Biases in white noise analysis due to non-Poisson spike generation

Jonathan W. Pillow, Eero P. Simoncelli

Abstract

White noise analysis methods for receptive field characterization typically ignore the dynamics of neural spike generation (i.e., they assume Poisson spike generation). We show that a linear integrate-and-fire spike generation model can induce significant bias in the estimation of the linear kernel, and leads to reverse correlation estimates that depend on stimulus amplitude, as has been observed in experimental data. We develop a modified estimator for linear characterization of such neurons, and demonstrate its effectiveness in simulation. We also apply it to physiological data, demonstrating that the estimated linear kernel is stable with respect to changes in stimulus amplitude.
BibTeX
@article{pillow2003biases,
  author = {Jonathan W. Pillow and Eero P. Simoncelli},
  title = {{Biases in white noise analysis due to non-Poisson spike generation}},
  journal = {Neurocomputing},
  year = {2003},
  doi = {10.1016/s0925-2312(02)00822-6},
}

Loading related content...

Figure