Engineering SelfOrganization in MAS Complex adaptive systems using situated MAS

1
Engineering Self-Organization in MASComplex
adaptive systems using situated MAS

• Salima Hassas
• LIRIS-CNRS
• Lyon I University, France

2
Outline

• Introduction
• Self-organizing Computing systems as
  self-organizing Situated MAS
• Coupling to the environment
• Co-evolution of social and spatial organizations
• A complex adaptive system perspective
• Some illustrations
• Conclusion

3
Introduction

• Evolution of computing systems High complexity
• Complex environments, sophisticated applications
• Complex data and usages/practices
• Emergence of new needs, new practices,
• Computing system a system open on its
  environment
• Complexity of the environment distributed,
  dynamic, evolving, uncertain,

4
Examples

• Evolution of the Internet and the Web
• A complex dynamic network, exhibiting a
  self-organizing character
• Evolution of Software Engineering
• Awareness of dynamic changes of the environment
• Design at run-time
• User more and more present
• User centered systems
• Capture usages/practices through system use

5
Issue

• How to design computing systems exhibiting
  intelligence while embodied in their environment,
  considered at its widest meaning?
• Widest meaning
• physical as well as conceptual environment

6
Issue

• Environment is thus put at the heart of the
  engineering of the computing system
• Conceptual environment related to uses
  (practices)
• Place of materialization of uses
• (ex Virtual communities)
• Physical environment
• Place of materialization (embodiment) of the
  computing system gta complex network of resources
• Place of inscription of traces of uses related to
  actions and interactions
• gt Ex web/Internet topology expresses usages

7
Approach

• The system is considered through its coupling
  with its environment
• A double articulation
• Physical articulation Structural Coupling
• Conceptual articulation Behavioral Coupling
• Retroactive effects of one coupling on another
• Organizational articulation

8
Implications on a MAS

• Using Situated MAS to implement this kind of
  computing systems (complex systems aware of their
  environment)
• The MAS is subject to the same coupling with its
  environment
• Structural Coupling Physical articulation
• Spatial organization of the MAS / physical
  environment
• Behavioral Coupling Conceptual articulation
• Social organization of the MAS / conceptual
  environment
• Retroactive effects of one coupling on another
• Co-evolution of spatial and social organizations
  of the MAS

9
Implications on a MAS

• The design of the situated MAS must address
• its spatial organization
• its social organization
• And the co-evolution of both organizations
  through the MAS dynamics
• Self-organization is mandatory
• The eternal ants foraging example
• Emergent Structures shortest paths from nest to
  food source
• Physical materialization of the spatial
  organization
• Emergent behavior self-catalytic frequentation
  of paths
• Conceptual materialization of the social
  organization
• Self-organization is the mechanism which allows
  co-evolution of social and spatial organization
• Need for a glue between both organizations
  stigmergy mechanism

10
Our vision

• The computing system as a Complex Adaptive System
  
• A set of interconnected components (agents),
  strongly interacting with one another at
  different levels
• Micro level retroactive interactions between
  agents
• (local behaviors)
• Macro level emerging structure and organization
  of the system
• (global behaviors)
• System Dynamics maintaining the system
  organization
• Non linear dynamic (retroactions and emergences)
• Coupling to the environment autopoïetic vision
• Co-evolution structures-their generating
  processes
• reflective loop
• Co-evolution of spatial organization-social
  organization in MAS

11
Positioning
Complex Adaptive Systems
Non Linear Dynamic Systems
General System Theory Cybernetics
Chaos Theory, statistical mechanics,..
Self-Organizing Computing systems
Situated Multi-Agents
Artificial life Embodied intelligence Nature-inspi
red computing
12
Propositions
13
A Guiding framework

• A framework for developing self-organizing
  computing systems
• Physical materialization of the environment and
  its spatial representation
• A complex dynamic network of resources
  importance of topology
• Embodied Intelligence using situated agents
• Population of situated agents embodied in a
  physical (spatial) environment gt incarnation
  of the computing system
• Stigmergy
• Spatial structure for coding control and
  meta-control information
• case of the electronic pheromone
• Individual behaviors Correlation
• Strategy balancing exploitation (reinforcement)
  /exploration (diversity)

14
Topology

• Topology of networks produced by human activities
  / nature
• (exhibiting self-organization ..)

• Scale Free Networks and  small world  property
• Scale free
• Small number of highly connected nodes,
  distributed randomly
• High number of nodes weakly connected
• Small world
• Small average length between any couple of nodes

15
Illustration
(IEEE Swarm Intelligence 03 publication)

• A computing ecosystem on the web WACO system
  
• A multi-agents system An ecosystem composed of
  Web Ants (mobile agents),
• mapped on the web
• Using a social insects paradigm (stigmergy)
• Combining foraging and collective sorting
• Specialization/population regulation following
  the web content
• Dynamics of population Energy mechanism
  (order/disorder of web content)

16
Illustration

• Experiment1 Disorder decreasing
• Disorder decreases while new documents are
  created
• Disordernumber of scattered documents
• Negative value of disorder multiple clustering
  of a same document

Scattered documents are those created (order
emergence)
17
Illustration

• Experiment 2 Clusters forming
• Effectiveness of clustering
• Size of clusters increases regularly
• Sudden (small) decrease of mean clusters sizes
  near time 80000
• Order emergence disturbed by new creations

Scattered documents are those created
18
Illustration

• Experiment3 Energy evolution
• Energy evolution follows the disorder evolution
• Decrease near time 80000
• gt order emergence
• Decrease near time 100000
• gt new clustering operation specialists creation

order emergence
Specialists creation
19
Illustration

• Energy of specialized agents
• Specialists energy increase during clusters
  forming
• Near order emergence (near time 80000) energy 0
• Sudden increasing near time 100000, new clusters
  apparition

20
Illustration

• ECoNET
• Dynamic multi-criteria balancing on a network of
  processors
• Problem
• On a network of processors, processes must find
  dynamically a spatial repartition allowing the
  satisfaction of the 4 following criteria
• Balancing the average of the perceived load
• Spatial clustering of processes belonging to the
  same application (sharing of same data,
  resources)
• Spatial clustering of processes belonging to
  highly communicating different applications
  (minimize communications delays)
• Spatial repulsion of concurrent processes
  accessing the same resources (resources access
  conflicts)
• Note
• Environment is subject to perturbations and
  criteria may evolve during time..

21
Conclusion

• Towards a methodolgy of self-organizing computing
  systems
• Environment A central point for the system
• Situated MAS paradigm incarnation of the
  computing system
• The MAS is subject to the same coupling with
  respect to its environment
• Deployment of the MAS in its physical environment
  spatial organization
• Maintaining the spatial organization through the
  social organization of the MAS
• Retro-active effects of one organization on the
  other

22
Conclusion

• Necessary to study
• Relation between spatial organization and the
  environment topology (and their retro-active
  effects)
• Reflective coupling structure-processes
  (autopoïesis)
• Relation between spatial organization and social
  organization (and their retro-active effects)
• Structure-environment coupling (self-organization)
  
• Reflective effects between the two coupling
• Co-evolution of both (emergent) organizations and
  the environment topology

23
Thank you ))