Neural codes in early sensory areas maximize fitness

Abstract

It has generally been presumed that sensory information encoded by a nervous system should be as accurate as its biological limitations allow. However, perhaps counter intuitively, accurate representations of sensory signals do not necessarily maximize the organism's chances of survival. To test this hypothesis, we developed a unified normative framework for fitness-maximizing encoding by combining theoretical insights from neuroscience, computer science, and economics. Initially, we applied predictions of this model to neural responses from large monopolar cells (LMCs) in the blowfly retina. We found that neural codes that maximize reward expectation---and not accurate sensory representations---account for retinal LMC activity. We also conducted experiments in humans and find that early sensory areas flexibly adopt neural codes that promote fitness maximization in a retinotopically-specific manner, which impacted decision behavior. Thus, our results provide evidence that fitness-maximizing rules imposed by the environment are applied at the earliest stages of sensory processing.

Bias and variance of the Bayesian-mean decoder

“…( 16)). This prediction is consistent with that of Wei and Stocker [1], who also predict a relation of proportionality, although only when the objective is to maximize the mutual information, which is not the objective in many tasks (moreover, actual encodings do not seem to be always consistent with this objective, even in early sensory areas [27]). Wei and Stocker call the relation a "law of human perception", and find strong empirical support for it; but the proportionality constant is left unspecified.…”

“…Thus in all the tasks mentioned above, the optimal Fisher information is proportional to a power of the prior (I(x) ∝ π γ (x)), i.e., efficient coding allocates greater precision to the more frequent stimuli, although the exact allocation depends on the specifics of the objective function [22,23,[25][26][27].…”

Efficient numerosity estimation under limited time

“…Taken together, our findings suggest the fruitfulness of studying optimal models with resource limitations, which can serve as a departing point to understand the neurocomputational mechanisms underlying human behaviour without ignoring the fact that biological systems are limited in their capacity to process information (34)(35)(36)(37). This highlights that understanding behavior in terms of its objectives while taking into account cognitive limitations, alongside encoding, decoding, and inference processes is likely to be essential to elucidate the mechanisms underlying human cognition.…”

Individual risk attitudes arise from noise in neurocognitive magnitude representations

“…However, the current form of the NLC is actually agnostic to whether choices are information-maximising (i.e., perceptual choice) or maximising expected reward or payoff (i.e., economic choice). Previous results have suggested that models can be set up to explicitly dissociate these two behavioural goals (Heng et al, 2020; Park and Pillow, 2020; Rustichini et al, 2017; Schaffner et al, 2021). However, while the NLC may not be as computationally detailed as these more recent optimal coding models, our approach establishes it as one of the few choice models for whom there is an empirical correspondence between model parameters and independent measures of neural processing.…”

“…However, the current form of the NLC is actually agnostic to whether choices are information-maximising (i.e., perceptual choice) or maximising expected reward or payoff (i.e., economic choice). Previous results have suggested that models can be set up to explicitly dissociate these two behavioural goals(Heng et al, 2020;Park and Pillow, 2020;Rustichini et al, 2017;Schaffner et al, 2021…”