A general decoding strategy explains the relationship between behavior and correlated variability

Abstract

Increases in perceptual performance correspond to decreases in the correlated variability of sensory neuron responses. No sensory information decoding mechanism has yet explained this relationship. We hypothesize that when observers must respond to a stimulus change of any magnitude, decoders prioritize generality: a single set of neuronal weights to decode any stimulus response. Our mechanistic circuit model supports that a general decoding strategy explains the inverse relationship between perceptual performance and V4 correlated variability observed in two rhesus monkeys performing a visual attention task. Further, based on the recorded V4 population responses, a monkey’s decoding mechanism was more closely matched the more broad the range of stimulus changes used to compute a sensory information decoder. These results support that observers use a general sensory information decoding strategy based on a single set of decoding weights, capable of decoding neuronal responses to the wide variety of stimuli encountered in natural vision.

Not so spontaneous: Multi-dimensional representations of behaviors and context in sensory areas

“…This stimulus response modulation observed across various studies is a form of gain modulation that multiplicatively changes all stimulus responses, rather than specifically enhancing the responses to certain stimuli (Ferguson and Cardin, 2020;Figures 2A and 2B). Gain modulation in many cases does not impair linear decoding of sensory stimuli but instead may increase the overall sensory information available in downstream decision-making areas (Abdolrahmani et al, 2021;Allen et al, 2019;Engel et al, 2016;Kanashiro et al, 2017;Ni et al, 2021;Stringer et al, 2021).…”

Adaptive coding across visual features during free-viewing and fixation conditions

“…Additionally, analysis of iso-feature adaptation data (adapter and test stimuli belonging to the same feature), revealed similar qualitative changes in tuning strength, discriminability and population response variance (Supplementary Figure 11). This indicates that decorrelation could be a general mechanism at the population level to efficiently encode a wide variety of sensory stimuli after adaptation 28…”

“…Additionally, analysis of iso-feature adaptation data (adapter and test stimuli belonging to the same feature), revealed similar qualitative changes in tuning strength, discriminability and population response variance (Supplementary Figure 11). This indicates that decorrelation could be a general mechanism at the population level to efficiently encode a wide variety of sensory stimuli after adaptation 28 Surprisingly, there is a complete lack of studies on either rapid or prolonged adaptation in the cross-feature domain with respect to changes in stimulus discriminability. Previous studies have revealed cross-orientation 29 and cross-sensory interactions 30,31 , however the role of such interactions on stimulus discriminability has not been explored.…”

The impact of functional correlations on task information coding

“…These studies revealed that during task and attentional states, correlations are reduced among pairs of neurons (Cohen and Maunsell, 2009), cortical regions accessed with wide-field calcium imaging data (Pinto et al, 2019), mean-field multi-unit recording across cortical regions in non-human primates (Ito et al, 2020), and human fMRI correlations (Ito et al, 2020). While prior literature has demonstrated that spike count correlations impact information coding in non-human primates (Cohen and Maunsell, 2009;Ni et al, 2022), it was an open question as to whether these intuitions would scale to larger spatial organizations and broader cognitive tuning curves. Our findings affirm that the generic statistical principles developed to understand neural coding in spiking units are translatable to different data modalities, and naturally scale up to broader spatial and cognitive levels of organization.…”