next up previous
Next: Acknowledgments Up: Why Have Lateral Connections Previous: Lateral comparisons on

Discussion

  The combined bottom-up/lateral mode of information processing, assumed throughout this paper, may account only for a part of the computational prowess of the visual system, much of the rest being probably due to the top-down propagation of information [28,46]. Nevertheless, the bottom-up/lateral processing view of perception and, indeed, of cognition, has been the mainstay of brain science since the times of John Locke and David Hume, and up to the present day. Locke's theory of representation by covariation, or, in his words, ``conformity'' between the world and the representation [22] remains influential, despite attacks by generations of philosophers [5]. In sensory physiology, the notion of functional units selectively and reliably responsive to external stimuli (that is, units whose output covaries with the sensory input in a well-defined and predictable manner) has found a wide acceptance in the form of the feature detector doctrine. The most common objections to this doctrine (the claims of causal inadequacy and of combinatorial explosion in the required number of detectors) have been swept aside by appealing to coarse coding and similarity-based representation [2,6,8,42]; see Figure 10.

The complementary side of Locke's world view, echoed in [19] --- the formation of complex ``ideas'' by associating simple ones (those directly available in the sensory input) --- has received recently stunningly direct experimental support in Miyashita's results (see the Physiology of ... section). This and other examples from low-level and high-level vision, in which lateral connections turned out to be indispensable to the modeler, indicate that a common functional principle may underlie the existence of lateral connections in the cortex. I propose that the unifying principle is associative formation of structured coarse-coded representations. Given the computational power of lateral information processing, it should not come as a surprise that association of ideas (whether simple, of the form ``contrast element C is present at orientation '', or complex, of the form ``stimulus resembles shape prototype '') leads to the emergence of a representation that is, in a sense, greater than the sum of its parts.

  
Figure 10: Lateral thinking in the philosophy of representation. Locke's idea of representation by covariation (that is, by a causal connection between the world and the mind) has been criticized as incapable of providing a sufficiently firm basis for veridical representation (see the discussion in [5]). Representation by covariation, which appears to be the central tacit assumption behind much of the contemporary brain science, becomes viable if combined with Shepard's notion of ``second-order'' isomorphism [6]. Shepard notes that ``... the isomorphism should be sought --- not in the first-order relation between (a) an individual object, and (b) its corresponding internal representation --- but in the second-order relation between (a) the relations among alternative external objects, and (b) the relations among their corresponding internal representations. Thus, although the internal representation for a square need not itself be square, it should (whatever it is) at least have a closer functional relation to the internal representation for a rectangle than to that, say, for a green flash or the taste of a persimmon'' [37]. Lateral links among object representations may help to make second-order isomorphism work, as argued in the Lateral Comparisons ... section.


next up previous
Next: Acknowledgments Up: Why Have Lateral Connections Previous: Lateral comparisons on