Prefrontal cortical contributions to working memory loading, maintenance and recall are parsed by hippocampal-prefrontal oscillatory assembly dynamics

Working memory enables incorporation of recent experience into subsequent decision-making. This processing recruits both prefrontal cortex and hippocampus, where neurons encode task cues, rules and outcomes. However, precisely which information is carried when, and by which neurons, remains unclear. Using population decoding of activity in rat medial prefrontal cortex (mPFC) and dorsal hippocampal CA1, we confirm that mPFC populations lead in maintaining sample information across delays of an operant non-match to sample task, despite individual neurons firing only transiently. During sample encoding, distinct mPFC subpopulations joined distributed CA1-mPFC cell assemblies hallmarked by 4-5Hz rhythmic modulation; CA1-mPFC assemblies re-emerged during choice episodes, but were not 4-5Hz modulated. Delay-dependent errors arose when attenuated rhythmic assembly activity heralded collapse of sustained mPFC encoding; pharmacological disruption of CA1-mPFC assembly rhythmicity impaired task performance. Our results map component processes of memory-guided decisions onto heterogeneous CA1-mPFC subpopulations and the dynamics of physiologically distinct, distributed cell assemblies.

Frontal but not parietal cortex is required for decisions under risk

Neurons in frontal and parietal cortex encode task variables during decision-making, but causal manipulations of the two regions produce strikingly different results. For example, silencing the posterior parietal cortex (PPC) in rats and monkeys produces minimal effects in perceptual decisions requiring integration of sensory evidence, but silencing frontal cortex profoundly impairs the same decisions. Here, we tested, for the first time, the causal roles of the rat frontal orienting field (FOF) and PPC in economic choice under risk. On each trial, rats chose between a lottery and a small but guaranteed surebet. The magnitude of the lottery was independently varied across trials and was indicated to the rat by the pitch of an auditory cue. As in perceptual decisions, both unilateral and bilateral PPC muscimol inactivations produced weak effects. FOF inactivations produced substantial changes in behavior even though our task had no working memory component. We quantified control and bilateral inactivation behavior with a multi-agent model consisting of a mixture of a 'rational' utility-maximizing agent (U=Vρ) with two `habitual' agents that either choose surebet or lottery. Silencing PPC produced no significant shifts in any parameters relative to controls. Effects of FOF silencing were best explained by a decrease in ρ, the exponent of the utility function. This effect was parsimoniously explained by a dynamical model where the FOF is part of network that performs sensory-to-value transformations.

A recurrent neural network framework for flexible and adaptive decision making based on sequence learning

The brain makes flexible and adaptive responses in a complicated and ever-changing environment for an organism’s survival. To achieve this, the brain needs to understand the contingencies between its sensory inputs, actions, and rewards. This is analogous to the statistical inference that has been extensively studied in the natural language processing field, where recent developments of recurrent neural networks have found many successes. We wonder whether these neural networks, the gated recurrent unit (GRU) networks in particular, reflect how the brain solves the contingency problem. Therefore, we build a GRU network framework inspired by the statistical learning approach of NLP and test it with four exemplar behavior tasks previously used in empirical studies. The network models are trained to predict future events based on past events, both comprising sensory, action, and reward events. We show the networks can successfully reproduce animal and human behavior. The networks generalize the training, perform Bayesian inference in novel conditions, and adapt their choices when event contingencies vary. Importantly, units in the network encode task variables and exhibit activity patterns that match previous neurophysiology findings. Our results suggest that the neural network approach based on statistical sequence learning may reflect the brain’s computational principle underlying flexible and adaptive behaviors and serve as a useful approach to understand the brain.

Fine Spike Timing in Hippocampal–Prefrontal Ensembles Predicts Poor Encoding and Underlies Behavioral Performance in Healthy and Malformed Brains

Spatial working memory (SWM) is a central cognitive process during which the hippocampus and prefrontal cortex (PFC) encode and maintain spatial information for subsequent decision-making. This occurs in the context of ongoing computations relating to spatial position, recall of long-term memory, attention, among many others. To establish how intermittently presented information is integrated with ongoing computations we recorded single units, simultaneously in hippocampus and PFC, in control rats and those with a brain malformation during performance of an SWM task. Neurons that encode intermittent task parameters are also well modulated in time and incorporated into a functional network across regions. Neurons from animals with cortical malformation are poorly modulated in time, less likely to encode task parameters, and less likely to be integrated into a functional network. Our results implicate a model in which ongoing oscillatory coordination among neurons in the hippocampal–PFC network describes a functional network that is poised to receive sensory inputs that are then integrated and multiplexed as working memory. The background temporal modulation is systematically altered in disease, but the relationship between these dynamics and behaviorally relevant firing is maintained, thereby providing potential targets for stimulation-based therapies.

