Effective learning is accompanied by high-dimensional and efficient representations of neural activity

Abstract

A fundamental cognitive process is the ability to map value and identity onto objects as we learn about them. Exactly how such mental constructs emerge and what kind of space best embeds this mapping remains incompletely understood. Here we develop tools to quantify the space and organization of such a mapping, thereby providing a framework for studying the geometric representations of neural responses as reflected in functional MRI. Considering how human subjects learn the values of novel objects, we show that quick learners have a higher dimensional geometric representation than slow learners, and hence more easily distinguishable whole-brain responses to objects of different value. Furthermore, we find that quick learners display a more compact embedding of their neural responses and hence have a higher ratio of their task-based dimension to their embedding dimension -consistent with a greater efficiency of cognitive coding. Lastly, we investigate the neurophysiological drivers of high dimensional patterns at both regional and voxel levels, and we complete our study with a complementary test of the distinguishability of associated whole-brain responses. Our results demonstrate a spatial organization of neural responses characteristic of learning, and offer a suite of geometric measures applicable to the study of efficient coding in higher-order cognitive processes more broadly.

Map making: Constructing, combining, and inferring on abstract cognitive maps

“…Another recent study (Tang et al, 2019) in which human participants learned the association between novel visual stimuli and reward values over multiple days used a linear support vector machine to show that the brains of fast learners were more likely to use an efficient coding scheme to represent stimuli. Their study suggested that multidimensional coding in the brain helps participants to discriminate different stimuli while simultaneously embedding the stimuli in an efficient low-dimensional task-relevant structure.…”

Rich and lazy learning of task representations in brains and neural networks

“…One recently popular theory proposes that stimulus and context signals are projected into a high-dimensional neural code, permitting linear decoding of exhaustive combinations of task variables [16]. Indeed many neurons, especially in prefrontal and parietal cortex, exhibit nonlinear mixed selectivity, multiplexing information over several potentially relevant task variables [17][18][19], with errors heralded by a collapse in dimensionality [17]. This highdimensional random mixed selectivity offers great behavioural flexibility because it maximises the potential for discrimination among diverse combinations of inputs, but also implies that neural codes should be relatively unstructured and task-agnostic.…”

Efficient Coding in the Economics of Human Brain Connectomics

“…Compression efficiency of hippocampal and sensory pathways should predict the speed, accuracy, and efficient cognitive coding of high-dimensional visuospatial stimuli in sensorimotor learning [25,19]. Such studies could illuminate how the representational structure of information drives the selective loss of redundant or core information in convolutional feedforward network models of sensorimotor information processing where triangular structural motifs of path transitivity resemble feedforward loops [13,19]. Lastly, our findings invite further development and application of well-studied information routing models and coding schemes to brain network communication [26,51,40].…”

“…In contrast to these models, the efficient coding hypothesis proposes that the brain represents information in a metabolically economical or compressed form by taking advantage of redundancy in the structure of information [18,2]. Coding efficiency characterizes low-dimensional neural representations and dynamics supporting cognition [19,20,21]. New models should therefore demonstrate metabolic and information transfer efficiency that predictably differ according to variation in brain network structure across the protracted development of structural connectivity [3,12,5,22,17,7,16].…”