Some synaptic weights cause inhibitory neurons to excite past original firing

[nmsutton]

I have found in different networks that one-to-one bi-directional inhibitory neuron connections to excitatory neurons at certain weights cause an increase beyond original (non-inhibited) firing levels. This does not appear to be physiologically accurate based on what I have heard from someone knowledgeable about neural activities. I have included simple example code here (weight set with "inhib_level" variable; tested on CARLsim6).

For example:
3 layer network, 8x8 size, layer 1: excit.(L1), 2: inhib.(L2), 3: excit.(L3). Connections (all one-to-one): L1->L3, L3->L2, and L2->L3. Firing with weight settings:

L3 firing   |   L2 connection weights
24644       |   0.0
2286        |   1.0
1646        |   1.5
4977        |   2.0
12222       |   2.05
221366      |   3.0
353728      |   4.0

[bainro]

Hey Nate, that's a really weird bug. I'll see if CARLsim6 still shows this behavior! Thanks for the code snippet too.