Using Locally Competitive Algorithms to Model Top-Down and Lateral Interactions between Cortical Neurons

Cortical connections consist of feedforward, feedback and lateral pathways. However, most functional models of visual cortex only account for feedforward connections. Additionally, most models of visual cortex fail to account both for the thalamic projections to non-striate areas and the reciprocal connections from extrastriate areas back to the thalamus. In this talk, I will describe how a modified Locally Competitive Algorithm (LCA; Rozell et al, Neural Comp, 2008) can be used as a unifying framework for exploring the role of top-down, lateral and poly-thalamocortical pathways within the context of deep, sparse, generative models. LCA is expressed entirely in terms of simple elements all performing the same local input-output operations and is thus amenable to massively parallel implementation in custom hardware. We project that algorithms based on LCA will achieve orders of magnitude reductions in size, weight and power (SWaP). We are currently applying LCA to the problem of object detection and tracking in aerial video with the goal of enabling unmanned and other mobile platforms to achieve state-of-the-art performance on difficult pattern-recognition tasks in environments where limits on SWaP represent the dominant constraints.
Garrett Kenyon received his PhD in Physics from the University of Washington in 1990 but has worked primarily as a computational neuroscientist throughout his career. He is presently a staff member at the Los Alamos National Laboratory. His current research interests include large-scale simulations of retinal and cortical circuits, development of open-source high-performance neural simulation tools (PetaVision), and biologically-inspired distributed sensor systems.