A neural circuit model of flexible sensorimotor mapping: learning and forgetting on multiple timescales

Flexible gating between subspaces in a neural network model of internally guided task switching

Behavioral flexibility relies on the brain's ability to switch rapidly between multiple tasks, even when the task rule is not explicitly cued but must be inferred through trial and error. The underlying neural circuit mechanism remains poorly understood. We investigated recurrent neural networks (RNNs) trained to perform an analog of the classic Wisconsin Card Sorting Test. The networks consist of two modules responsible for rule representation and sensorimotor mapping, respectively, where each module is comprised of a circuit with excitatory neurons and three major types of inhibitory neurons. We found that rule representation by self-sustained persistent activity across trials, error monitoring and gated sensorimotor mapping emerged from training. Systematic dissection of trained RNNs revealed a detailed circuit mechanism that is consistent across networks trained with different hyperparameters. The networks' dynamical trajectories for different rules resided in separate subspaces of population activity; the subspaces collapsed and performance was reduced to chance level when dendrite-targeting somatostatin-expressing interneurons were silenced, illustrating how a phenomenological description of representational subspaces is explained by a specific circuit mechanism.

Fast adaptation to rule switching using neuronal surprise

In humans and animals, surprise is a physiological reaction to an unexpected event, but how surprise can be linked to plausible models of neuronal activity is an open problem. We propose a self-supervised spiking neural network model where a surprise signal is extracted from an increase in neural activity after an imbalance of excitation and inhibition. The surprise signal modulates synaptic plasticity via a three-factor learning rule which increases plasticity at moments of surprise. The surprise signal remains small when transitions between sensory events follow a previously learned rule but increases immediately after rule switching. In a spiking network with several modules, previously learned rules are protected against overwriting, as long as the number of modules is larger than the total number of rules-making a step towards solving the stability-plasticity dilemma in neuroscience. Our model relates the subjective notion of surprise to specific predictions on the circuit level.

Reconstructing computational system dynamics from neural data with recurrent neural networks

Computational models in neuroscience usually take the form of systems of differential equations. The behaviour of such systems is the subject of dynamical systems theory. Dynamical systems theory provides a powerful mathematical toolbox for analysing neurobiological processes and has been a mainstay of computational neuroscience for decades. Recently, recurrent neural networks (RNNs) have become a popular machine learning tool for studying the non-linear dynamics of neural and behavioural processes by emulating an underlying system of differential equations. RNNs have been routinely trained on similar behavioural tasks to those used for animal subjects to generate hypotheses about the underlying computational mechanisms. By contrast, RNNs can also be trained on the measured physiological and behavioural data, thereby directly inheriting their temporal and geometrical properties. In this way they become a formal surrogate for the experimentally probed system that can be further analysed, perturbed and simulated. This powerful approach is called dynamical system reconstruction. In this Perspective, we focus on recent trends in artificial intelligence and machine learning in this exciting and rapidly expanding field, which may be less well known in neuroscience. We discuss formal prerequisites, different model architectures and training approaches for RNN-based dynamical system reconstructions, ways to evaluate and validate model performance, how to interpret trained models in a neuroscience context, and current challenges.

Training biologically plausible recurrent neural networks on cognitive tasks with long-term dependencies

Training recurrent neural networks (RNNs) has become a go-to approach for generating and evaluating mechanistic neural hypotheses for cognition. The ease and efficiency of training RNNs with backpropagation through time and the availability of robustly supported deep learning libraries has made RNN modeling more approachable and accessible to neuroscience. Yet, a major technical hindrance remains. Cognitive processes such as working memory and decision making involve neural population dynamics over a long period of time within a behavioral trial and across trials. It is difficult to train RNNs to accomplish tasks where neural representations and dynamics have long temporal dependencies without gating mechanisms such as LSTMs or GRUs which currently lack experimental support and prohibit direct comparison between RNNs and biological neural circuits. We tackled this problem based on the idea of specialized skip-connections through time to support the emergence of task-relevant dynamics, and subsequently reinstitute biological plausibility by reverting to the original architecture. We show that this approach enables RNNs to successfully learn cognitive tasks that prove impractical if not impossible to learn using conventional methods. Over numerous tasks considered here, we achieve less training steps and shorter wall-clock times, particularly in tasks that require learning long-term dependencies via temporal integration over long timescales or maintaining a memory of past events in hidden-states. Our methods expand the range of experimental tasks that biologically plausible RNN models can learn, thereby supporting the development of theory for the emergent neural mechanisms of computations involving long-term dependencies.

Individual prefrontal neurons contribute to sensory-to-motor information transformation by rotating reference frames during spatial working memory performance

Performing working memory tasks correctly requires not only the temporary maintenance of information but also the visual-to-motor transformation of information. Although sustained delay-period activity is known to be a mechanism for temporarily maintaining information, the mechanism for information transformation is not well known. An analysis using a population of delay-period activities recorded from prefrontal neurons visualized a gradual change of maintained information from sensory to motor as the delay period progressed. However, the contributions of individual prefrontal neurons to this process are not known. In the present study, we used a version of the delayed-response task, in which monkeys needed to make a saccade 90o clockwise from a visual cue after a 3-s delay, and examined the temporal change in the preferred directions of delay-period activity during the delay period for individual neurons. One group of prefrontal neurons encoded the cue direction by a retinotopic reference frame and either maintained it throughout the delay period or rotated it 90o counterclockwise to adjust visual information to saccade information, whereas other groups of neurons encoded the cue direction by a saccade-based reference frame and rotated it 90o clockwise. The results indicate that visual-to-motor information transformation is achieved by manipulating the reference frame to adjust visual coordinates to motor coordinates.

Schema formation in a neural population subspace underlies learning-to-learn in flexible sensorimotor problem-solving

Learning-to-learn, a progressive speedup of learning while solving a series of similar problems, represents a core process of knowledge acquisition that draws attention in both neuroscience and artificial intelligence. To investigate its underlying brain mechanism, we trained a recurrent neural network model on arbitrary sensorimotor mappings known to depend on the prefrontal cortex. The network displayed an exponential time course of accelerated learning. The neural substrate of a schema emerges within a low-dimensional subspace of population activity; its reuse in new problems facilitates learning by limiting connection weight changes. Our work highlights the weight-driven modifications of the vector field, which determines the population trajectory of a recurrent network and behavior. Such plasticity is especially important for preserving and reusing the learned schema in spite of undesirable changes of the vector field due to the transition to learning a new problem; the accumulated changes across problems account for the learning-to-learn dynamics.

