CIS 140 Lecture Notes - Lecture 5: Connectionism, Classical Architecture, Activation Function
Document Summary
What are some the arguments made for why the classical architecture for the study of cognition fails? (e. g. , pp. Serial machines: ram -> cpu computational time grows increasingly large as size of machine increases. Dawson likens pdp/connectionist architecture to a computer simulation of the simplified brain features identified by neuroscientists. Summarize the structure and components of connectionist architecture. Input units (stimulus of environment turned into pattern of activity)-> hidden units (detect features) -> output units ( response of the system represented by pattern of activity in these) Contrast feed forward networks and recurrent networks. Define or describe the generalized delta rule. How does it facilitate learning in a connectionist system: give inputs, generates output, and calculates error, which is sent backwards through the layers and each unit"s weights are adjusted according to that. Explain the concept of graceful degradation: rather than failing completely with noisy inputs, machine"s outputs also become noisier- it still works, just not as well.