Design of SelfOrganizing Learning Array for Intelligent Machines

1
Design of Self-Organizing Learning Array for
Intelligent Machines
Janusz Starzyk School of Electrical Engineering
and Computer Science Heidi Meeting June 3 2005
Motivation How a new understanding of the brain
will lead to the creation of truly intelligent
machines from J. Hawkins On Intelligence
2
Elements of Intelligence

• Abstract thinking and action planning
• Capacity to learn and memorize useful things
• Spatio-temporal memories
• Ability to talk and communicate
• Intuition and creativity
• Consciousness
• Emotions and understanding others
• Surviving in complex environment and adaptation
• Perception
• Motor skills in relation to sensing and
  anticipation

3
Problems of Classical AI

• Lack of robustness and generalization
• No real-time processing
• Central processing of information by a single
  processor
• No natural interface to environment
• No self-organization
• Need to write software

4
Intelligent Behavior

• Emergent from interaction with environment
• Based on large number of sparsely connected
  neurons
• Asynchronous
• Self-timed
• Interact with environment through sensory-motor
  system
• Value driven
• Adaptive

5
Design principles of intelligent systems
from Rolf Pfeifer Understanding of Intelligence

• Design principles
• synthetic methodology
• time perspectives
• emergence
• diversity/compliance
• frame-of-reference

• Agent design
• complete agent principle
• cheap design
• ecological balance
• redundancy principle
• parallel, loosely coupled processes
• sensory-motor coordination
• value principle

6
The principle of cheap design

• intelligent agents cheap
• exploitation of ecological niche
• economical (but redundant)
• exploitation of specific physical properties of
  interaction with real world

7
Principle of ecological balance

• balance / task distribution between
• morphology
• neuronal processing (nervous system)
• materials
• environment
• balance in complexity
• given task environment
• match in complexity of sensory, motor, and neural
  system

8
The redundancy principle

• redundancy prerequisite for adaptive behavior
• partial overlap of functionality in different
  subsystems
• sensory systems different physical processes
  with information overlap

9
Generation of sensory stimulation through
interaction with environment

• multiple modalities
• constraints from morphology and materials
• generation of correlations through physical
  process
• basis for cross-modal associations

10
The principle of sensory-motor coordination
Holk Cruse no central control only local
neuronal communication global communication
through environment neuronal connections

• self-structuring of sensory data through
  interaction with environment
• physical process not computational
• prerequisite for learning

11
The principle of parallel, loosely coupled
processes

• Intelligent behavior emergent from
  agent-environment interaction
• Large number of parallel, loosely coupled
  processes
• Asynchronous
• Coordinated through agents
• sensory-motor system
• neural system
• interaction with environment

12
Neuron Structure and Self-Organizing Principles
12
13
Neuron Structure and Self-Organizing Principles
(Contd)
14
While we learn its functions can we emulate its
operation?
Brain Organization
15
Minicolumn Organization and Self Organizing
Learning Arrays

• V. Mountcastle argues that all regions of the
  brain perform the same algorithm
• SOLAR combines many groups of neurons
  (minicolumns) in a pseudorandom way
• Each microcolumn has the same structure
• Thus it performs the same computational algorithm
  satisfying Mountcastles principle
• VB Mountcastle (2003). Introduction to a
  special issue of Cerebral Cortex on columns.
  Cerebral Cortex, 13, 2-4. 

16
Cortical Minicolumns

• The basic unit of cortical operation is the
  minicolumn It contains of the order of 80-100
  neurons except in the primate striate cortex,
  where the number is more than doubled. The
  minicolumn measures of the order of 40-50 ?m in
  transverse diameter, separated from adjacent
  minicolumns by vertical, cell-sparse zones The
  minicolumn is produced by the iterative division
  of a small number of progenitor cells in the
  neuroepithelium. (Mountcastle, p. 2)
•  
• Stain of cortex in planum temporale.

17
Groupping of Minicolumns

•  
• Groupings of minicolumns seem to form the
  physiologically observed functional columns.
  Best known example is orientation columns in V1.
• They are significantly bigger than minicolumns,
  typically around 0.3-0.5 mm and have 4000-8000
  neurons
• Mountcastles summation
• Cortical columns are formed by the binding
  together of many minicolumns by common input and
  short range horizontal connections. The number
  of minicolumns per column varies between 50 and
  80. Long range intracortical projections link
  columns with similar functional properties. (p.
  3)

18
Sparse Connectivity

• The brain is sparsely connected.
• (Unlike most neural nets.)
•  
• A neuron in cortex may have on the order of
  100,000 synapses. There are more than 1010
  neurons in the brain. Fractional connectivity is
  very low 0.001.
• Implications 
• Connections are expensive biologically since they
  take up space, use energy, and are hard to wire
  up correctly.
• Therefore, connections are valuable.
• The pattern of connection is under tight control.
• Short local connections are cheaper than long
  ones.
• Our approximation makes extensive use of local
  connections for computation.