Neural learning rules for generating flexible predictions and computing the successor representation

The predictive nature of the hippocampus is thought to be useful for memory-guided cognitive behaviors. Inspired by the reinforcement learning literature, this notion has been formalized as a predictive map called the successor representation (SR). The SR captures a number of observations about hippocampal activity. However, the algorithm does not provide a neural mechanism for how such representations arise. Here, we show the dynamics of a recurrent neural network naturally calculate the SR when the synaptic weights match the transition probability matrix. Interestingly, the predictive horizon can be flexibly modulated simply by changing the network gain. We derive simple, biologically plausible learning rules to learn the SR in a recurrent network. We test our model with realistic inputs and match hippocampal data recorded during random foraging. Taken together, our results suggest that the SR is more accessible in neural circuits than previously thought and can support a broad range of cognitive functions.

A characterization of the neural representation of confidence during probabilistic learning

Learning in a stochastic and changing environment is a difficult task. Models of learning typically postulate that observations that deviate from the learned predictions are surprising and used to update those predictions. Bayesian accounts further posit the existence of a confidence-weighting mechanism: learning should be modulated by the confidence level that accompanies those predictions. However, the neural bases of this confidence are much less known than the ones of surprise. Here, we used a dynamic probability learning task and high-field MRI to identify putative cortical regions involved in the representation of confidence about predictions during human learning. We devised a stringent test based on the conjunction of four criteria. We localized several regions in parietal and frontal cortices whose activity is sensitive to the confidence of an ideal observer, specifically so with respect to potential confounds (surprise and predictability), and in a way that is invariant to which item is predicted. We also tested for functionality in two ways. First, we localized regions whose activity patterns at the subject level showed an effect of both confidence and surprise in qualitative agreement with the confidence-weighting principle. Second, we found neural representations of ideal confidence that also accounted for subjective confidence. Taken together, those results identify a set of cortical regions potentially implicated in the confidence-weighting of learning.

Recurrent networks endowed with structural priors explain suboptimal animal behavior

The strategies found by animals facing a new task are determined both by individual experience and by structural priors evolved to leverage the statistics of natural environments. Rats quickly learn to capitalize on the trial sequence correlations of two-alternative forced choice (2AFC) tasks after correct trials but consistently deviate from optimal behavior after error trials. To understand this outcome-dependent gating, we first show that recurrent neural networks (RNNs) trained in the same 2AFC task outperform rats as they can readily learn to use across-trial information both after correct and error trials. We hypothesize that, although RNNs can optimize their behavior in the 2AFC task without any a priori restrictions, rats' strategy is constrained by a structural prior adapted to a natural environment in which rewarded and non-rewarded actions provide largely asymmetric information. When pre-training RNNs in a more ecological task with more than two possible choices, networks develop a strategy by which they gate off the across-trial evidence after errors, mimicking rats' behavior. Population analyses show that the pre-trained networks form an accurate representation of the sequence statistics independently of the outcome in the previous trial. After error trials, gating is implemented by a change in the network dynamics that temporarily decouple the categorization of the stimulus from the across-trial accumulated evidence. Our results suggest that the rats' suboptimal behavior reflects the influence of a structural prior that reacts to errors by isolating the network decision dynamics from the context, ultimately constraining the performance in a 2AFC laboratory task.

Flexible sensory-motor mapping rules manifest in correlated variability of stimulus and action codes across the brain

Humans and non-human primates can flexibly switch between different arbitrary mappings from sensation to action to solve a cognitive task. It has remained unknown how the brain implements such flexible sensory-motor mapping rules. Here, we uncovered a dynamic reconfiguration of task-specific correlated variability between sensory and motor brain regions. Human participants switched between two rules for reporting visual orientation judgments during fMRI recordings. Rule switches were either signaled explicitly or inferred by the participants from ambiguous cues. We used behavioral modeling to reconstruct the time course of their belief about the active rule. In both contexts, the patterns of correlations between ongoing fluctuations in stimulus- and action-selective activity across visual- and action-related brain regions tracked participants' belief about the active rule. The rule-specific correlation patterns broke down around the time of behavioral errors. We conclude that internal beliefs about task state are instantiated in brain-wide, selective patterns of correlated variability.

Neural Mechanisms That Make Perceptual Decisions Flexible

Neural mechanisms of perceptual decision making have been extensively studied in experimental settings that mimic stable environments with repeating stimuli, fixed rules, and payoffs. In contrast, we live in an ever-changing environment and have varying goals and behavioral demands. To accommodate variability, our brain flexibly adjusts decision-making processes depending on context. Here, we review a growing body of research that explores the neural mechanisms underlying this flexibility. We highlight diverse forms of context dependency in decision making implemented through a variety of neural computations. Context-dependent neural activity is observed in a distributed network of brain structures, including posterior parietal, sensory, motor, and subcortical regions, as well as the prefrontal areas classically implicated in cognitive control. We propose that investigating the distributed network underlying flexible decisions is key to advancing our understanding and discuss a path forward for experimental and theoretical investigations.

Human inference reflects a normative balance of complexity and accuracy

We must often infer latent properties of the world from noisy and changing observations. Complex, probabilistic approaches to this challenge such as Bayesian inference are accurate but cognitively demanding, relying on extensive working memory and adaptive processing. Simple heuristics are easy to implement but may be less accurate. What is the appropriate balance between complexity and accuracy? Here we model a hierarchy of strategies of variable complexity and find a power law of diminishing returns: increasing complexity gives progressively smaller gains in accuracy. The rate of diminishing returns depends systematically on the statistical uncertainty in the world, such that complex strategies do not provide substantial benefits over simple ones when uncertainty is either too high or too low. In between, there is a complexity dividend. In two psychophysical experiments, we confirm specific model predictions about how working memory and adaptivity should be modulated by uncertainty.

