A note on minimum-variance theory and beyond

Abstract

We revisit the minimum-variance theory proposed by Harris and Wolpert (1998 Nature 394 780–4), discuss the implications of the theory on modelling the firing patterns of single neurons and analytically find the optimal control signals, trajectories and velocities. Under the rate coding assumption, input control signals employed in the minimum-variance theory should be Fitts processes rather than Poisson processes. Only if information is coded by interspike intervals, Poisson processes are in agreement with the inputs employed in the minimum-variance theory. For the integrate-and-fire model with Fitts process inputs, interspike intervals of efferent spike trains are very irregular. We introduce diffusion approximations to approximate neural models with renewal process inputs and present theoretical results on calculating moments of interspike intervals of the integrate-and-fire model. Results in Feng, et al (2002 J. Phys. A: Math. Gen. 35 7287–304) are generalized. In conclusion, we present a complete picture on the minimum-variance theory ranging from input control signals, to model outputs, and to its implications on modelling firing patterns of single neurons.