PSYCO258 Lecture Notes - Lecture 20: Connectionism, Fault Tolerance, Somatotopic Arrangement
Document Summary
Connectionist approaches involve creating computer models for representing cognitive processes these are sometimes referred to as parallel distributed processing" (pdp) networks. Knowledge is represented in the distributed activity of many units (similar to the concept of neurons) Weights at each connection determine how strongly an incoming signal will activate the next unit. Pdp networks consist of 3 kinds of units: input units: activated by stimulation from environment: hidden units: receive input from input units, output units: receive input from hidden units. Learning in a pdp system typically progresses as follows: 1. network responds to stimulus, 2. provided with correct response, 3. modifies responding to match correct response. The error signal reflects the difference between actual activity of each output unit and the correct activity. Indicate how weights should be changed to allow the output signal to match the correct signal. This signal is sent back through the circuit (via back-propagation) The entire process repeats until error signal = 0.