Geometry of neural computation unifies working memory and planning

Abstract

Real-world tasks require coordination of working memory, decision-making, and planning, yet these cognitive functions have disproportionately been studied as independent modular processes in the brain. Here, we propose that contingency representations, defined as mappings for how future behaviors depend on upcoming events, can unify working memory and planning computations. We designed a task capable of disambiguating distinct types of representations. In task-optimized recurrent neural networks, we investigated possible circuit mechanisms for contingency representations and found that these representations can explain neurophysiological observations from the prefrontal cortex during working memory tasks. Our experiments revealed that human behavior is consistent with contingency representations and not with traditional sensory models of working memory. Finally, we generated falsifiable predictions for neural data to identify contingency representations in neural data and to dissociate different models of working memory. Our findings characterize a neural representational strategy that can unify working memory, planning, and context-dependent decision-making.

The representational geometry of cognitive maps under dynamic cognitive control

“…Based on empirical and theoretical evidence that task demands during learning might shape a task space’s subjective representation (Barsalou, 1983; Bernardi et al, 2020; Ehrlich & Murray, 2022; Ho et al, 2022; Mack et al, 2016; Nieh et al, 2021; Tang et al, 2019), we considered to what extent cognitive control demands might influence the relational configuration between entities and hence the representational geometry of the 2D cognitive map. Specifically, we test the hypothesis that increasing control demands for incongruent compared to congruent pairs produces warping along the congruent axis in the subjective representation of the 2D social hierarchy.…”

“…Recent computational models of cognitive control suggest that the brain trades off the benefits of low-dimensional representations for fast learning and generalization across multiple tasks against high-dimensional representations for flexible decision making ( Fig. 1a ) (Badre et al, 2020; Bernardi et al, 2020; Buschman, 2021; Ehrlich & Murray, 2022; Fusi et al, 2016; Garner & Dux, 2022; Musslick & Cohen, 2021; Rigotti et al, 2013). An intriguing but unexplored possibility is that an analogous tradeoff between generalized and flexible representations may determine the geometry of cognitive maps defined by multiple attributes.…”

“…Recent findings suggest that the representational geometry of task representations might be sculpted by task demands to facilitate generalization to similar contexts that share a decision policy (i.e. choice rule or mapping from entities to decision values) (Barsalou, 1983; Bernardi et al, 2020; Ehrlich & Murray, 2022; Ho et al, 2022; Mack et al, 2016; Nieh et al, 2021; Tang et al, 2019). If control demands resulting from different levels of congruence play a role in shaping the resulting subjective representations, we would expect the brain to adopt a versatile representational strategy whereby congruence promotes sharing, and therefore generalization, across dimensions of the task, while incongruence promotes segregation, thereby allowing greater flexibility across contexts or tasks with divergent responses.…”

Cue-specific neuronal ensembles span intermittent rate coding of working memory

“…Working memory allows us to retain and manipulate information on short timescales and it is central to complex cognitive processing and adaptive behavior (Baddeley, 2000; Ehrlich & Murray, 2022). Foundational work in the 1970s revealed that working memory is associated with sustained neuronal spiking activity in primate prefrontal cortex (Fuster & Alexander, 1971).…”

“…Working memory allows us to retain and manipulate information on short timescales and it is central to complex cognitive processing and adaptive behavior (Baddeley, 2000;Ehrlich & Murray, 2022).…”