Skip to content

Advertisement

  • Poster presentation
  • Open Access

Identification of nonlinear-nonlinear neuron models and stimulus decoding

BMC Neuroscience201314 (Suppl 1) :P367

https://doi.org/10.1186/1471-2202-14-S1-P367

  • Published:

Keywords

  • Spike Train
  • Action Potential Generation
  • Conditional Phase
  • Volterra Series
  • Phase Response Curve

The majority of neural encoding models employed today consist of a linear feature-selection stage followed by a static memoryless nonlinearity generating the neuronal rate of response. Although such linear-nonlinear (LN) models have been proven useful in characterizing computation in many neurons (see [1] and references therein), they exclude the biophysics of action potential generation and, with a few exceptions [24], are limited to describing only linear computations. A large body of evidence [5] suggests that various synaptic microcircuits and complex dendritic morphologies exhibit a rich repertoire of nonlinear computations on a single neuron level. Importantly, such computations are observed not only for individual inputs [2, 3] but also between inputs, as manifested by multiplicative interactions between them [5]. At the same time, the highly nonlinear nature of action potential generation has been shown to adversely affect many existing identification methodologies and typically presents itself as artifactual stimulus processing [3, 6].

Here we investigate nonlinear-nonlinear neuron models that combine both biophysics and dendritic computation. Specifically, we consider models in which the nonlinear dendritic processing of input stimuli is described by a Volterra series [4] and the aggregate dendritic current is encoded into a sequence of action potentials by a biophysical conductance-based model. We first present a novel experimental approach for characterizing the biophysical spike generator using conditional phase response curves [7]. We then prove that only a projection of Volterra kernels onto the space of input signals can be identified and present an efficient algorithm for estimating these projections. The identification is robust in the presence of noise, scales favorably with the order of nonlinear interactions and affords high-resolution identification due to a basis decomposition of the projections. Furthermore, either synthetic or naturalistic stimuli can be used to identify the model. Finally, we consider a nonlinear-nonlinear population encoding scheme and construct a nonlinear decoder that is capable of reconstructing the original stimulus directly from spike times. This should be contrasted with linear decoders applied to firing rates of LN models [8, 9]. To the best of our knowledge, this is the first demonstration of a nonlinear decoder capable of high-fidelity reconstruction from a single multidimensional spike train in the nonlinear-nonlinear setting.

Declarations

Acknowledgements

The research reported here was supported in part by AFOSR under grant #FA9550-12-1-0232 and in part by NIH under grant #R021 DC012440001.

Authors’ Affiliations

(1)
Department of Electrical Engineering, Columbia University, New York, NY 10027, USA

References

  1. Wu MC-K, David SV, Gallant JL: Complete functional characterization of sensory neurons by system identification. Annual Reviews of Neuroscience. 2006, 477-505. 29Google Scholar
  2. Sharpee T, Rust NC, Bialek W: Analyzing Neural Responses to Natural Signals: Maximally Informative Dimensions. Neural Computation. 2004, 223-250. 16Google Scholar
  3. Schwartz O, Pillow JW, Rust NC, Simoncelli E: Spike-triggered neural characterization. Journal of Vision. 2006, 6: 484-507.View ArticlePubMedGoogle Scholar
  4. Marmarelis VZ: Nonlinear Dynamic Modeling of Physiological Systems. Wiley-IEEE Press. 2004Google Scholar
  5. Koch C: Biophysics of Computation: Information Processing in Single Neurons. Oxford University Press. 1999Google Scholar
  6. Jonathan Pillow, Eero Simoncelli: Biases in white noise analysis due to non-Poisson spike generation. Neurocomputing. 2003, 109-115. 52-54Google Scholar
  7. Kim AJ, Lazar AA: Recovery of Stimuli Encoded with a Hodgkin-Huxley Neuron Using Conditional PRCs. Phase Response Curves in Neuroscience. 2011, SpringerGoogle Scholar
  8. Stanley GB, Li FF, Dan Y: Reconstruction of natural scenes from ensemble responses in the lateral geniculate nucleus. J Neurosci. 1999, 8036-8042. 19Google Scholar
  9. Warland DK, Reinagel P, Meister M: Decoding visual information from a population of retinal ganglion cells. J Neurophysiol. 1997, 78 (5): 2336-2350.PubMedGoogle Scholar

Copyright

Advertisement