Gated recurrence enables simple and accurate sequence prediction in stochastic, changing, and structured environments

From decision making to perception to language, predicting what is coming next is crucial. It is also challenging in stochastic, changing, and structured environments; yet the brain makes accurate predictions in many situations. What computational architecture could enable this feat? Bayesian inference makes optimal predictions but is prohibitively difficult to compute. Here, we show that a specific recurrent neural network architecture enables simple and accurate solutions in several environments. This architecture relies on three mechanisms: gating, lateral connections, and recurrent weight training. Like the optimal solution and the human brain, such networks develop internal representations of their changing environment (including estimates of the environment’s latent variables and the precision of these estimates), leverage multiple levels of latent structure, and adapt their effective learning rate to changes without changing their connection weights. Being ubiquitous in the brain, gated recurrence could therefore serve as a generic building block to predict in real-life environments.

Information capacity and robustness of encoding in the medial prefrontal cortex are modulated by the bioavailability of serotonin and the time elapsed from the cue during a reward-driven task

Serotonin (5-HT) is a key neuromodulator of medial prefrontal cortex (mPFC) functions. Pharmacological manipulation of systemic 5-HT bioavailability alters the electrical activity of mPFC neurons. However, 5-HT modulation at the population level is not well characterized. In the present study, we made single neuron extracellular recordings in the mPFC of rats performing an operant conditioning task, and analyzed the effect of systemic administration of fluoxetine (a selective serotonin reuptake inhibitor) on the information encoded in the firing activity of the neural population. Chronic (longer than 15 days), but not acute (less than 15 days), fluoxetine administration reduced the firing rate of mPFC neurons. Moreover, fluoxetine treatment enhanced pairwise entropy but diminished noise correlation and redundancy in the information encoded, thus showing how mPFC differentially encodes information as a function of 5-HT bioavailability. Information about the occurrence of the reward-predictive stimulus was maximized during reward consumption, around 3 to 4 s after the presentation of the cue, and it was higher under chronic fluoxetine treatment. However, the encoded information was less robust to noise corruption when compared to control conditions.

Flexible categorization in perceptual decision making

Perceptual decisions rely on accumulating sensory evidence. This computation has been studied using either drift diffusion models or neurobiological network models exhibiting winner-take-all attractor dynamics. Although both models can account for a large amount of data, it remains unclear whether their dynamics are qualitatively equivalent. Here we show that in the attractor model, but not in the drift diffusion model, an increase in the stimulus fluctuations or the stimulus duration promotes transitions between decision states. The increase in the number of transitions leads to a crossover between weighting mostly early evidence (primacy) to weighting late evidence (recency), a prediction we validate with psychophysical data. Between these two limiting cases, we found a novel flexible categorization regime, in which fluctuations can reverse initially-incorrect categorizations. This reversal asymmetry results in a non-monotonic psychometric curve, a distinctive feature of the attractor model. Our findings point to correcting decision reversals as an important feature of perceptual decision making.

Hippocampal Network Reorganization Underlies the Formation of a Temporal Association Memory

Episodic memory requires linking events in time, a function dependent on the hippocampus. In "trace" fear conditioning, animals learn to associate a neutral cue with an aversive stimulus despite their separation in time by a delay period on the order of tens of seconds. But how this temporal association forms remains unclear. Here we use two-photon calcium imaging of neural population dynamics throughout the course of learning and show that, in contrast to previous theories, hippocampal CA1 does not generate persistent activity to bridge the delay. Instead, learning is concomitant with broad changes in the active neural population. Although neural responses were stochastic in time, cue identity could be read out from population activity over longer timescales after learning. These results question the ubiquity of seconds-long neural sequences during temporal association learning and suggest that trace fear conditioning relies on mechanisms that differ from persistent activity accounts of working memory.

Neural Correlates of Reinforcement Learning in Mid-lateral Cerebellum

The role of the cerebellum in non-motor learning is poorly understood. Here, we investigated the activity of Purkinje cells (P-cells) in the mid-lateral cerebellum as the monkey learned to associate one arbitrary symbol with the movement of the left hand and another with the movement of the right hand. During learning, but not when the monkey had learned the association, the simple spike responses of P-cells reported the outcome of the animal's most recent decision without concomitant changes in other sensorimotor parameters such as hand movement, licking, or eye movement. At the population level, P-cells collectively maintained a memory of the most recent decision throughout the entire trial. As the monkeys learned the association, the magnitude of this reward-related error signal approached zero. Our results provide a major departure from the current understanding of cerebellar processing and have critical implications for cerebellum's role in cognitive control.

Dissociation of nicotinic α7 and α4/β2 sub-receptor agonists for enhancing learning and attentional filtering in nonhuman primates

RATIONALE

Nicotinic acetylcholine receptors (nAChRs) modulate attention, memory, and higher executive functioning, but it is unclear how nACh sub-receptors mediate different mechanisms supporting these functions.

OBJECTIVES

We investigated whether selective agonists for the alpha-7 nAChR versus the alpha-4/beta-2 nAChR have unique functional contributions for value learning and attentional filtering of distractors in the nonhuman primate.

METHODS

Two adult rhesus macaque monkeys performed reversal learning following systemic administration of either the alpha-7 nAChR agonist PHA-543613 or the alpha-4/beta-2 nAChR agonist ABT-089 or a vehicle control. Behavioral analysis quantified performance accuracy, speed of processing, reversal learning speed, the control of distractor interference, perseveration tendencies, and motivation.

RESULTS

We found that the alpha-7 nAChR agonist PHA-543613 enhanced the learning speed of feature values but did not modulate how salient distracting information was filtered from ongoing choice processes. In contrast, the selective alpha-4/beta-2 nAChR agonist ABT-089 did not affect learning speed but reduced distractibility. This dissociation was dose-dependent and evident in the absence of systematic changes in overall performance, reward intake, motivation to perform the task, perseveration tendencies, or reaction times.

CONCLUSIONS

These results suggest nicotinic sub-receptor specific mechanisms consistent with (1) alpha-4/beta-2 nAChR specific amplification of cholinergic transients in prefrontal cortex linked to enhanced cue detection in light of interferences, and (2) alpha-7 nAChR specific activation prolonging cholinergic transients, which could facilitate subjects to follow-through with newly established attentional strategies when outcome contingencies change. These insights will be critical for developing function-specific drugs alleviating attention and learning deficits in neuro-psychiatric diseases.

