How can connectionist models explain facial recognition?

In the context of cognitive science, connectionist models—also known as parallel-distributed processing—explain facial recognition as a result of firing activity generated within an artificial neural network.1

Rather than relying on step-by-step logical inferences, these models perform cognitive tasks through the collective, parallel processing of information. The process functions as follows:1

• Learning through Experience: The network’s ability to recognize a face is determined by the weights of the connections between its processing units (artificial neurons). These weights are forged through past experience, often using a back-propagation algorithm to adjust the network in response to representative examples.21
• Pattern Activation: Facial recognition occurs when the network generates a specific pattern of simultaneous neural firing activity that corresponds to a known face.1
• Constraint Satisfaction: The computation is accomplished by the simultaneous satisfaction of multiple constraints, allowing the system to identify complex patterns of association that would be difficult to map through explicit rules.21

By distributing knowledge across these connections rather than hard-coding it, connectionist models emulate the brain's ability to recognize patterns efficiently.2