Semantic Consistency in Information Exchange

1
Semantic Consistency in Information Exchange
Pleiades Project

• Dwork, Kannan, Lee, Lincoln, Mitchell, Rubinfeld,
  Scedrov
• Cervesato, Sokolsky, Stern
• Gordon

2
Simulations

• Diana Gordon
• Navy Center for Applied Research in AI
• Naval Research Laboratory
• MURI Board Meeting
• 10/26/99

3
Outline

• MAVs
• Resource competition
• Virus vs anti-virus

4
Self-Assembly, Monitoring, Checking, and Steering
of MAV Sensing Grids

• Diana Gordon and William Spears
• Naval Research Laboratory
• Insup Lee and Oleg Sokolsky
• University of Pennsylvania
• Tugkan Batu, Ronitt Rubinfeld, and Patrick White
• Cornell University

5
Objective

• Combine
• Artificial Physics framework
• Monitoring and Checking framework
• Program checking
• Steering
• to solve an important Navy problem.

6
MAVs Scenario
Global monitoring UAV
Using Artificial Physics, MAVs form a hexagonal
lattice sensing grid
7
Severe Disturbance to Lattice
Global monitoring UAV
suspend repulsion!
8
Steering restores lattice
Global monitoring UAV
9
Empirical Evaluation of Steering
A smaller difference is preferable. Difference
between means is statistically significant.
10
Latest Progress

• Last meeting
• MAV simulation with MaC emulator.
• This meeting
• MAV simulation with real MaC -- see demo!

11
Evolving Better Strategies for Resource
Competition

• Diana Gordon and William Spears
• Naval Research Laboratory
• Insup Lee and Oleg Sokolsky
• University of Pennsylvania

12
Evolving Finite-State Automaton Strategies

• Population of
• strategies

Perturbation operators
Simulation
Evaluation
Selection
fitness scores
13
Resource Competition Simulation
2D grid of squares.
Squares correspond to resources
Two agents defense and adversary
Resources have north, south, east and west
neighbors
Blue square means protected by the defense
Red circle means controlled by the adversary
- Each move consists of going to a neighboring
resource.
- Once an agent occupies a resource, it
controls/protects that resource forever.
14
Resource Competition Simulation
27 occupied by the defense 15 occupied
by the adversary
15
Resource Competition Simulation

• Competition ends when
• all squares are occupied
• or time runs out
• Agent with most resources wins

16
Experimental Results

• 10 ? 10 grid
• An evolved defensive agent with 3 or more states
  in its plan wins at least 85 of the games
  against an opportunistic adversary
• Best average results over 10 runs
• With 5 states,
• defense wins 90 after 2000 generations
• Still room for strategy improvement
• unproductive cyclic behavior limits success

17
Combine with Monitoring, Checking and Steering

• Monitor
• checks for unproductive cyclic behavior
• Steering
• can suggest move that breaks the cycle
• makes progress toward an unoccupied region
• (in progress)

18
Virus and Anti-virus Simulation
(in progress)

• Communication topology
• 2D grid of squares
• Every square is an agent
• A PC or a small high-connected cluster of PCs
• At each time step
• Each agent has virus, anti-virus, or neither.
• Agent that has (anti-)virus can spread it to one
  neighbor
• Anti-virus strategy is evolved virus is
  opportunistic.
• Simulation results
• spreading frontier of virus and of anti-virus.

19
Future Directions for Virus Simulation

• Integrate virus simulation with MaC
• Co-evolving virus and anti-virus
• More realistic communication topologies
• Sparsely-connected, hierarchically-clustered
  topology (Kephart White 1993)
• Virus and anti-virus topologies differ (Meadows)
• Study variations in epidemiological model
• Vary model of sensing, actions and their effects
  (based on suggestions from Cathy Meadows).

20
Applications Current and Future

• MAV
• Virus and anti-virus propagation through
  population
• Intrusion detection
• Statistical monitoring of large datasets
• Trust management
• Schedulability analysis
• Wireless network routing
• Distributed control, monitoring, checking and
  steering of formations of a large number of real
  (rather than simulated) mini-robots