Abstract
Most psychometric models are defined in terms of free parameters which lack a clear neurobiological interpretation. This limits the ability of most psychophysical experiments to speak directly to issues of neural encoding mechanisms, or questions of how neural responses are decoded to guide behavior. In this study, we introduce a general computational methodology for defining a psychometric model in terms of an assumed underlying neural encoding model and neural decoding model. We demonstrate that this methodology can be used to define a psychometric model whose free parameters are those defining the neural encoding and decoding models, and which does not depend on unobserved neuronal responses. We apply our method to the problem of estimating the parameters of the neural contrast gain function for a hypothetical population of orientation-tuned V1 neurons. By fitting a psychometric function derived using our method to psychophysical data obtained from an orientation-discrimination experiment, we accurately estimate neural contrast gain function parameters (half-saturation, shape) in the known physiological range. Furthermore, we demonstrate that it is possible to use psychophysical data to distinguish between two qualitatively similar (but quantitatively different) candidate models of neural contrast gain. We show that this process of model comparison is greatly aided by adaptive stimulus generation methods, where a stimulus optimized for discriminating competing models is generated during the course of the experimental session based on the best fit of each model to data collected earlier in the session. We suggest that our methodology may in many cases permit psychophysical experiments to more directly inform and guide neurophysiological investigations.
Meeting abstract presented at VSS 2016