Fig. 1: Blowfly LMC responses are better explained by fitness than information maximization coding schemes.

a, Responses measured from the LMCs (black dots) and the CDF (blue line) of contrasts in the natural environment of the blowfly. If accurate perception of the environment is maximized by the LMCs, then the line indicating the CDF should lie on top of the dots reflecting the empirical data. The data points were averaged from n = 6 cells; the range bars show the total scatter (data of the LMC responses was reproduced from ref. ³⁶). b, The black dots represent the same empirical data and the blue line the same contrast stimulus CDF as in a. The grey dashed line represents the predicted response function from an accuracy maximization code. The orange line indicates a coding rule that maximizes fitness. It matches the data better than the grey line. The data points were averaged from n = 6 cells; the range bars show the total scatter (data of the LMC responses was reproduced from ref. ³⁶, work distributed with license CC BY-NC-ND 3.0). c, Neural response probability density distributions predicted by a fitness maximization rule (orange) also align better with the empirical data (black) than those predicted by infomax coding (dashed grey). This suggests that the fitness maximization model describes the empirical data more accurately than the accuracy maximization model. The same fitness-maximizing solution emerges when studying the L_p reconstruction error penalty, with optimal solution p = 0.5, which is the error penalty that best describes the LCM neural response data⁴⁰.