Balanced Nets

• A more commonly explored type of net is a balanced net.
• A simple way to do this is to have about as much excitatory synaptic strength as inhibitory synaptic strength.
• This used to be commonly done with 80% excitatory neurons, and 20% inhibitory with the inhibitory being four times as strong.
• This means that the system is near the border of chaos and can do some interesting calculations.
• A more recent trend is to keep the voltage of the neurons close to firing.
• Then when they go over the edge and fire, you need to balance the synaptic weight so the remaining neurons stay near the edge, and
• the just fired neuron quickly returns to the edge.
• There seems to be some evidence that the brain does this.
• Balanced firing systems are also being used for machine learning.