PSYC 2145 Lecture Notes - Lecture 18: Semantic Dementia, Fault Tolerance, Backpropagation
Document Summary
Clicker question the output of a unit within a neural network model is . The result of applying an activation function to the weighted sum of its inputs. The outputs of these units can be either 0 or 1 and those are called boolean values. We display these simple logical functions because they are easy to understand, but that is not all these networks do! The brain often computes functions with continuous-valued outputs, and which are probabilistic, rather than logical. Network learns to activate appropriate output units for each pair. Input layer (left side) hidden layer (middle) output later (right side) Hidden because you can"t actually see its contents from the outside of the model. Difficult to see from the outside unless you do some investigation. The fact that you"re indicating to the model what the correct answer ought to be, makes this supervised learning.