Table 1.

Pseudocode for two-stage, unsupervised training of the corticotectal model of multisensory enhancement

Set unit numbers, bias, and sensitivity; set input activation and target state probabilities; set learning rates, neighborhood properties, numbers of iterations, and thresholds; initialize primary and modulatory weights.

For the required number of stage-one training iterations, do

Determine target modality according to the target state probabilities

Determine primary input activation using input activation probabilities

Compute responses of DSC units to the input

Find the DSC unit with the maximal response (winning DSC unit)

Train primary weights of winning DSC unit and neighbors using Hebb's rule

Eliminate primary weights of any modality with values below the threshold

For the required number of stage-two training iterations, do

Determine target modality according to the target state probabilities

Determine primary and modulatory input activation using activation probabilities

Modulate primary weights according to modulatory inputs and weights

Compute responses of DSC units to primary input with modulation

Find active primary and modulatory inputs and DSC units by thresholding

Train modulatory weights of DSC units using the correlation-anti-correlation rule

If a modulatory input and a DSC unit are both active, then increase the modulatory input weights to inactive primary inputs and decrease the modulatory input weights to active primary inputs

If a modulatory input is active but a DSC unit is inactive, then decrease the modulatory weights to all primary inputs

End algorithm