Transformations of sensory information in the brain suggest changing criteria for optimality

Efficient coding in biophysically realistic excitatory-inhibitory spiking networks

The principle of efficient coding posits that sensory cortical networks are designed to encode maximal sensory information with minimal metabolic cost. Despite the major influence of efficient coding in neuroscience, it has remained unclear whether fundamental empirical properties of neural network activity can be explained solely based on this normative principle. Here, we derive the structural, coding, and biophysical properties of excitatory-inhibitory recurrent networks of spiking neurons that emerge directly from imposing that the network minimizes an instantaneous loss function and a time-averaged performance measure enacting efficient coding. We assumed that the network encodes a number of independent stimulus features varying with a time scale equal to the membrane time constant of excitatory and inhibitory neurons. The optimal network has biologically plausible biophysical features, including realistic integrate-and-fire spiking dynamics, spike-triggered adaptation, and a non-specific excitatory external input. The excitatory-inhibitory recurrent connectivity between neurons with similar stimulus tuning implements feature-specific competition, similar to that recently found in visual cortex. Networks with unstructured connectivity cannot reach comparable levels of coding efficiency. The optimal ratio of excitatory vs inhibitory neurons and the ratio of mean inhibitory-to-inhibitory vs excitatory-to-inhibitory connectivity are comparable to those of cortical sensory networks. The efficient network solution exhibits an instantaneous balance between excitation and inhibition. The network can perform efficient coding even when external stimuli vary over multiple time scales. Together, these results suggest that key properties of biological neural networks may be accounted for by efficient coding.

Efficient coding in biophysically realistic excitatory-inhibitory spiking networks

The principle of efficient coding posits that sensory cortical networks are designed to encode maximal sensory information with minimal metabolic cost. Despite the major influence of efficient coding in neuroscience, it has remained unclear whether fundamental empirical properties of neural network activity can be explained solely based on this normative principle. Here, we derive the structural, coding, and biophysical properties of excitatory-inhibitory recurrent networks of spiking neurons that emerge directly from imposing that the network minimizes an instantaneous loss function and a time-averaged performance measure enacting efficient coding. We assumed that the network encodes a number of independent stimulus features varying with a time scale equal to the membrane time constant of excitatory and inhibitory neurons. The optimal network has biologically-plausible biophysical features, including realistic integrate-and-fire spiking dynamics, spike-triggered adaptation, and a non-specific excitatory external input. The excitatory-inhibitory recurrent connectivity between neurons with similar stimulus tuning implements feature-specific competition, similar to that recently found in visual cortex. Networks with unstructured connectivity cannot reach comparable levels of coding efficiency. The optimal ratio of excitatory vs inhibitory neurons and the ratio of mean inhibitory-to-inhibitory vs excitatory-to-inhibitory connectivity are comparable to those of cortical sensory networks. The efficient network solution exhibits an instantaneous balance between excitation and inhibition. The network can perform efficient coding even when external stimuli vary over multiple time scales. Together, these results suggest that key properties of biological neural networks may be accounted for by efficient coding.

Orthogonalization of spontaneous and stimulus-driven activity by hierarchical neocortical areal network in primates

How biological neural networks reliably process information in the presence of spontaneous activity remains controversial. In mouse primary visual cortex (V1), stimulus-evoked and spontaneous activity show orthogonal (dissimilar) patterns, which is advantageous for separating sensory signals from internal noise. However, studies in carnivore and primate V1, which have functional columns, have reported high similarity between stimulus-evoked and spontaneous activity. Thus, the mechanism of signal-noise separation in the columnar visual cortex may be different from that in rodents. To address this issue, we compared spontaneous and stimulus-evoked activity in marmoset V1 and higher visual areas. In marmoset V1, spontaneous and stimulus-evoked activity showed similar patterns as expected. However, in marmoset higher visual areas, spontaneous and stimulus-evoked activity were progressively orthogonalized along the cortical hierarchy, eventually reaching levels comparable to those in mouse V1. These results suggest that orthogonalization of spontaneous and stimulus-evoked activity is a general principle of cortical computation.

Measuring Stimulus Information Transfer Between Neural Populations through the Communication Subspace

Sensory processing arises from the communication between neural populations across multiple brain areas. While the widespread presence of neural response variability shared throughout a neural population limits the amount of stimulus-related information those populations can accurately represent, how this variability affects the interareal communication of sensory information is unknown. We propose a mathematical framework to understand the impact of neural population response variability on sensory information transmission. We combine linear Fisher information, a metric connecting stimulus representation and variability, with the framework of communication subspaces, which suggests that functional mappings between cortical populations are low-dimensional relative to the space of population activity patterns. From this, we partition Fisher information depending on the alignment between the population covariance and the mean tuning direction projected onto the communication subspace or its orthogonal complement. We provide mathematical and numerical analyses of our proposed decomposition of Fisher information and examine theoretical scenarios that demonstrate how to leverage communication subspaces for flexible routing and gating of stimulus information. This work will provide researchers investigating interareal communication with a theoretical lens through which to understand sensory information transmission and guide experimental design.

A Bayesian observer model reveals a prior for natural daylights in hue perception

Incorporating statistical characteristics of stimuli in perceptual processing can be highly beneficial for reliable estimation from noisy sensory measurements but may generate perceptual bias. According to Bayesian inference, perceptual biases arise from the integration of internal priors with noisy sensory inputs. In this study, we used a Bayesian observer model to derive biases and priors in hue perception based on discrimination data for hue ensembles with varying levels of chromatic noise. Our results showed that discrimination thresholds for isoluminant stimuli with hue defined by azimuth angle in cone-opponent color space exhibited a bimodal pattern, with lowest thresholds near a non-cardinal blue-yellow axis that aligns closely with the variation of natural daylights. Perceptual biases showed zero crossings around this axis, indicating repulsion away from yellow and attraction towards blue. These biases could be explained by the Bayesian observer model through a non-uniform prior with a preference for blue. Our findings suggest that visual processing takes advantage of knowledge of the distribution of colors in natural environments for hue perception.