Flexible sensorimotor computations through rapid reconfiguration of cortical dynamics

Abstract

SummarySensorimotor computations can be flexibly adjusted according to internal states and contextual inputs. The mechanisms supporting this flexibility are not understood. Here, we tested the utility of a dynamical system perspective to approach this problem. In a dynamical system whose state is determined by interactions among neurons, computations can be rapidly and flexibly reconfigured by controlling the system‘s inputs and initial conditions. To investigate whether the brain employs such control strategies, we recorded from the dorsomedial frontal cortex (DMFC) of monkeys trained to measure time intervals and subsequently produce timed motor responses according to multiple context-specific stimulus-response rules. Analysis of the geometry of neural states revealed a control mechanism that relied on the system‘s inputs and initial conditions. A tonic input specified by the behavioral context adjusted firing rates throughout each trial, while the dynamics in the measurement epoch allowed the system to establish initial conditions for the ensuing production epoch. This initial condition in turn set the speed of neural dynamics in the production epoch allowing the animal to aim for the target interval. These results provide evidence that the language of dynamical systems can be used to parsimoniously link brain activity to sensorimotor computations.

Bridging large-scale neuronal recordings and large-scale network models using dimensionality reduction

“…This approach was recently used to study how the brain flexibly controls the timing of behavior [48,49]. By applying dimensionality reduction to neuronal activity recorded from medial frontal cortex, Wang et al found that population activity time courses for different time intervals followed a stereotypical path, but traversed that path at different speeds ( Fig.…”

Neural state space alignment for magnitude generalisation in humans and recurrent networks

“…One view is that conceptual knowledge relies on neural ensembles that code for relations among stimuli but are invariant to their physical properties [5][6][7][8][9][10][11][12][13] . Recent evidence hints that when sets of stimuli share relational structure across contexts, they are embedded on parallel low-dimensional neural manifolds, so that a linear decoder learned in one context can be readily repurposed for another [14][15][16][17][18] . By aligning neural state spaces between contexts in this way, one can generalise relational knowledge, for example applying a criterion that distinguishes fast and slow animals to discriminate fast and slow vehicles, such as space rockets and bicycles (Fig.…”

Statistical structure of the trial-to-trial timing variability in synfire chains

“…For each pool in the chain, we have a single readout neuron, which receives synaptic input from all neurons of that pool along with a white Gaussian noise input with mean zero and autocorrelation σ readout δ(t − t ) ( Figure 1a). The dynamics of membrane potential and synaptic inputs for the readout neurons are similar to that in (18) and (19) with the appropriate inputs.…”

“…As behavior is controlled by the nervous system, it is natural to look for the source of some of this variability in the variable activity of neural circuits involved in the production of behavior [2,[4][5][6][7]. Indeed, the neural mechanisms underlying behavioral timing have been extensively studied experimentally [2,[8][9][10][11][12], establishing links between temporal variations of behavior and that of neural activity [13][14][15][16][17][18][19][20]. This neural variability could result from multiple sources such as stochastic events at the level of ion channels [21], synapses [22], and neurons [23]; chaotic activity of neural networks [24,25]; and sensory inference errors [5].…”