Elliot Anshelevich

1

• Elliot Anshelevich
• Department of Computer Science
• Interests
• Design and analysis of algorithms,
  especially for large decentralized
  networks.
• Strategic agents in networks and
  algorithmic game theory.
• Approximation algorithms.

2
Networks in Theoretical CS

• A major focus of Theoretical Computer Science is
  the study of networks
• Networks arise in many contexts,
  with many different properties

• The Internet
• Networks of processors
• Distributed Databases
• Social networks
• Control-Flow Networks
• Biological networks
• . . .

3
Networks with Independent Agents

• Internet is not centrally controlled
• Transportation Networks
• Social Networks
• Peer-to-peer Networks
• Business relationships
• To understand these, cannot assume centralized
  control
• Algorithmic Game Theory studies such agents

4
Transportation Networks

• Traffic patterns are not
  centrally controlled
• Behavior can be very different from centrally
  controlled traffic
• Braess Paradox sometimes building new roads can
  increase congestion

5
Transportation Networks

• Traffic patterns are not
  centrally controlled
• Price of anarchy quality lost because of
  agents being self-interested
• What do equilibria look like?
  How to improve them?

6
Agents in Network Design

• What if network is built
  by many self-interested
  agents?
• Properties of resulting network
  may be very different from
  
  the globally optimum one
• Connection Game
  (e.g. construction of roads and
  bus stations)
• Autonomous Systems and Contracts

7
Agents in Network Design

• What if network is built
  by many self-interested
  agents?
• Properties of resulting network
  may be very different from
  
  the globally optimum one
• Connection Game
• In general, converges to solution within log of
  optimal
• In multicast (single-source) case, can form a
  good solution
• True even for survivable networks

8
Agents in Network Design

• What if network is built
  by many self-interested
  agents?
• Properties of resulting network
  may be very different from
  
  the globally optimum one
• Connection Game
• Autonomous Systems and Contracts
• Characterize stable systems of contracts
• Can get the ASs to agree on a solution within
  factor 2 of optimal

peer
peer
customer
provider
9
Diffusion and Epidemiology

• Graph social network (or computer network)
• Nodes people/computers
• Edges relationships/links
• Diffusive network process
  disease,
  idea, computer virus, forest fire

10
Diffusion and Epidemiology

• Graph social network (or computer network)
• Nodes people/computers
• Edges relationships/links
• Diffusive network process
  disease,
  idea, computer virus, forest fire

11
Diffusion and Epidemiology

• Graph social network (or computer network)
• Nodes people/computers
• Edges relationships/links
• Diffusive network process
  disease,
  idea, computer virus, forest fire

12
Diffusion and Epidemiology

• Graph social network (or computer network)
• Nodes people/computers
• Edges relationships/links
• Diffusive network process
  disease,
  idea, computer virus, forest fire

13
Immunization

• Stop the spread by immunizing/protecting
  nodes/edges
• Goal immunize few, protect many from infection

14
Immunization

• Stop the spread by immunizing/protecting
  nodes/edges
• Goal immunize few, protect many from infection
• Somewhat know what to do if immunizing in advance
• What if immunizing in real-time?

15
Thank you.If want to learn more,
takeAlgorithmic Game Theory Spring 09