## Info

Timestep

FIGURE 2.6 Hidden layer size. This figure demonstrates the effects of changing the size of the hidden layer. Each line represents the average of five independent peak procedure simulations. The heavy black line represents the standard value for this parameter in all other simulations in this work. In general, increasing the size of the hidden layer improves the discrimination of the timing function.

### 2.3.4 Parameters

Like any mathematical model, the general timing model includes a number of parameters representing assumptions about the way the model should behave. The parameter values described here were used in all simulations in this work except where specifically stated otherwise.

First, there is the number of nodes in each of the three layers. The model uses two input nodes, ten hidden nodes, and one output node. Of the two input nodes, the first represents the time marker stimulus, going on at the beginning of the interval and off at the end. The second is always on and represents background stimuli in the model's environment. While the input and output layer sizes are determined by the environment and the task, the hidden layer could be any size. The basic peak-interval timing function using various numbers of hidden nodes is illustrated in Figure 2.6.

Second, there are parameters governing the initial values of the weights of the network. The weights between the input and hidden layers began with values randomly distributed between 0.03 and -0.02. The weights between the hidden and output layers began with values randomly distributed between 0.06 and -0.04. These values were then modified by the learning algorithm over the course of the simulation. The initial values were slightly skewed toward the positive in order to create some amount of initial responding that could then be modified by the learning algorithm. How changes in these initial weight ranges can affect interval timing functions is shown is Figures 2.7 and 2.8. Each node in the hidden and output layers also has a decay parameter that controls the rate at which it loses activation. This was set at 0.935 in all simulations described in this work. How altering this value affects the shape of the interval timing functions is shown in Figure 2.9. 0 0