CGSC170 Study Guide - Midterm Guide: Learning Rule, Connectionism, Step Function
Document Summary
Generally constructed with general purpose learning algorithms. These learning algorithms work by changing the connection weights between units. This eventually yield the desired output for the appropropriate input. Basic principle of connectionist networks is that many different units are active at a given time. Increasing number of steps that can be performed. If we think of each unit as performing an information processing step. Key to computational power of artificial neural networks. Inputs to weights to weighted sum to unit step function. Input values cannot really be manipulated because they are real world. If we change the strength of the weights then we can determine what the output will be. Attributing human like qualities to something else scares humans. Sometimes these computational networks can perform tasks far better than humans can. Error is propagated backwards through the network from the output units to the hidden units.