A neural circuit mechanism of categorical perception: top-down signaling in the primate cortex

Abstract

In contrast to feedforward architecture commonly used in deep networks at the core of today's AI revolution, the biological cortex is endowed with an abundance of feedback projections. Feedback signaling is often difficult to differentially identify, and its computational roles remain poorly understood. Here, we investigated a cognitive phenomenon, called categorical perception (CP), that reveals the influences of high-level category learning on low-level feature-based perception, as a putative signature of top-down signaling. By examining behavioral data from a visual motion delayed matching experiment in non-human primates, we found that, after categorization training, motion directions closer to (respectively, away from) a category center became more (less) difficult to discriminate. This distance-dependent discrimination performance change along the dimension relevant to the learned categories provides direct evidence for the CP phenomenon.To explain this experimental finding, we developed a neural circuit model that incorporated key neurophysiological findings in visual categorization, working memory and decision making. Our model accounts for the behavioral data indicative of CP, pinpoints its circuit basis, suggests novel experimentally testable predictions and provides a functional explanation for its existence. Our work shows that delayed matching paradigms in non-human primates combined with biologicallybased modeling can serve as a promising model system for elucidating the neural mechanisms of CP, as a manifestation of top-down signaling in the cortex. 2 Significant StatementCategorical perception is a cognitive phenomenon revealing the influences of high-level category learning on low-level feature-based perception. However, its underlying neural mechanisms are largely unknown. Here, we found behavioral evidence for this phenomenon from a visual motion delayed matching experiment in non-human primates. We developed a neural circuit model that can account for this behavioral data, pinpoints its circuit basis, suggests novel experimentally testable predictions and provides a functional explanation for its existence. Our work shows that delayed matching paradigms in non-human primates combined with biologically-based modeling can serve as a promising model system for elucidating the neural mechanisms of categorical perception, as a manifestation of top-down signaling in the cortex.3

Confidence-controlled Hebbian learning efficiently extracts category membership from stimuli encoded in view of a categorization task

“…The adaptation of the coding layer has also been studied computationally [Bonnasse-Gahot and Nadal, 2008, Engel et al, 2015, Tajima et al, 2016, Min et al, 2020. Whereas most models of decision-making consider an uniform coding of the stimulus before the decision part [Beck et al, 2008, Drugowitsch et al, 2019, few models analyse the nature of a stimulus coding layer optimized in view of a categorization task [Bonnasse-Gahot and Nadal, 2008].…”

“…Authors have proposed an alternative type of model, assuming a top-down modulation of the decision layer onto the coding layer [Tajima et al, 2016[Tajima et al, , 2017. To obtain the adaptation of the tuning curves, authors use a top-down modulation under the form of a reward-modulated Hebbian learning [Engel et al, 2015, Min et al, 2020. However, the efficiency of the resulting neural coding has not been discussed.…”

“…However, such rule requires to keep track of the rewards associated to the previous trials to generate the prediction error. RMHL has in particular be used in models of categorical perception to adapt the weights between the encoding and the decision layers [Engel et al, 2015, Min et al, 2020.…”

“…As a bio-physically plausible Hebbian learning rule [Hebb, 1949, Miller andMacKay, 1994] modulated by a reward signal [Schultz et al, 1997, Loewenstein and Seung, 2006, Reward-modulated Hebbian learning (RMHL) [Legenstein et al, 2008[Legenstein et al, , 2010 has been proposed and used in various contexts. RMHL has in particular be used in models of categorical perception to adapt the weights between the encoding and the decision layers [Engel et al, 2015, Min et al, 2020. However, such rule requires to keep track of the rewards associated to the previous trials to generate the prediction error.…”

Across-area synchronization supports feature integration in working memory

“…Our multi-area model adds to a large body of computational work (Min et al, 2020;Froudist-Walsh et al, 2020;Mejias and Wang, 2019;Engel and Wang, 2011;Ardid et al, 2007;Edin et al, 2009;Murray et al, 2017;Novikov et al, 2021;Ardid et al, 2010;Bouchacourt and Buschman, 2019) attempting to account for the distributed nature of working memory (Christophel et al, 2017). While several of these models have implemented across-area interactions through oscillatory dynamics (Ardid et al, 2010;Novikov et al, 2021), they did not attribute a clear mechanistic role to inter-area synchronization dynamics.…”