Biologically inspired computing and control

1
Biologically inspired computing and control

• Mark Neal

2
Why look at biology?

• Biology makes things that are very robust,
  flexible and reliable
• Engineers make things that are fragile,
  inflexible and unreliable (in comparison)
• Biological systems display some other endearing
  traits
• Modularity
• Component reuse
• Redundancy
• Self-organization
• Growth and development (and more!)

3
Goals

• Engineering
• Robustness, reliability and flexibility
• Self-organization
• Growth and development
• Learning
• Evolution
• Biological
• Simulation
• Theory testing

4
How to do it?

• Levels
• Detailed/faithful (individual mechanisms)
• Organizational (groups of mechanisms and
  interactions)
• Broad principles
• Granularity
• Molecular pathways
• Cells
• Organisms
• Populations

5
Existing biologically inspired techniques (1)

• Artificial neural networks (ANN)
• Based on animal nervous systems
• Very simplified models of neural function
• Very simplified models of interconnection
• Multi-layer perceptrons
• Detailed mechanism
• Self-organizing feauture maps
• Organization (of memory)
• Strengths and weaknesses (learning, applications,
  transparency, time)

6
Existing biologically inspired techniques (2)

• Genetic algorithms (GA)
• Based on Darwinian evolution
• Simplified models of genetic material
• Reasonable models of genetic recombination and
  variation
• Direct encoding
• Phenotype generation
• Genetic programming (GP)
• Strengths and weaknesses (search, representation,
  viability, population sizes)

7
Existing biologically inspired techniques (3)

• Flocking, swarming and ants
• Pheromone trails
• Follow your neighbour
• Run away from your neighbour (etc)
• Strengths and weaknesses (applications,
  representation, problem mapping, population sizes)

8
My approach

• Look closer at organism control systems
• Neural
• Endocrine
• Immune
• In biological systems are
• Closely integrated (with each other and the body)
• Varied timescales
• Amazingly effective
• In robots might be
• Closely integrated with each other and the robot
  hardware
• Able to deal with varied timescales
• Amazingly effective

9
Immunoneuroendocrine control (INE)

• Biologists already consider the three as a
  Supersystem
• Immune system is connected to neurons
• Neurons stimulate hormone release/production
• The immune system affects the nervous system
• How can we model this in artificial systems?
• Reuse some components (ANN AIS)
• Need an artificial endocrine system (AES)

10
Combining things

• Need to examine the biological systems
• and their interaction mechanisms
• Make tractable computational models
• That capture the biological properties
• And hopefully advantages
• Choose some applications
• And build some prototypes

11
Next time

• Details of
• Artificial neural network functions
• Artificial immune system functions
• Artificial endocrine system functions
• Trying to combine them sensibly
• Look at the paper on Timidity on the web-site