19
Introducing Self-Organizing Learning Array SOLAR

• SOLAR is a regular array of identical processing
  cells, connected to programmable routing
  channels.
• Each cell in the array has ability to
  self-organize by adapting its functionality in
  response to information contained in its input
  signals.
• Cells choose their input signals from the
  adjacent routing channels and send their output
  signals to the routing channels.
• Processing cells can be structured to implement
  minicolumns

20
SOLAR Hardware Architecture
21
SOLAR Routing Scheme
22
PCB SOLAR
XILINX VIRTEX XCV 1000
23
System SOLAR
24
Wiring in SOLAR
Initial wiring and final wiring selection for
credit card approval problem
25
SOLAR Classification Results
26
Associative SOLAR
27
Associations made in SOLAR
28
Brain Structure with Value System Properties

• Interacts with environment through sensors and
  actuators
• Uses distributed processing in sparsely connected
  neurons organized in minicolumns
• Uses spatio-temporal associative learning
• Uses feedback for input prediction and screening
  input information for novelty
• Develops an internal value system to evaluate its
  state in environment using reinforcement learning
• Plans output actions for each input to maximize
  the internal state value in relation to
  environment
• Uses redundant structures of sparsely connected
  processing elements

29
Possible Minicolumn Organization
Understanding
Improvement Detection
Expectation
Comparison
Inhibition
Novelty Detection
Anticipated Response
Reinf. Signal
Motor Outputs
Sensory Inputs
30
Postulates for Minicolumn Organization

• Learning should be restricted to unexpected
  situation or reward
• Anticipated response should have expected value
• Novelty detection should also apply to the value
  system
• Need mechanism to improve and compare the value
• Anticipated response block should learn the
  response that improves the value
• A RL optimization mechanism may be used to learn
  the optimum response for a given value system and
  sensory input
• Random perturbation should be applied to the
  optimum response to explore possible states and
  learn their the value
• New situation will result in new value and WTA
  will chose the winner

31
Minicolumn Selective Processing

• Sensory inputs are represented by more and more
  abstract features in the sensory inputs hierarchy
• Possible implementation is to use winner takes
  all or Hebbian circuits to select the best match
• Sameness principle of the observed objects to
  detect and learn feature invariances
• Time overlap of feature neuron activation to
  store temporal sequences
• Random wiring may be used to preselect sensory
  features
• Uses feedback for input prediction and screening
  input information for novelty
• Uses redundant structures of sparsely connected
  processing elements

32
Minicolumn Organization
superneuron
Value
Positive Reinforcement
Negative Reinforcement
Sensory
Motor
Sensory Inputs
Motor Outputs
33
Minicolumn Organization

• Sensory neurons are primarily responsible for
  providing information about environment
• They receive inputs from sensors or other sensory
  neurons on lower level
• They interact with motor neurons to represent
  action and state of environment
• They provide an input to reinforcement neurons
• They help to activate motor neurons
• Motor neurons are primarily responsible for
  activation of motor functions
• They are activated by reinforcement neurons with
  the help from sensory neurons
• They activate actuators or provide an input to
  lower level motor neurons
• They provide an input to sensory neurons
• Reinforcement neurons are primarily responsible
  for building the internal value system
• They receive inputs from reinforcement learning
  sensors or other reinforcement neurons on lower
  level
• They receive inputs from sensory neurons
• They provide an input to motor neurons
• They help to activate sensory neurons

34
Sensory Neurons Functions

• Sensory neurons
• Represent inputs from environment by
• Responding to activation from lower level
  (summation)
• Selecting most likely scenario (WTA)
• Interact with motor functions by
• Responding to activation from motor outputs
  (summation)
• Anticipate inputs and screen for novelty by
• Correlation to sensory inputs from higher level
• Inhibition of outputs to higher level
• Select useful information by
• Correlating its outputs with reinforcement
  neurons
• Identify invariances by
• Making spatio-temporal associations between
  neighbor sensory neurons