next up previous
Next: The Read-Out Layer Up: The Coherence Network Previous: The Model Neuron

The Coherence Layer

All neurons in a single coherence layer \( \mathcal{C}(i,j) \) of a disparity stack \( \mathcal{D}(i,j) \) are numbered by the shift parameter \( s \) (compare A.1.4). The total input current of a neuron \( s\in \mathcal{C}(i,j) \) in the coherence layer is given by


\begin{displaymath}
I_{s}(t)=\sum _{\bar{s}\in \mathcal{C}(i,j)}w^{\mathcal{C}}_...
...}}\cdot i_{\bar{s}}(t;\tau _{\mathcal{C}})+S(d_{(i,j)}(s))\, ,
\end{displaymath} (11)

where the raw disparity estimates \( d_{(i,j)}(s) \) are converted to actual input currents \( S(d_{(i,j)}(s)) \) by a linear function, \( S(x)=3.0+0.65*(x+10) \). The synaptic decay constant of the input currents is set to \( \tau _{\mathcal{C}}=0.01. \)

The interlayer coupling between the neurons in a coherence layer is an all-to-all coupling without self-interaction, i.e., the synaptic links are given by

\begin{displaymath}
w^{\mathcal{C}}_{s\bar{s}}=\left\{ \begin{array}{cc}
w_{\mat...
...f}\, \, s\neq \bar{s}\\
0 & \mbox {else}
\end{array}.\right.
\end{displaymath} (12)

Here \( N_{\mathcal{C}} \) is the number of neurons in a single disparity stack, and varies between 20 and 100 neurons in the reported simulations.

The value of the interlayer coupling \( w_{\mathcal{CC}} \) is the most important parameter of the whole system; it is this coupling constant which corresponds to the coherence threshold \( \epsilon \) in the abstract coherence detection scheme, defined in equation (1). To ensure that network operations realize the dynamical coherence detection scheme, \( w_{\mathcal{CC}} \) can vary between \( w_{\mathcal{CC}}\approx 0.5-2.0\protect \) (compare Fig. 4E, 4F and Fig. 9D). In the simulations reported here, a value of \( w_{\mathcal{CC}}=0.7 \) was used.


next up previous
Next: The Read-Out Layer Up: The Coherence Network Previous: The Model Neuron

2000-11-20