Separable codes for read-out of mouse primary visual cortex across attentional states

Attentional modulation of neuronal activity in sensory cortex could alter perception by enhancing the local representation of attended stimuli or its behavioral read-out downstream. We tested these hypotheses using a task in which mice are cued on interleaved trials to attend visual or auditory targets. Neurons in primary visual cortex (V1) that encode task stimuli have larger visually-evoked responses when attention is directed toward vision. To determine whether the attention-dependent changes in V1 reflect changes in representation or read-out, we decoded task stimuli and choices from population activity. Surprisingly, both visual and auditory choices can be decoded from V1, but decoding takes advantage of unique activity patterns across modalities. Furthermore, decoding of choices, but not stimuli, is impaired when attention is directed toward the opposite modality. The specific effect on choice suggests behavioral improvements with attention are largely due to targeted read-out of the most informative V1 neurons.

Flexible and accurate decoding of neural populations through stochastic comodulation

Sensory-guided behavior requires reliable encoding of information (from stimuli to neural responses) and flexible decoding (from neural responses to behavior). In typical decision tasks, a small subset of cells within a large population encode task-relevant stimulus information and need to be identified by later processing stages for relevant information to be transmitted. A statistically optimal decoder (e.g., maximum likelihood) can utilize task-relevant cells for any given task configuration, but relies on complete knowledge of the relationship between the task and the stimulus-response and noise properties of the encoding population. The brain could learn an optimal decoder for a task through supervised learning (i.e., regression), but this typically requires many training trials, and thus lacks the flexibility of humans or animals, that can rapidly adjust to changes in task parameters or structure. Here, we propose a novel decoding solution based on functionally targeted stochastic modulation. Population recordings during different discrimination tasks have revealed that a substantial portion of trial-to-trial variability in cell responses can be explained by stochastic modulatory signals that are shared, and that seem to preferentially target task-informative neurons (Rabinowitz et al., 2015). The variability introduced by these modulators corrupts the encoded stimulus signal, but we propose that it also serves as a label for the informative neurons, allowing the decoder to solve the identification problem. We show in simulations of a modulated Poisson spiking model that a linear decoder with readout weights proportional to the estimated neuron-specific strength of modulation achieves performance close to an optimal decoder.

Layer 5 circuits in V1 differentially control visuomotor behavior

SummarySensory areas of the mammalian neocortex are thought to act as distribution hubs, transmitting information about the external environment to various cortical and subcortical structures in order to generate adaptive behavior. However, the exact role of cortical circuits in sensory perception remains unclear. Within primary visual cortex (V1), various populations of pyramidal neurons (PNs) send axonal projections to distinct targets, suggesting multiple cellular networks that may be independently engaged during the generation of behavior. Here, we investigated whether PN subpopulations differentially support sensory-guided performance by training mice in a visual detection task based on eyeblink conditioning. Applying 2-photon calcium imaging and optogenetic manipulation of anatomically-defined PNs in behaving animals, we show that layer 5 corticopontine neurons strongly encode task information and are selectively necessary for performance. These results contrast with recent observations of operant behavior showing a limited role for the neocortex in sensory detection. Instead, our findings support a model in which target-specific cortical subnetworks form the basis for adaptive behavior by directing relevant information to downstream brain areas. Overall, this work highlights the potential for neurons to form physically interspersed but functionally segregated networks capable of parallel, independent control of perception and behavior.

A flexible optimization-based method for synthesizing intent-expressive robot arm motion

We present an approach to synthesize robot arm trajectories that effectively communicate the robot’s intent to a human collaborator while achieving task goals. Our approach uses nonlinear constrained optimization to encode task requirements and desired motion properties. Our implementation allows for a wide range of constraints and objectives. We introduce a novel objective function to optimize robot arm motions for intent-expressiveness that works in a range of scenarios and robot arm types. Our formulation supports experimentation with different theories of how viewers interpret robot motion. Through a series of human-subject experiments on real and simulated robots, we demonstrate that our method leads to improved collaborative performance against other methods, including the current state of the art. These experiments also show how our perception heuristic can affect collaborative outcomes.

Nominally non-responsive frontal and sensory cortical cells encode task-relevant variables via ensemble consensus-building

Neurons recorded in behaving animals often do not discernibly respond to sensory input and are not overtly task-modulated. These nominally non-responsive neurons are difficult to interpret and are typically neglected from analysis, confounding attempts to connect neural activity to perception and behavior. Here we describe a trial-by-trial, spike-timing-based algorithm to reveal the hidden coding capacities of these neurons in auditory and frontal cortex of behaving rats. Responsive and nominally non-responsive cells contained significant information about sensory stimuli and behavioral decisions, and network modeling indicated that nominally non-responsive cells are important for task performance. Sensory input was more accurately represented in frontal cortex than auditory cortex, via ensembles of nominally non-responsive cells coordinating the behavioral meaning of spike timings on correct but not error trials. This unbiased approach allows the contribution of all recorded neurons - particularly those without obvious task-modulation - to be assessed for behavioral relevance on single trials.