Dynamic Cue Combination in Distributional Population Code Networks

Rama Natarajan, Richard S. ZemelOxford 2011 · DOI · OpenAlex

Abstract

This chapter considers how a neural system could properly combine cues even as their reliabilities change dynamically and continuously in time. It describes a framework for characterizing how uncertain information may be represented in population codes, and how the representation may be optimized to facilitate optimal decoding. It then motivates and presents a coding scheme within this framework that applies to time-varying stimulus variables. Next, it describes a hierarchical network setup that utilizes this coding approach, and demonstrates an application to a dynamic cue-combination task—a novel adaptation of a sensorimotor task. Specifically, it considers how a neural population might recursively integrate dynamic inputs from multiple sensory modalities with learned prior information, to determine appropriate motor commands that control behavior; the results of simulated experiments with this network are presented. Finally, the implications of this scheme and suggest some future directions are discussed.
BibTeX
@inproceedings{natarajan2011dynamic,
  author = {Rama Natarajan and Richard S. Zemel},
  title = {{Dynamic Cue Combination in Distributional Population Code Networks}},
  booktitle = {Oxford},
  year = {2011},
  doi = {10.1093/acprof:oso/9780195387247.003.0020},
}

Loading related content...

Figure