Organic Computing

1
Organic Computing

• A Summary of Issues Raised During the Symposium

A Summary of Issues Raised During the
Symposiumbut also those that were left out!
2
Why Organic Computing?

• The Software Crisis
• Many software projects fail or exceed projected
  costs
• The algorithmic schema is doomed where too many
  processes need to be coordinated/anticipated by a
  programmer
• Distributed processing across networks
• Sensor-rich processing
• Strongly dynamic environments

There seems to be consensus that a coordinated
research initiative towards Organic Computing
would be useful!
3
Where Is Organic Computing Useful?

• Computational processing characterized by
• High precisions is not the primary purpose (?)
• Processes with some fault tolerance
• Sensor-rich processing
• Dynamic environments
• Processes where manual decomposition is tough
• E.g., HCI, computer action,perception, language,
  networking, complex process control
• Organic computing may be less useful for
• Simple computer tools (e.g., text editors,
  graphic programs)
• Zero-fault tolerance tools (accounting,
  databases, etc.)

4
What is Organic Computing?

• Computing with a hierarchy/nesting of active
  modules
• Dynamics of module formation and growth
• Search in module space
• Teaching of OC modules (Bootstrapping)
• Communication between OC modules
• Programming by teaching, biasing, and re-using
  modules
• The Physics of Biology
• An Evolvable Computer

5
Properties of Organic Computing

• Robustness
• Structural Robustness (-gtredundancy)
• Process Robustness (-gtfeedback loops)
• Noise Rejection (-gtredundancy)
• Generalization (-gtlarge basins of attraction)
• Flexibility
• Stability/Plasticity dilemma
• Exploration vs. Exploitation
• Modularity
• Hierarchies
• Adaptation, Learning, Self-Organization
• Templates Instantiation
• Specialization

6
OC Modularity-Research?

• Modularity is good!
• Reduces search space (data hiding, dimensionality
  reduction), allows re-use of modules, allows
  later global optimization
• Communication between Modules
• Embodiment can serve a communication device
• Processing within a module Some form of
  self-organization
• What are modules?
• Inputs, outputs, reinforcement, internal state
  (memory)
• Specific computational abilities
• Confidence (probabilistic modules?)
• Module Competition vs. Restructuring of Modules
• Module selection (based on confidence,
  reinforcement, self-consistency, inter-module
  consistency?)

7
Formalizing the OC Module
Reinforcement(Environment)
Computational Specialization Instantiation
OC Module
Inputs afferent efferent output from
other modules confidence values values may be
grouped
Outputs transformed values confidences
8
Self-Organization

• Within Modules Comp. Specialization
• Unsupervised learning
• Feature extraction, density estimation
• Dimensionality reduction
• Self-consistency recognition-generation networks
  (statistical or functional inversion)
• (Self-)Supervised Learning
• Nonlinear correlations-gtdimensionality reduction
• Reinforcement Learning
• Somehow we need to encode a goal
• Dynamical Systems (e.g., feedback loops?)
• Between Modules
• Contribution to the goal
• Consistency with other modules
• With who to interact?

9
OC-Related Computing Areas

• Statistical Learning
• Reinforcement learning (in particular modular RL)
• (Self-)supervised learning
• Unsupervised learning
• Committee machines
• Genetic/Evolutionary Programming
• Artificial Life
• Soft-Computation, Natural Computation
• Distributed Computing
• Agents community
• Computer networking

10
Research Approaches Towards Organic Computing

• Methods of Self-Organization Self-Consistency
• Within the module
• Theory of Self-organization of Modules
• Communication(Interaction) between modules
• Propagation of probabilities across modules?
• Selection
• Task-oriented Experiments
• Manual creation of OC modules and their
  interaction
• Exploration of principles and constraints
• Generalization to other tasks
• Compare systems programmed with different
  paradigms
• Model experimental data