Temporal chunking as a mechanism for unsupervised learning of task-sets

Depending on environmental demands, humans can learn and exploit multiple concurrent sets of stimulus-response associations. Mechanisms underlying the learning of such task-sets remain unknown. Here we investigate the hypothesis that task-set learning relies on unsupervised chunking of stimulus-response associations that occur in temporal proximity. We examine behavioral and neural data from a task-set learning experiment using a network model. We first show that task-set learning can be achieved provided the timescale of chunking is slower than the timescale of stimulus-response learning. Fitting the model to behavioral data on a subject-by-subject basis confirmed this expectation and led to specific predictions linking chunking and task-set retrieval that were borne out by behavioral performance and reaction times. Comparing the model activity with BOLD signal allowed us to identify neural correlates of task-set retrieval in a functional network involving ventral and dorsal prefrontal cortex, with the dorsal system preferentially engaged when retrievals are used to improve performance.

A Minimum Free Energy Model of Motor Learning

Even highly trained behaviors demonstrate variability, which is correlated with performance on current and future tasks. An objective of motor learning that is general enough to explain these phenomena has not been precisely formulated. In this six-week longitudinal learning study, participants practiced a set of motor sequences each day, and neuroimaging data were collected on days 1, 14, 28, and 42 to capture the neural correlates of the learning process. In our analysis, we first modeled the underlying neural and behavioral dynamics during learning. Our results demonstrate that the densities of whole-brain response, task-active regional response, and behavioral performance evolve according to a Fokker-Planck equation during the acquisition of a motor skill. We show that this implies that the brain concurrently optimizes the entropy of a joint density over neural response and behavior (as measured by sampling over multiple trials and subjects) and the expected performance under this density; we call this formulation of learning minimum free energy learning (MFEL). This model provides an explanation as to how behavioral variability can be tuned while simultaneously improving performance during learning. We then develop a novel variant of inverse reinforcement learning to retrieve the cost function optimized by the brain during the learning process, as well as the parameter used to tune variability. We show that this population-level analysis can be used to derive a learning objective that each subject optimizes during his or her study. In this way, MFEL effectively acts as a unifying principle, allowing users to precisely formulate learning objectives and infer their structure.

Replays of spatial memories suppress topological fluctuations in cognitive map

The spiking activity of the hippocampal place cells plays a key role in producing and sustaining an internalized representation of the ambient space-a cognitive map. These cells do not only exhibit location-specific spiking during navigation, but also may rapidly replay the navigated routs through endogenous dynamics of the hippocampal network. Physiologically, such reactivations are viewed as manifestations of "memory replays" that help to learn new information and to consolidate previously acquired memories by reinforcing synapses in the parahippocampal networks. Below we propose a computational model of these processes that allows assessing the effect of replays on acquiring a robust topological map of the environment and demonstrate that replays may play a key role in stabilizing the hippocampal representation of space.

Dorsolateral prefrontal neurons mediate subjective decisions and their variation in humans

Subjective decisions play a vital role in human behavior because, while often grounded in fact, they are inherently based on personal beliefs that can vary broadly within and between individuals. While these properties set subjective decisions apart from many other sensorimotor processes and are of wide sociological impact, their single-neuronal basis in humans is unknown. Here we find cells in the dorsolateral prefrontal cortex (dlPFC) that reflect variations in the subjective decisions of humans when performing opinion-based tasks. These neurons changed their activities gradually as the participants transitioned between choice options but also reflected their unique point of conversion at equipoise. Focal disruption of the dlPFC, by contrast, diminished gradation between opposing decisions but had little effect on sensory perceptual choices or their motor report. These findings suggest that the human dlPFC plays an important role in subjective decisions and propose a mechanism for mediating their variation during opinion formation.

Dissociating task acquisition from expression during learning reveals latent knowledge

Performance on cognitive tasks during learning is used to measure knowledge, yet it remains controversial since such testing is susceptible to contextual factors. To what extent does performance during learning depend on the testing context, rather than underlying knowledge? We trained mice, rats and ferrets on a range of tasks to examine how testing context impacts the acquisition of knowledge versus its expression. We interleaved reinforced trials with probe trials in which we omitted reinforcement. Across tasks, each animal species performed remarkably better in probe trials during learning and inter-animal variability was strikingly reduced. Reinforcement feedback is thus critical for learning-related behavioral improvements but, paradoxically masks the expression of underlying knowledge. We capture these results with a network model in which learning occurs during reinforced trials while context modulates only the read-out parameters. Probing learning by omitting reinforcement thus uncovers latent knowledge and identifies context- not “smartness”- as the major source of individual variability.

Performance is generally used as a metric to assay whether an animal has learnt a particular perceptual task. Here the authors demonstrate that in the context of probe trials without the possibility of reward, animals perform the correct instrumental response suggesting a latent knowledge of the task much before it is manifest in their performance.

Deviation from the matching law reflects an optimal strategy involving learning over multiple timescales

Behavior deviating from our normative expectations often appears irrational. For example, even though behavior following the so-called matching law can maximize reward in a stationary foraging task, actual behavior commonly deviates from matching. Such behavioral deviations are interpreted as a failure of the subject; however, here we instead suggest that they reflect an adaptive strategy, suitable for uncertain, non-stationary environments. To prove it, we analyzed the behavior of primates that perform a dynamic foraging task. In such nonstationary environment, learning on both fast and slow timescales is beneficial: fast learning allows the animal to react to sudden changes, at the price of large fluctuations (variance) in the estimates of task relevant variables. Slow learning reduces the fluctuations but costs a bias that causes systematic behavioral deviations. Our behavioral analysis shows that the animals solved this bias-variance tradeoff by combining learning on both fast and slow timescales, suggesting that learning on multiple timescales can be a biologically plausible mechanism for optimizing decisions under uncertainty.

Recent experience can only provide limited information to guide decisions in a volatile environment. Here, the authors report that the choices made by a monkey in a dynamic foraging task can be better explained by a model that combines learning on both fast and slow timescales.

