cognitive science - levels of analysis

what was Marr’s approach to how the mind performs information-processing?

distinguished different explanatory tasks at different levels and gave a general theoretical framework for combining these levels

what were Marr’s levels?

computational, algorithmic, and physical implementation

what is being identified at the computational level?

the specific IP problem that the system is configured to solve and general constraints upon any solution to that problem

what about an IP problem might be identified at the computational level?

inputs, outputs, and what is being computed or maximized

what does the algorithmic level explain?

how the system actually performs the information-processing task

what is identified at the algorithmic level?

input info, output info, and the algorithm for transforming input into the required output

what is specified at the algorithmic level?

how info is represented

what is identified at the physical implementation level?

neural structures implementing the basic representations the algorithm applies to and neural mechanisms that transform those representations according to the algorithm

how could you analyze a calculator at the physical implementation level?

explain the physical operations of the calculator by appealing to the properties of its transistors, resistors, and capacitors

how could you analyze a calculator at the computational level?

specify the laws of arithmetic and demonstrate that the behavior of the calculator conforms to these abstract laws

how could you analyze a calculator at the algorithmic level?

explain the information-processing steps carried out by the addition algorithm

what two key jobs of the visual system did Marr identify?

providing a 3D representation of the visual environment and an object-centered rather than viewer-centered frame of reference

what is discovered by analyzing the visual system at the computational level?

the input is light arriving at the retina and the output is a 3D representation of the environment

what questions are asked when analyzing the visual system at the algorithmic level?

what kind of info is extracted from the light at the retina, how is it represented, and how does the system get from this info to a 3D representation of the environment

what questions are asked when analyzing the visual system at the physical implementation level?

how does the lateral geniculate nucleus implement this process, what types of neurons are involved and how do their firing patterns carry out the algorithm

what are Marr’s stages of visual processing?

image-based processing, surface-based processing, object-based processing, and category-based processing

what are representational primitives?

the building blocks of the image representation at each level of processing

what do representational primitives allow for?

structure to be imposed at the next level of processing

what are the representational primitives of the retinal image?

the intensity values of light at each point

what are representational primitives at the surface-based processing level?

zero crossings which are computed as sudden intensity changes at the image-based level

how do you define a sudden intensity change algorithmically?

calculate intensity as you move horizontally across one line of points in the image

what can be used to detect the presence of an edge?

the magnitude of the first derivative of the intensity across the line of points in the image

what does the second derivative of the intensity across the line of points in the image produce?

two values at every edge which, if connected, would cross zero near the midpoint of the edge

how are zero-crossing detectors implemented in the LGN?

if adjacent P and Q cells are active at the same time, the AND cell detects a zero-crossing and several AND cells can be wired in tandem to detect an oriented zero-crossing

what are P cells?

aka on-center cells, fire when the center of their receptive field is stimulated

what are Q cells?

aka off-center, fire if the center is not stimulated but the surround is

what are AND cells?

respond to the firing of adjacent P and Q cells when they are firing at the same time

how does Marr’s theory of edge detection address the computational level?

goal is to infer surface boundaries from 2D image

how does Marr’s theory of edge detection address the algorithmic level?

compute zero-crossings over the image

how does Marr’s theory of edge detection address the physical implementation level?

wiring of P, Q, and AND cells in LGN

what is the criticism of the Marrian approach to vision?

Marr’s assumptions do not actually reflect natural constraints

what natural constraint of vision did Marr not address?

underdetermination, when there is insufficient info in the image to invert the process and recover a full description of the scene, necessitating assumptions about the world to solve the problem