Protective and Resilience Factors

Not only can the network model be used to establish potential vulnerability factors for depression, but manipulation of network parameters can be done in order to find configurations of parameters which prevent the network from incurring information processing biases after overtraining on negative stimuli. Potentially identification of such configurations can aid in the understanding of protective and resilience factors for depression.

For example, when the network's learning rate is relatively low ($\eta$=0.09) the network is able to learn positive stimuli adequately after being trained on negative stimuli. When the learning rate is higher ($\eta$=0.2), it has difficulty. This behavior might suggest that slow deliberate consideration of information, rather than immediate assimilation into one's knowledge structure is a resilience factor, allowing depression to be overcome efficiently.

One factor which was expected to protect the network but did not is strong initial learning. The network did assume the information processing biases characteristic of the depressed network, even when it was given large amounts of initial training (e.g., 100 epochs, and the error threshold for all patterns learned is below .0005). The clinical analog of this situation is that even people who have had large numbers of relatively positive experience might become depressed under the right circumstances.