • t: duration of potential
interaction time of co-presence • I: infectivity • S: susceptivity • T: Transmissibility
Infectious
Uninfected
S
I
t
P1
P2
P3
P4
Location Social contact network L1
L2
• Agent-based model • Realistic Population • Complex Interventions • Co-evolution between
• Contact Network • Individual Behavior • Public Policy • Individual Health
§ Pandemic Influenza Planning – Fall 2006 § How can we prepare for a likely influenza pandemic?
§ Emergence of H1N1 Influenza – Spring -‐ Fall 2009 § What are the characterisDcs of this influenza strain and their likely impact on US populaDons?
§ OpDmizing Household Response to Epidemics – Spring 2011 § Do behaviors within households, while harder to simulate, have a significant impact on the
course of an epidemic?
§ Adenovirus Pandemics SimulaDon and Analysis – August 2013 § How can decision makers become familiar with the challenges and decisions they are likely to
encounter during a naDonal pandemic, mainly centered on the allocaDon of scarce resources?
§ Emergence of Middle East Respiratory Syndrome – Now § We are exploring how simulaDon combined with a syntheDc informaDcs analyDc approach
might shed light on what may be driving this outbreak.
Selected Case Studies
§ Model individual decision-‐making in socio-‐economic context
§ Incorporate the diagnosDc process (false + / -‐) into intervenDon applicaDons
§ Dynamic applicaDon of anDvirals based on current state of infecDon
§ IntegraDng economics of health care decisions with dynamic simulaDon
§ Effect of individual decision-‐making on epidemics
AnDviral DistribuDon Planning
Would retail dispensing of an6virals reduce morbidity and mortality?
§ Early days of 2009 H1N1 pandemic § Transmissibility and Virulence not
known § Analysts at DTRA Reachback were
running simulaDons and interpreDng results on a 24 hour cycle to provide input to a daily briefing to HHS as the epidemic was unfolding
§ Regular training of analysts played criDcal role in rapid response
Emergence of H1N1 “Swine Flu”
§ Goal: Flu Forecast similar to current weather forcasts § Unlike weather people can actually do something to change outcome § Harness social media to infer populaDon level senDments and idenDfy
events of interest § Processing of alternaDve data sources upstream from event causaDon § Fusing mulDple predicDve data streams to dynamically opDmize
predicDons § IntegraDng Dme series analysis with simulaDon models § EvaluaDng social media as an informaDon source
Open Source Data based ForecasDng
Can freely available data be integrated with sta6s6cal and simula6on models to forecast the near term future?
§ Large, highly irregular, dynamic network § Large number of small messages § Large number of runs needed for study, even more for forecasts § Atypical instrucDon mix (not floaDng point)
Key Challenges
Instruction Type Portion Load/Store (L1-D) 58%
Branch 19% (70% Conditional) Floating Point < 1%
0.1
1
10
100
256 512 1K 2K 4K 8K 16K 32K 64K 128K 256KSi
mul
atio
n tim
e pe
r da
y (s
)Number of core-modules
Strong ScalingUS Population
§ System: NCSA Blue Waters § 352k Cray XE6 core modules § US PopulaDon: 280 million § Simulated Dme: 120 days § WallDme: 12 seconds § Granularity: < 800 people/core
EpiSimdemics Strong Scaling
Model 280 million people on 352k cores for 120 days: 12 seconds
Projected Global Popula6on 6me: 6 minutes To appear IPDPS 14
§ World DominaDon § More complex behavior § Long distance travel model § Co-‐circulaDng diseases (e.g., Influenza and ILI) § Complex Contagion
Ongoing Work