L64 PNP 200 Lecture Notes - Lecture 8: Fault Tolerance, Turing Machine, Symbolic Computation
8 Mar 2018
School
Department
Course
Professor
Document Summary
Parallel rather than serial processing as activation spreads through a network (t serial) Information is being passed to different parts all at once. Knowledge distributed across a network (rather than stored in discrete symbol. Processing does not rely on explicit rules (other than those governing how act the network) Concerns about rule-based and serial models of simple cognitive abilities. Gap in neuroscientist"s toolkit: they were missing a level of analysis. The components don"t have to represent individual neurons, they can be. Worries about biological plausibility of physical symbol systems. Can be used to model multiple satisfactions of soft constraints. Can still perform well as it is damaged slowly. Intended as models of information-processing at the algorithmic level. The basic principle of connectionist networks are that many different units are. If we think of each unit as performing an information processing step, this vas number of steps that can be performed in a short time-span.
