Cellular Wave Computing for (artificial and natural) Visual Processing
Abstract: The Cellular Wave Computing paradigm is based on two roots: the standard Cellular Nonlinear/neural Network (CNN) Dynamics and the two fundamental brain related inspirations: laterality and receptive fields.
The elementary instruction of a Cellular Wave Computer is a spatial-temporal nonlinear wave. This is generated by a cellular dynamical system described by a cell array, the cells are placed on a grid, each cell is mainly locally connected. These dynamical system cells, as well as the local interaction patterns are programmable. The various elementary instructions are combined in space and time, fully parallel. The algorithm, described by a so called alpha-recursive function operating on topographic flows, opens new vistas in spatial-temporal computing and programming. Some specific versions of the related new computational infrastructure have been implemented in connection with various visual microprocessors. Special cases describing various neuronal constructs in the brain, mostly for the visual pathway, will be highlighted.
A few mission critical applications will be shown, such as surveillance on high speed scenes, and a wearable navigation device: the bionic eyeglass. As to the latter, it is a prototype application of a spatial-temporal logic with semantic embedding.
Two new directions will be highlighted: the Oscillatory CNN architecture for spintronic devices, and the spatial-temporal event detections without frame by frame analysis.
It is foreseen that cellular wave computing principles will spread in many areas of nanoscale technology, as well as in neuromorphic modeling of some parts of the brain. A new way of algorithmic thinking will prevail considering the algorithmic actions in space and time, establishing prototype physical and virtual cellular machines and related algorithmic techniques.