The chapters in this book have been organized around four main topics: (1) effects of lateral interactions on the receptive fields of neurons; (2) their role in establishing global effects such as inhibition, synchronization, and Gestalt phenomena; (3) the development of lateral connections and cortical columns; and (4) lateral interactions in high-level vision including object recognition and depth perception.
The first three chapters focus on receptive fields, each using a different approach. Polat, Norcia, and Sagi [35] present psychophysical and neurophysiological evidence for facilitation and suppression of responses beyond classically defined receptive fields, proposing that long-range lateral interactions could be responsible for such phenomena. They also find evidence for Gestalt grouping of these effects. Sabatini [36] analyzes the effects of medium-range clustered connections in an orientation map mathematically and shows that they give rise to Gabor-like receptive fields observed in the visual cortex. Somers et el. [40] present a computational simulation showing how fixed-strength lateral connections can either facilitate or suppress a group of neurons depending on their level of activation. They also present experimental evidence for such gain control, and postulate that it could play a role in perceptual filling-in and discrimination phenomena.
The next three chapters discuss emergent effects of lateral connections. First Usher, Stemmler, and Niebur [42] show how lateral connections may establish oscillations at the level of local field variables even though single cell recordings do not show oscillatory activity. They go on to suggest that such connections may play an essential role in binding and grouping, pop-out, line completetion, and tilt-illusion phenomena. Next, Wang [45] shows how neural oscillators can achieve global synchronization through only local lateral connections, and how such synchronization could be used to perform perceptual grouping and scene segmentation. Third, Taylor and Alavi [41] propose two solutions to the problem of how long-range competition could arise without long-range inhibitory connections. These lateral inhibitory nets include nuclei of thalamus and basal ganglia, and could also take part in generating attention and active memory.
The third part of the book addresses the role of lateral interactions in the development of cortical structures. Sirosh, Miikkulainen, and Bednar [39] present simulations where lateral connections self-organize synergetically with cortical orientation columns, and also mediate reorganization of receptive fields in response to lesions. Lateral connections are shown to decorrelate cortical activity patterns, and thereby play a central role in forming a sparse and redundancy-reduced representation of visual input. Dong [10] analyzes the decorrelating role of lateral interactions further, and develops a general theory of development and adaptation in laterally connected networks. This theory can account for the development of orientation selectivity as well as other phenomena such as adaptation to color.
The final part explores lateral interactions as a substrate for high-level visual processing. Marshall and Alley's [31] laterally connected model learns to represent depth based on motion sequences where one object becomes temporarily occluded by another. Wiskott and von der Malsburg [47] show how face recognition could take place through a dynamic mapping of the image to a set of models. The topological constraints that define a match are implemented through lateral interactions. Finally, Edelman [11] shows how recognizing 3-D objects from different viewpoints could be mediated by lateral connections between view-specific representations. Concluding the book, Edelman surveys general computational reasons for lateral interactions, underlining the power of laterally coupled mechanisms over feedforward processes.
The chapters of this book thereby address, at some level, all the issues brought up in the previous section. Although the answers are by no means final, the book serves as a survey of the state of the art, as well as a foundation for further computational explorations on lateral interactions in the cortex.