Anisotropic connectivity implements motion-basedprediction in a spiking neural network
2013 (English)In: Frontiers in Computational Neuroscience, ISSN 1662-5188Article in journal (Refereed) Published
Predictive coding hypothesizes that the brain explicitly infers upcoming sensory inputto establish a coherent representation of the world. Although it is becoming generallyaccepted, it is not clear on which level spiking neural networks may implementpredictive coding and what function their connectivity may have. We present a networkmodel of conductance-based integrate-and-fire neurons inspired by the architectureof retinotopic cortical areas that assumes predictive coding is implemented throughnetwork connectivity, namely in the connection delays and in selectiveness for the tuningproperties of source and target cells. We show that the applied connection pattern leadsto motion-based prediction in an experiment tracking a moving dot. In contrast to ourproposed model, a network with random or isotropic connectivity fails to predict the pathwhen the moving dot disappears. Furthermore, we show that a simple linear decodingapproach is sufficient to transform neuronal spiking activity into a probabilistic estimatefor reading out the target trajectory.
Place, publisher, year, edition, pages
motion detection, motion extrapolation, probabilistic representation, predictive coding, network of spiking neurons, large-scale neuromorphic systems
Bioinformatics (Computational Biology) Neurosciences
IdentifiersURN: urn:nbn:se:kth:diva-136251DOI: 10.3389/fncom.2013.00112ISI: 000324633400001ScopusID: 2-s2.0-84884678746OAI: oai:DiVA.org:kth-136251DiVA: diva2:675641
QC 201401212013-12-042013-12-042015-05-04Bibliographically approved