Feature-specific prediction errors and surprise across macaque fronto-striatal circuits

To adjust expectations efficiently, prediction errors need to be associated with the precise features that gave rise to the unexpected outcome, but this credit assignment may be problematic if stimuli differ on multiple dimensions and it is ambiguous which feature dimension caused the outcome. Here, we report a potential solution: neurons in four recorded areas of the anterior fronto-striatal networks encode prediction errors that are specific to feature values of different dimensions of attended multidimensional stimuli. The most ubiquitous prediction error occurred for the reward-relevant dimension. Feature-specific prediction error signals a) emerge on average shortly after non-specific prediction error signals, b) arise earliest in the anterior cingulate cortex and later in dorsolateral prefrontal cortex, caudate and ventral striatum, and c) contribute to feature-based stimulus selection after learning. Thus, a widely-distributed feature-specific eligibility trace may be used to update synaptic weights for improved feature-based attention.

In order to adjust expectations efficiently, prediction errors need to be associated with the features that gave rise to the unexpected outcome. Here, the authors show that neurons in anterior fronto-striatal networks encode prediction errors that are specific to feature values of different stimulus dimensions.

Forget-me-some: General versus special purpose models in a hierarchical probabilistic task

Humans build models of their environments and act according to what they have learnt. In simple experimental environments, such model-based behaviour is often well accounted for as if subjects are ideal Bayesian observers. However, more complex probabilistic tasks require more sophisticated forms of inference that are sufficiently computationally and statistically taxing as to demand approximation. Here, we study properties of two approximation schemes in the context of a serial reaction time task in which stimuli were generated from a hierarchical Markov chain. One, pre-existing, scheme was a generically powerful variational method for hierarchical inference which has recently become popular as an account of psychological and neural data across a wide swathe of probabilistic tasks. A second, novel, scheme was more specifically tailored to the task at hand. We show that the latter model fit significantly better than the former. This suggests that our subjects were sensitive to many of the particular constraints of a complex behavioural task. Further, the tailored model provided a different perspective on the effects of cholinergic manipulations in the task. Neither model fit the behaviour on more complex contingencies that competently. These results illustrate the benefits and challenges that come with the general and special purpose modelling approaches and raise important questions of how they can advance our current understanding of learning mechanisms in the brain.

Corticospinal correlates of fast and slow adaptive processes in motor learning

Recent computational theories and behavioral observations suggest that motor learning is supported by multiple adaptation processes, operating on different timescales, but direct neural evidence is lacking. We tested this hypothesis by applying transcranial magnetic stimulation over motor cortex in 16 human subjects during a validated reach adaptation task. Motor-evoked potentials (MEPs) and cortical silent periods (CSPs) were recorded from the biceps brachii to assess modulations of corticospinal excitability as indices for corticospinal plasticity. Guided by a two-state adaptation model, we show that the MEP reflects an adaptive process that learns quickly but has poor retention, while the CSP correlates with a process that responds more slowly but retains information well. These results provide a physiological link between models of motor learning and distinct changes in corticospinal excitability. Our findings support the relationship between corticospinal gain modulations and the adaptive processes in motor learning. NEW & NOTEWORTHY Computational theories and behavioral observations suggest that motor learning is supported by multiple adaptation processes, but direct neural evidence is lacking. We tested this hypothesis by applying transcranial magnetic stimulation over human motor cortex during a reach adaptation task. Guided by a two-state adaptation model, we show that the motor-evoked potential reflects a process that adapts and decays quickly, whereas the cortical silent period reflects slow adaptation and decay.

An effect of serotonergic stimulation on learning rates for rewards apparent after long intertrial intervals

Serotonin has widespread, but computationally obscure, modulatory effects on learning and cognition. Here, we studied the impact of optogenetic stimulation of dorsal raphe serotonin neurons in mice performing a non-stationary, reward-driven decision-making task. Animals showed two distinct choice strategies. Choices after short inter-trial-intervals (ITIs) depended only on the last trial outcome and followed a win-stay-lose-switch pattern. In contrast, choices after long ITIs reflected outcome history over multiple trials, as described by reinforcement learning models. We found that optogenetic stimulation during a trial significantly boosted the rate of learning that occurred due to the outcome of that trial, but these effects were only exhibited on choices after long ITIs. This suggests that serotonin neurons modulate reinforcement learning rates, and that this influence is masked by alternate, unaffected, decision mechanisms. These results provide insight into the role of serotonin in treating psychiatric disorders, particularly its modulation of neural plasticity and learning.

Serotonin (5-HT) plays many important roles in reward, punishment, patience and beyond, and optogenetic stimulation of 5-HT neurons has not crisply parsed them. The authors report a novel analysis of a reward-based decision-making experiment, and show that 5-HT stimulation increases the learning rate, but only on a select subset of choices.

Medial Frontal Circuit Dynamics Represents Probabilistic Choices for Unfamiliar Sensory Experience

Neurons in medial frontal cortex (MFC) receive sensory signals that are crucial for decision-making behavior. While decision-making is easy for familiar sensory signals, it becomes more elaborative when sensory signals are less familiar to animals. It remains unclear how the population of neurons enables the coordinate transformation of such a sensory input into ambiguous choice responses. Furthermore, whether and how cortical oscillations temporally coordinate neuronal firing during this transformation has not been extensively studied. Here, we recorded neuronal population responses to familiar or unfamiliar auditory cues in rat MFC and computed their probabilistic evolution. Population responses to familiar sounds organize into neuronal trajectories containing multiplexed sensory, motor, and choice information. Unfamiliar sounds, in contrast, evoke trajectories that travel under the guidance of familiar paths and eventually diverge to unique decision states. Local field potentials exhibited beta- (15-20 Hz) and gamma-band (50-60 Hz) oscillations to which neuronal firing showed modest phase locking. Interestingly, gamma oscillation, but not beta oscillation, increased its power abruptly at some timepoint by which neural trajectories for different choices were near maximally separated. Our results emphasize the importance of the evolution of neural trajectories in rapid probabilistic decisions that utilize unfamiliar sensory information.

