The Classification Challenge

Print
Category: Cognition

[an early excerpt from my theory of cognition]

The pervasiveness of classification in human experience makes it a natural starting point if one is seeking to explain cognitive processes at a high level.  The challenge comes in the mechanics of developing these classifications.

  For example, if you take Hawkins’ assumption that the basic unit of memory is a quasi-photographic representation of the world, the problem of classification becomes fatal.  In the extreme case, we are forced to do a pair-wise comparison of each of the millions of objects that we perceive in an attempt to create classes.  Several issues arise from this approach:

  • Immense Memory Requirements.  This approach requires a substantial and accurate memory.  While the brain is mathematically capable of maintaining such a memory, its accuracy is not supported by experience.
  • Incompatible with Brain Design.  The brain does not provide the dynamic wiring to make possible each and every pair-wise connection proposed in this approach.
  • Computational Complexity.  Even if the brain somehow contained this data and permitted this pair-wise comparison, it would be computationally infeasible (N2).  Not only do we need to compare a million objects to a million other objects, but we must compare each of the thousands of characteristics that could be used as a basis of classification to determine which are relevant.

 

More critically, this simplistic computational complexity argument assumes that the brain is able to innately select and compare objects as wholes.  If this is not the case, the objects themselves must first be distilled by comparing various “clippings” out of our quasi-photographic experience.  Since a single object could be close and large (100 pixels by 100 pixels) or far and small (5 pixels by 5 pixels) and located anywhere in our visual field, this comparison problem would be immense.

 

Joomla Template designed by Joomla Themes.co