Dyslexia - No Input

Project: dyslex.proj
Emergent version: 4.0.19

The Semantics layer was disconnected from other layers (only self-connections were present). There was also no input data ("clear" data) so there was no external influence on the Semantics layer.

All units from the Semantics layer were divided into 4 groups (35 units each) on account of their location on the layer (left-bottom, right-bottom, left-top or right-top). Every group was associated with a Gaussian function given by the formula:

(1)

\begin{align} g_k(t) = \exp{ \left( - \frac{\sum_{i | w_i \neq 0 }{ \beta \left( w_k^{(i)} X^{(i)}(t) - \mu_k^{(i)} \right) ^2 }} {2 \sigma_k^2} \right) } \qquad k = 1, \dots, 4 \end{align}

where $\mu_k$, $\sigma_k$ are center and dispersion of Gaussian function respectively and $w_k$ is weight vector which has ones only in these dimensions that match unit numbers from a given group (other dimensions contain zeros).

Each Gaussian function is projected in a specific direction on the 2D space:

• left-bottom ($g_1$): $[-1,-1]$
• right-bottom ($g_2$): $[1,-1]$
• left-top ($g_3$): $[-1,1]$
• right-top ($g_4$): $[1,1]$

$g_3$	$g_4$
$g_1$	$g_2$

(so direction corresponds to group location on the layer)

The formula for such mapping is as follows:

(2)

\begin{equation} x(t) = -g_1(t) + g_2(t) - g_3(t) + g_4(t), \end{equation}

(3)

\begin{equation} y(t) = -g_1(t) - g_2(t) + g_3(t) + g_4(t), \end{equation}

where $g_1(t)$, $g_2(t)$, $g_3(t)$, $g_4(t)$ are values of the Gaussian functions and $(x(t),y(t))$ is a point on the 2D space which represents a point of trajectory at moment $t$.

Vm Noise (Gaussian, var = 0.008)

There were 10 trials with the same model parameters.
Each trial last 1000 cycles (so there are 1000 trajectory points in every trail).

All dispersions of Gaussian functions was equal 1.0.

All 10 trials together

Each trial on a different plot