Abstract
In sensory neuroscience, rate coding is the simplest and most common assumption; many psychophysical functions behave in a way that is roughly monotonic to the firing rates of some particular class of geniculate or cortical neuron. Somewhat less attention has been paid to the specific nature of the implied coding. An often cited example of rate coding in human color vision is that color naming seems to be predictable from power law transforms of specific geniculate neuron spike rates (De Valois et al., JOSA, 1966). More specifically, Young (JOSA, 1986) and Romney & DAndrade (PNAS, 2005) assert that a cubic power law maps LGN firing rate inputs to cortex into psychophysical responses. I probed this curious assertion for luminosity and for chromatic valence. (1) Modeling psychophysical luminance sensitivity from the firing rate of broadband LGN cells. Depending on which datasets are matched, this yields power laws with exponents between 2.68 and 3.05. (2) Modeling chromatic valences from P-cell firing rates is trickier (because of mutual opponency between responding mechanisms) but is straightforward for wavelengths away from the valence crosspoints/unique hues. Preliminary calculations with the r-g psychophysical channel and various P-cell datasets indicate power law exponents in the range of 2.45-3.03. The nearly cubic relationship between some neural spike rates and some psychophysical behaviors is likely a consequence of many cortical neurons Naka-Rushton expansive nonlinearities resembling cubic functions over a significant portion of their operating ranges. In contrast-transducing neurons the average Naka-Rushton exponent increases from 1.5 in LGN to 2.4 in V1 to 3.0 in mid-temporal cortex (Sclar et al., Vision Research, 1990). If wavelength-tuned neurons follow the same trend, a cubic function would be compatible with evidence of color-information processing along pathways that span V1 and temporal cortex (Conway, Visual Neuroscience, 2013).
Meeting abstract presented at VSS 2014