Raglio, S. | Search bioRxiv

bioRxiv2024

Learning to infer transitively: serial ordering on a mental line in premotor cortex

Transitive inference (TI) is a form of deductive reasoning that allows to infer unknown relationships among premises. It is hypothesized that this cognitive task is accomplished by mapping stimuli onto a linear workspace, referred to as the mental line, based on their arbitrarily assigned ranks. However, open questions remain: does this mental line have a neural correlate, and if so, where and how is it represented and learned in the brain? In this study, we investigate the role of monkeys dorsal premotor cortex (PMd) in encoding the hypothesized mental line during the acquisition of item relationships. Our findings provide evidence that the TI task can be solved through a linear transformation of the neural representations of arbitrarily ranked items. We show that PMd multi-unit activity organizes along a theoretically informed direction, implementing a geometrical solution that effectively explains animal behavior. Our results suggest that the premotor cortex plays a crucial role in integrating item representations into a geometric mental line, where the symbolic distance (i.e., rank difference) between items influences the related motor decisions. Furthermore, we observe an ongoing learning process characterized by a rotation of this mental line, which aligns to the linear manifold where motor plan unfolds. This elucidates a cortical optimization strategy based on the statistical structure of the task.

animal behavior and cognition↗

bioRxiv2023

Ranking and serial thinking: A geometric solution

A general mathematical description of the way the brain encodes ordinal knowledge of sequences is still lacking. Coherently with the well-established idea of mixed selectivity in high-dimensional state spaces, we conjectured the existence of a linear solution for serial learning tasks. In this theoretical framework, the neural representation of the items in a sequence are read out as ordered projections along a suited "geometric" mental line learned via classical conditioning (delta rule learning). We show that the derived model explains all the behavioral effects observed in humans and other animal species performing the transitive inference task in presence of noisy sensory information and stochastic neural activity. This result is generalized to the case of recurrent neural networks performing motor decision, where the same geometric mental line is learned showing a tight correlation with the motor plan of the responses. Network activity is then eventually modulated according to the symbolic distance of presented item pairs, as observed in associative cortices of nonhuman primates. Serial ordering is thus predicted to emerge as a linear mapping between sensory input and behavioral output, highlighting a possible pivotal role of motor-related associative cortices in the transitive inference task.

neuroscience↗