COGS 2160 Lecture Notes - Lecture 10: Artificial Neuron, Connectionism, Parallel Computing
Document Summary
Lecture 10: neural networks i basics & motivations. Stimulation from the environment stimulates the input units: the input units then stimulate the hidden units which then stimulate the output layer. Building logic gates: assume a binary activation function: the cell either fires or it doesn"t, only 2 values: 0 and 1. Aggregating statistical information: e. g. if you want to know if something is a vegetable, the inputs are indicative of whether something is a vegetable or not. If something is green, it increases the probability of it being a vegetable. Motivations for connectionism: biological plausibility, modeled after neurons as opposed to turing machines, speed of processing, graceful degradation, graceful learning. Graceful degradation: brains respond in distinctive and flexible ways to damage and impairment, e. g. people lose the ability to recognize faces, characteristic partial breakdown patterns in response to local damage/lesions. Cogs 2160: graceful degradation when cognitive abilities slowly deteriorate.