Normal CA1 Place Fields but Discoordinated Network Discharge in a Fmr1-Null Mouse Model of Fragile X Syndrome

Silence of FMR1 causes loss of fragile X mental retardation protein (FMRP) and dysregulated translation at synapses, resulting in the intellectual disability and autistic symptoms of fragile X syndrome (FXS). Synaptic dysfunction hypotheses for how intellectual disabilities like cognitive inflexibility arise in FXS predict impaired neural coding in the absence of FMRP. We tested the prediction by comparing hippocampus place cells in wild-type and FXS-model mice. Experience-driven CA1 synaptic function and synaptic plasticity changes are excessive in Fmr1-null mice, but CA1 place fields are normal. However, Fmr1-null discharge relationships to local field potential oscillations are abnormally weak, stereotyped, and homogeneous; also, discharge coordination within Fmr1-null place cell networks is weaker and less reliable than wild-type. Rather than disruption of single-cell neural codes, these findings point to invariant tuning of single-cell responses and inadequate discharge coordination within neural ensembles as a pathophysiological basis of cognitive inflexibility in FXS. VIDEO ABSTRACT.

Sure enough: efficient Bayesian learning and choice

Probabilistic decision-making is a general phenomenon in animal behavior, and has often been interpreted to reflect the relative certainty of animals' beliefs. Extensive neurological and behavioral results increasingly suggest that animal beliefs may be represented as probability distributions, with explicit accounting of uncertainty. Accordingly, we develop a model that describes decision-making in a manner consistent with this understanding of neuronal function in learning and conditioning. This first-order Markov, recursive Bayesian algorithm is as parsimonious as its minimalist point-estimate, Rescorla-Wagner analogue. We show that the Bayesian algorithm can reproduce naturalistic patterns of probabilistic foraging, in simulations of an experiment in bumblebees. We go on to show that the Bayesian algorithm can efficiently describe the behavior of several heuristic models of decision-making, and is consistent with the ubiquitous variation in choice that we observe within and between individuals in implementing heuristic decision-making. By describing learning and decision-making in a single Bayesian framework, we believe we can realistically unify descriptions of behavior across contexts and organisms. A unified cognitive model of this kind may facilitate descriptions of behavioral evolution.

Semantic integration by pattern priming: experiment and cortical network model

Neural network models describe semantic priming effects by way of mechanisms of activation of neurons coding for words that rely strongly on synaptic efficacies between pairs of neurons. Biologically inspired Hebbian learning defines efficacy values as a function of the activity of pre- and post-synaptic neurons only. It generates only pair associations between words in the semantic network. However, the statistical analysis of large text databases points to the frequent occurrence not only of pairs of words (e.g., "the way") but also of patterns of more than two words (e.g., "by the way"). The learning of these frequent patterns of words is not reducible to associations between pairs of words but must take into account the higher level of coding of three-word patterns. The processing and learning of pattern of words challenges classical Hebbian learning algorithms used in biologically inspired models of priming. The aim of the present study was to test the effects of patterns on the semantic processing of words and to investigate how an inter-synaptic learning algorithm succeeds at reproducing the experimental data. The experiment manipulates the frequency of occurrence of patterns of three words in a multiple-paradigm protocol. Results show for the first time that target words benefit more priming when embedded in a pattern with the two primes than when only associated with each prime in pairs. A biologically inspired inter-synaptic learning algorithm is tested that potentiates synapses as a function of the activation of more than two pre- and post-synaptic neurons. Simulations show that the network can learn patterns of three words to reproduce the experimental results.

Fast and slow transitions in frontal ensemble activity during flexible sensorimotor behavior

The ability to shift between repetitive and goal-directed actions is a hallmark of cognitive control. Previous studies have reported that adaptive shifts in behavior are accompanied by changes of neural activity in frontal cortex. However, neural and behavioral adaptations can occur at multiple time scales, and their relationship remains poorly defined. Here we developed an adaptive sensorimotor decision-making task for head-fixed mice, requiring them to shift flexibly between multiple auditory-motor mappings. Two-photon calcium imaging of secondary motor cortex (M2) revealed different ensemble activity states for each mapping. When adapting to a conditional mapping, transitions in ensemble activity were abrupt and occurred before the recovery of behavioral performance. By contrast, gradual and delayed transitions accompanied shifts toward repetitive responding. These results demonstrate distinct ensemble signatures associated with the start versus end of sensory-guided behavior and suggest that M2 leads in engaging goal-directed response strategies that require sensorimotor associations.

Conflict Resolution as Near-Threshold Decision-Making: A Spiking Neural Circuit Model with Two-Stage Competition for Antisaccadic Task

Automatic responses enable us to react quickly and effortlessly, but they often need to be inhibited so that an alternative, voluntary action can take place. To investigate the brain mechanism of controlled behavior, we investigated a biologically-based network model of spiking neurons for inhibitory control. In contrast to a simple race between pro- versus anti-response, our model incorporates a sensorimotor remapping module, and an action-selection module endowed with a “Stop” process through tonic inhibition. Both are under the modulation of rule-dependent control. We tested the model by applying it to the well known antisaccade task in which one must suppress the urge to look toward a visual target that suddenly appears, and shift the gaze diametrically away from the target instead. We found that the two-stage competition is crucial for reproducing the complex behavior and neuronal activity observed in the antisaccade task across multiple brain regions. Notably, our model demonstrates two types of errors: fast and slow. Fast errors result from failing to inhibit the quick automatic responses and therefore exhibit very short response times. Slow errors, in contrast, are due to incorrect decisions in the remapping process and exhibit long response times comparable to those of correct antisaccade responses. The model thus reveals a circuit mechanism for the empirically observed slow errors and broad distributions of erroneous response times in antisaccade. Our work suggests that selecting between competing automatic and voluntary actions in behavioral control can be understood in terms of near-threshold decision-making, sharing a common recurrent (attractor) neural circuit mechanism with discrimination in perception.

We propose a novel neural circuit mechanism and construct a spiking neural network model for resolving conflict between an automatic response and a volitional one. In this mechanism the two types of responses compete against each other under the modulation of top-down control via multiple neural pathways. The model is able to reproduce a wide range of neuronal and behavioral features observed in various studies and provides insights into not just how subjects make correct responses and fast errors, but also why they make slow errors, a type of error often overlooked by previous modeling studies. The model suggests critical roles of tonic (non-racing) top-down inhibition and near-threshold decision-making in neural competition.

