Methods

  A simplified system with 2 input cells that individually respond to either the presence or absence of food directly in front of the ant was set up.  These two input cells can affect n ‘hidden’ cells (where n is variable) that can be arranged in any arrangement, i.e. each ‘hidden’ cell can affect every ‘hidden’ cell and output cell (one example is shown in Fig. 2).  The individual synapses would be modeled as having a property in the range of -1 to 1 where -1 is fully inhibitory, 1 is fully excitory and 0 would represent no connection.  Similarly each units state is represented by a property in the range of -1 to 1. A threshold of 0.7 was set as being required for a unit to activate its synaptic connections and for each unit the inputs would be summed up to see whether or not this unit will relay signal to its downstream cells.  Once this system was set up, a population was created and seeded with random values for the connections.  Each population was then tested for its performance in the Tracker task as described in the introduction.  The top 20% of the population were kept and ‘mutated’ by randomly varying each variable in the range of +/- 0.2.  And the process repeated.  Java was used for this and the specific code can be downloaded using the link above.