Slide 20 of 37
Notes:
A relaxation system solves a number of problems inherent in other classification models. It includes both bottom-up (stimulus feature driven) and top-down (hypothesis driven) processes. It includes features and representations of the central tendency of categories, but allows for more flexibility in recognizing members of each category. It does not require a perfect match for recognition as a template model might, but seeks to find the best match between category and exemplar.
Because the constraints are implemented as connections among neurons, the implementation of the system cannot be separated from the computations of the system. In contrast, because a Turing machine is defined relative to the computations it performs, and because it can be programmed by symbols presented to it, all Turing machines are computationally equivalent to one another. Furthermore, in a Turing machine, the relation between the symbols used in the computations and the items and features they represent is arbitrary, but the relation between activity in certain parts of the brain and the features that activity represents is fairly tightly determined by the structure of the brain. Biomimetic systems can be simulated, but veridical simulations must reflect the constraints imposed by the properties of the system in which the processes evolved.
