26 Feb 2013

Andy Clark. 7.3 of Being There, “Primate Vision: From Feature Detection to Tuned Filters,” summary

summary by
Corry Shores
Search Blog Here. Index-tags are found on the bottom of the left column.]
[Central Entry Directory]

[Posthumanism Entry Directory]

[Andy Clark, Being There, entry directory]

Andy Clark

Being There:
Putting Brain, Body, and World Together Again

The Neuroscientific Image

Part 7.3
Primate Vision: From Feature Detection to Tuned Filters

Brief Summary:

The neuronal visual systems in the brain have parts that are maximally tuned to process data for one visual parameter or another, meaning that rather than having cells responsible for dealing only with certain complex forms [like a spiral], many various cells in the system cooperatively play a role in understanding all the visual properties of something being seen [like something’s spirality, breadth, etc, perhaps].


Clark will discuss neuroscientific research into primate vision, especially work by David Van Essen. (133c.d)

Cognitive neuroscience examines neuronal responses.

Anatomically, the macaque monkey possesses at least 32 visual brain areas and over 300 connecting | pathways. Major areas include early cortical processing sites such as V1 and V2, intermediate sites such as V4 and MT, and higher sites such as IT (inferotemporal cortex) and PP (posterior parietal cortex) (plate 1). The connecting pathways tend to go both ways—e.g. from V1 to V2 and back again. In addition, there is some "sideways" connectivity—e.g. between subareas within VI. (133-134, boldface mine)


(From Clark p.170)

There are ten levels of cortical processing in the system, and we will look at some of the more important ones. There are three populations of sub-cortical cells from which the system receives input. One population is the magnocellular (M) and another is the Parvocellular (P). And there is a processing pathway for M, and another one for P. Each population specializes in a different type of low-level information. P cells “have high spatial and low temporal resolution”, while M cells have “high temporal resolution.” (134b) This means that M cells deal with rapid motion perception, while P cells deal with color discrimination (among other things). So when we selectively destroy a monkey’s P cells, it can no longer distinguish colors although it still recognizes motion. (134b)

So the magno M cells discern motion, and there is a magno-denominated (MD) stream of processing. This stream includes neuron populations that are sensitive to the direction of some motion, especially in area MT, which we said above was an intermediate cortical processing site. When we electrically stimulate a part of MT, the monkey might “perceive” left motion even if the target object is really moving to the right. There is a higher stage in the processing hierarchy, MSDT, where there are cells sensitive to spiral motion.

The MD stream is ultimately connected to the posterior parietal cortex, which appears to use spatial information to control such high level functions as deciding where objects are and planning eye movements. (134d)

There is also the task of object recognition, which is determining what things are. This is handled by a stream rooted in P inputs, moving through V1, V4, and posterior inferotemporal areas (PIT), and it leads into central and anterior inferotemporal areas. (134d) This pathway specializes in form and color. As we go up the hierarchy, we find sites capable of processing increasingly complex forms. At a high level, there are even cells that respond maximally to such complex geometrical visual stimuli as hands and faces. (135a). But although one cell responds maximally to one kind of form, like a spiral, it will also to a lesser extent respond to other sorts of patterns.


[This means that cells are not like yes-no sensors that detect the presence of one form or its absence, but rather each participate in contributing information about some property of what is being seen, with all working together cooperatively.]

Although a cell may respond maximally to (e.g.) a spiral pattern, the same cell will respond to some degree to multiple other patterns also. It is often the tuning of a cell to a whole set of stimuli that is most revealing. This overall tuning enables one cell to participate in a large number of distributed patterns of encoding, contributing information both by its being active and by its degree of activity. Such considerations lead Van Essen and others to treat cells not as simple feature detectors signaling the presence or absence of some fixed parameter but rather as filters tuned along several stimulus dimensions, so that differences in firing rate allow one cell to encode multiple types of information. There is also strong evidence that the responses of cells in the middle and upper levels of the processing hierarchy are dependent on attention and other shifting parameters (Motter 1994), and that even cells in VI have their response characteristics modulated by the effects of local context (Knierim and Van Essen 1992). Treating neurons as tunable and modulable filters provides a powerful framework in which to formulate and understand such complex profiles. (135a.b, boldface mine)

But even though visual systems are complex, they can still be analyzed. (135d)

Andy Clark. Being There: Putting Brain, Body, and World Together Again. Cambridge, Massachusetts/London: MIT, 1997.

No comments:

Post a Comment