A neural model of the frontal eye fields with reward-based learning

Decision-making is a flexible process dependent on the accumulation of various kinds of information; however, the corresponding neural mechanisms are far from clear. We extended a layered model of the frontal eye field to a learning-based model, using computational simulations to explain the cognitive process of choice tasks. The core of this extended model has three aspects: direction-preferred populations that cluster together the neurons with the same orientation preference, rule modules that control different rule-dependent activities, and reward-based synaptic plasticity that modulates connections to flexibly change the decision according to task demands. After repeated attempts in a number of trials, the network successfully simulated three decision choice tasks: an anti-saccade task, a no-go task, and an associative task. We found that synaptic plasticity could modulate the competition of choices by suppressing erroneous choices while enhancing the correct (rewarding) choice. In addition, the trained model captured some properties exhibited in animal and human experiments, such as the latency of the reaction time distribution of anti-saccades, the stop signal mechanism for canceling a reflexive saccade, and the variation of latency to half-max selectivity. Furthermore, the trained model was capable of reproducing the re-learning procedures when switching tasks and reversing the cue-saccade association.

Hebbian Wiring Plasticity Generates Efficient Network Structures for Robust Inference with Synaptic Weight Plasticity

In the adult mammalian cortex, a small fraction of spines are created and eliminated every day, and the resultant synaptic connection structure is highly nonrandom, even in local circuits. However, it remains unknown whether a particular synaptic connection structure is functionally advantageous in local circuits, and why creation and elimination of synaptic connections is necessary in addition to rich synaptic weight plasticity. To answer these questions, we studied an inference task model through theoretical and numerical analyses. We demonstrate that a robustly beneficial network structure naturally emerges by combining Hebbian-type synaptic weight plasticity and wiring plasticity. Especially in a sparsely connected network, wiring plasticity achieves reliable computation by enabling efficient information transmission. Furthermore, the proposed rule reproduces experimental observed correlation between spine dynamics and task performance.

Neural and neurochemical basis of reinforcement-guided decision making

Decision making is an adaptive behavior that takes into account several internal and external input variables and leads to the choice of a course of action over other available and often competing alternatives. While it has been studied in diverse fields ranging from mathematics, economics, ecology, and ethology to psychology and neuroscience, recent cross talk among perspectives from different fields has yielded novel descriptions of decision processes. Reinforcement-guided decision making models are based on economic and reinforcement learning theories, and their focus is on the maximization of acquired benefit over a defined period of time. Studies based on reinforcement-guided decision making have implicated a large network of neural circuits across the brain. This network includes a wide range of cortical (e.g., orbitofrontal cortex and anterior cingulate cortex) and subcortical (e.g., nucleus accumbens and subthalamic nucleus) brain areas and uses several neurotransmitter systems (e.g., dopaminergic and serotonergic systems) to communicate and process decision-related information. This review discusses distinct as well as overlapping contributions of these networks and neurotransmitter systems to the processing of decision making. We end the review by touching on neural circuitry and neuromodulatory regulation of exploratory decision making.

Neural substrates of cognitive biases during probabilistic inference

Decision making often requires simultaneously learning about and combining evidence from various sources of information. However, when making inferences from these sources, humans show systematic biases that are often attributed to heuristics or limitations in cognitive processes. Here we use a combination of experimental and modelling approaches to reveal neural substrates of probabilistic inference and corresponding biases. We find systematic deviations from normative accounts of inference when alternative options are not equally rewarding; subjects' choice behaviour is biased towards the more rewarding option, whereas their inferences about individual cues show the opposite bias. Moreover, inference bias about combinations of cues depends on the number of cues. Using a biophysically plausible model, we link these biases to synaptic plasticity mechanisms modulated by reward expectation and attention. We demonstrate that inference relies on direct estimation of posteriors, not on combination of likelihoods and prior. Our work reveals novel mechanisms underlying cognitive biases and contributions of interactions between reward-dependent learning, decision making and attention to high-level reasoning.

Humans are often biased in estimating the precise influence of probabilistic events on their decisions. Here, Khorsand and colleagues report a behavioural task that produces these biases in inference and describe a biophysically-plausible model that captures these behavioural deviations from optimal decision making.

Multiplicative and Additive Modulation of Neuronal Tuning with Population Activity Affects Encoded Information

Numerous studies have shown that neuronal responses are modulated by stimulus properties and also by the state of the local network. However, little is known about how activity fluctuations of neuronal populations modulate the sensory tuning of cells and affect their encoded information. We found that fluctuations in ongoing and stimulus-evoked population activity in primate visual cortex modulate the tuning of neurons in a multiplicative and additive manner. While distributed on a continuum, neurons with stronger multiplicative effects tended to have less additive modulation and vice versa. The information encoded by multiplicatively modulated neurons increased with greater population activity, while that of additively modulated neurons decreased. These effects offset each other so that population activity had little effect on total information. Our results thus suggest that intrinsic activity fluctuations may act as a "traffic light" that determines which subset of neurons is most informative.

Translational Research on Habit and Alcohol

Habitual actions enable efficient daily living, but they can also contribute to pathological behaviors that resistant change, such as alcoholism. Habitual behaviors are learned actions that appear goal-directed but are in fact no longer under the control of the action's outcome. Instead, these actions are triggered by stimuli, which may be exogenous or interoceptive, discrete or contextual. A major hallmark characteristic of alcoholism is continued alcohol use despite serious negative consequences. In essence, although the outcome of alcohol seeking and drinking is dramatically devalued, these actions persist, often triggered by environmental cues associated with alcohol use. Thus, alcoholism meets the definition of an initially goal-directed behavior that converts to a habit-based process. Habit and alcohol have been well investigated in rodent models, with comparatively less research in non-human primates and people. This review focuses on translational research on habit and alcohol with an emphasis on cross-species methodology and neural circuitry.

