The Disambiguation Experiment
The CANT model
- A network of leaky integrator neurons
- distance biased connections
- inhibitory and excitatory neurons
- fatiguing neurons
- hebbian learning
The experiment, learning the CAs:
- The network is a 20 by 20 network with a toroidal topology
- Stimuli is 20 random neurons from one of two patterns, either top or bottom
- Present stimuli for 20 cycles and allow 80 more cycles of running without stimuli
- Present each patern type 25 times to allow unambiguous CAs to be learned
The experiment, testing for disambiguation
- turn off learning
- present an ambiguous stimuli for 20 cycles,
10 neurons from the top, and 10 from the bottom
- allow the net to run for 80 more cycles
- see which CA is active