Adaptive learning and decision-making under uncertainty by metaplastic synapses guided by a surprise detection system

Recent experiments have shown that animals and humans have a remarkable ability to adapt their learning rate according to the volatility of the environment. Yet the neural mechanism responsible for such adaptive learning has remained unclear. To fill this gap, we investigated a biophysically inspired, metaplastic synaptic model within the context of a well-studied decision-making network, in which synapses can change their rate of plasticity in addition to their efficacy according to a reward-based learning rule. We found that our model, which assumes that synaptic plasticity is guided by a novel surprise detection system, captures a wide range of key experimental findings and performs as well as a Bayes optimal model, with remarkably little parameter tuning. Our results further demonstrate the computational power of synaptic plasticity, and provide insights into the circuit-level computation which underlies adaptive decision-making.

Abstract Context Representations in Primate Amygdala and Prefrontal Cortex

Neurons in prefrontal cortex (PFC) encode rules, goals, and other abstract information thought to underlie cognitive, emotional, and behavioral flexibility. Here we show that the amygdala, a brain area traditionally thought to mediate emotions, also encodes abstract information that could underlie this flexibility. Monkeys performed a task in which stimulus-reinforcement contingencies varied between two sets of associations, each defining a context. Reinforcement prediction required identifying a stimulus and knowing the current context. Behavioral evidence indicated that monkeys utilized this information to perform inference and adjust their behavior. Neural representations in both amygdala and PFC reflected the linked sets of associations implicitly defining each context, a process requiring a level of abstraction characteristic of cognitive operations. Surprisingly, when errors were made, the context signal weakened substantially in the amygdala. These data emphasize the importance of maintaining abstract cognitive information in the amygdala to support flexible behavior.

A neuro-computational model of economic decisions

Neuronal recordings and lesion studies indicate that key aspects of economic decisions take place in the orbitofrontal cortex (OFC). Previous work identified in this area three groups of neurons encoding the offer value, the chosen value, and the identity of the chosen good. An important and open question is whether and how decisions could emerge from a neural circuit formed by these three populations. Here we adapted a biophysically realistic neural network previously proposed for perceptual decisions (Wang XJ. Neuron 36: 955-968, 2002; Wong KF, Wang XJ. J Neurosci 26: 1314-1328, 2006). The domain of economic decisions is significantly broader than that for which the model was originally designed, yet the model performed remarkably well. The input and output nodes of the network were naturally mapped onto two groups of cells in OFC. Surprisingly, the activity of interneurons in the network closely resembled that of the third group of cells, namely, chosen value cells. The model reproduced several phenomena related to the neuronal origins of choice variability. It also generated testable predictions on the excitatory/inhibitory nature of different neuronal populations and on their connectivity. Some aspects of the empirical data were not reproduced, but simple extensions of the model could overcome these limitations. These results render a biologically credible model for the neuronal mechanisms of economic decisions. They demonstrate that choices could emerge from the activity of cells in the OFC, suggesting that chosen value cells directly participate in the decision process. Importantly, Wang's model provides a platform to investigate the implications of neuroscience results for economic theory.

Short-term plasticity as cause-effect hypothesis testing in distal reward learning

Asynchrony, overlaps, and delays in sensory-motor signals introduce ambiguity as to which stimuli, actions, and rewards are causally related. Only the repetition of reward episodes helps distinguish true cause-effect relationships from coincidental occurrences. In the model proposed here, a novel plasticity rule employs short- and long-term changes to evaluate hypotheses on cause-effect relationships. Transient weights represent hypotheses that are consolidated in long-term memory only when they consistently predict or cause future rewards. The main objective of the model is to preserve existing network topologies when learning with ambiguous information flows. Learning is also improved by biasing the exploration of the stimulus-response space toward actions that in the past occurred before rewards. The model indicates under which conditions beliefs can be consolidated in long-term memory, it suggests a solution to the plasticity-stability dilemma, and proposes an interpretation of the role of short-term plasticity.

Adaptation of short-term plasticity parameters via error-driven learning may explain the correlation between activity-dependent synaptic properties, connectivity motifs and target specificity

The anatomical connectivity among neurons has been experimentally found to be largely non-random across brain areas. This means that certain connectivity motifs occur at a higher frequency than would be expected by chance. Of particular interest, short-term synaptic plasticity properties were found to colocalize with specific motifs: an over-expression of bidirectional motifs has been found in neuronal pairs where short-term facilitation dominates synaptic transmission among the neurons, whereas an over-expression of unidirectional motifs has been observed in neuronal pairs where short-term depression dominates. In previous work we found that, given a network with fixed short-term properties, the interaction between short- and long-term plasticity of synaptic transmission is sufficient for the emergence of specific motifs. Here, we introduce an error-driven learning mechanism for short-term plasticity that may explain how such observed correspondences develop from randomly initialized dynamic synapses. By allowing synapses to change their properties, neurons are able to adapt their own activity depending on an error signal. This results in more rich dynamics and also, provided that the learning mechanism is target-specific, leads to specialized groups of synapses projecting onto functionally different targets, qualitatively replicating the experimental results of Wang and collaborators.

On self-feedback connectivity in neural mass models applied to event-related potentials

Neural mass models (NMMs) applied to neuroimaging data often do not emphasise intrinsic self-feedback within a neural population. However, based on mean-field theory, any population of coupled neurons is intrinsically endowed with effective self-coupling. In this work, we examine the effectiveness of three cortical NMMs with different self-feedbacks using a dynamic causal modelling approach. Specifically, we compare the classic Jansen and Rit (1995) model (no self-feedback), a modified model by Moran et al. (2007) (only inhibitory self-feedback), and our proposed model with inhibitory and excitatory self-feedbacks. Using bifurcation analysis, we show that single-unit Jansen-Rit model is less robust in generating oscillatory behaviour than the other two models. Next, under Bayesian inversion, we simulate single-channel event-related potentials (ERPs) within a mismatch negativity auditory oddball paradigm. We found fully self-feedback model (FSM) to provide the best fit to single-channel data. By analysing the posterior covariances of model parameters, we show that self-feedback connections are less sensitive to the generated evoked responses than the other model parameters, and hence can be treated analogously to "higher-order" parameter corrections of the original Jansen-Rit model. This is further supported in the more realistic multi-area case where FSM can replicate data better than JRM and MoM in the majority of subjects by capturing the finer features of the ERP data more accurately. Our work informs how NMMs with full self-feedback connectivity are not only more consistent with the underlying neurophysiology, but can also account for more complex features in ERP data.