Wave-like dopamine dynamics as a mechanism for spatiotemporal credit assignment.

Significant evidence supports the view that dopamine shapes learning by encoding reward prediction errors. However, it is unknown whether striatal targets receive tailored dopamine dynamics based on regional functional specialization. Here, we report wave-like spatiotemporal activity patterns in dopamine axons and release across the dorsal striatum. These waves switch between activational motifs and organize dopamine transients into localized clusters within functionally related striatal subregions. Notably, wave trajectories were tailored to task demands, propagating from dorsomedial to dorsolateral striatum when rewards are contingent on animal behavior and in the opponent direction when rewards are independent of behavioral responses. We propose a computational architecture in which striatal dopamine waves are sculpted by inference about agency and provide a mechanism to direct credit assignment to specialized striatal subregions. Supporting model predictions, dorsomedial dopamine activity during reward-pursuit signaled the extent of instrumental control and interacted with reward waves to predict future behavioral adjustments.

Subthalamic control of impulsive actions: insights from deep brain stimulation in Parkinson's disease.

Control of actions allows adaptive, goal-directed behaviour. The basal ganglia, including the subthalamic nucleus, are thought to play a central role in dynamically controlling actions through recurrent negative feedback loops with the cerebral cortex. Here, we summarize recent translational studies that used deep brain stimulation to record neural activity from and apply electrical stimulation to the subthalamic nucleus in people with Parkinson's disease. These studies have elucidated spatial, spectral and temporal features of the neural mechanisms underlying the controlled delay of actions in cortico-subthalamic networks and demonstrated their causal effects on behaviour in distinct processing windows. While these mechanisms have been conceptualized as control signals for suppressing impulsive response tendencies in conflict tasks and as decision threshold adjustments in value-based and perceptual decisions, we propose a common framework linking decision-making, cognition and movement. Within this framework subthalamic deep brain stimulation can lead to suboptimal choices by reducing the time that patients take for deliberation before committing to an action. However, clinical studies have consistently shown that the occurrence of impulse control disorders is reduced, not increased, after subthalamic deep brain stimulation surgery. This apparent contradiction can be reconciled when recognizing the multifaceted nature of impulsivity, its underlying mechanisms and modulation by treatment. While subthalamic deep brain stimulation renders patients susceptible to making decisions without proper forethought, this can be disentangled from effects related to dopamine comprising sensitivity to benefits vs. costs, reward delay aversion and learning from outcomes. Alterations in these dopamine-mediated mechanisms are thought to underlie the development of impulse control disorders, and can be relatively spared with reduced dopaminergic medication after subthalamic deep brain stimulation. Together, results from studies using deep brain stimulation as an experimental tool have improved our understanding of action control in the human brain and have important implications for treatment of patients with Neurological disorders.

Neural Index of Reinforcement Learning Predicts Improved Stimulus-Response Retention under High Working Memory Load.

Human learning and decision-making are supported by multiple systems operating in parallel. Recent studies isolating the contributions of reinforcement learning (RL) and working memory (WM) have revealed a trade-off between the two. An interactive WM/RL computational model predicts that although high WM load slows behavioral acquisition, it also induces larger prediction errors in the RL system that enhance robustness and retention of learned behaviors. Here, we tested this account by parametrically manipulating WM load during RL in conjunction with EEG in both male and female participants and administered two surprise memory tests. We further leveraged single-trial decoding of EEG signatures of RL and WM to determine whether their interaction predicted robust retention. Consistent with the model, behavioral learning was slower for associations acquired under higher load but showed parametrically improved future retention. This paradoxical result was mirrored by EEG indices of RL, which were strengthened under higher WM loads and predictive of more robust future behavioral retention of learned stimulus-response contingencies. We further tested whether stress alters the ability to shift between the two systems strategically to maximize immediate learning versus retention of information and found that induced stress had only a limited effect on this trade-off. The present results offer a deeper understanding of the cooperative interaction between WM and RL and show that relying on WM can benefit the rapid acquisition of choice behavior during learning but impairs retention.SIGNIFICANCE STATEMENT Successful learning is achieved by the joint contribution of the dopaminergic RL system and WM. The cooperative WM/RL model was productive in improving our understanding of the interplay between the two systems during learning, demonstrating that reliance on RL computations is modulated by WM load. However, the role of WM/RL systems in the retention of learned stimulus-response associations remained unestablished. Our results show that increased neural signatures of learning, indicative of greater RL computation, under high WM load also predicted better stimulus-response retention. This result supports a trade-off between the two systems, where degraded WM increases RL processing, which improves retention. Notably, we show that this cooperative interplay remains largely unaffected by acute stress.

Is human compositionality meta-learned?

Recent studies suggest that meta-learning may provide an original solution to an enduring puzzle about whether neural networks can explain compositionality - in particular, by raising the prospect that compositionality can be understood as an emergent property of an inner-loop learning algorithm. We elaborate on this hypothesis and consider its empirical predictions regarding the neural mechanisms and development of human compositionality.

Biophysical and Architectural Mechanisms of Subthalamic Theta under Response Conflict.

The cortico-basal ganglia circuit is needed to suppress prepotent actions and to facilitate controlled behavior. Under conditions of response conflict, the frontal cortex and subthalamic nucleus (STN) exhibit increased spiking and theta band power, which are linked to adaptive regulation of behavioral output. The electrophysiological mechanisms underlying these neural signatures of impulse control remain poorly understood. To address this lacuna, we constructed a novel large-scale, biophysically principled model of the subthalamopallidal (STN-globus pallidus externus) network and examined the mechanisms that modulate theta power and spiking in response to cortical input. Simulations confirmed that theta power does not emerge from intrinsic network dynamics but is robustly elicited in response to cortical input as burst events representing action selection dynamics. Rhythmic burst events of multiple cortical populations, representing a state of conflict where cortical motor plans vacillate in the theta range, led to prolonged STN theta and increased spiking, consistent with empirical literature. Notably, theta band signaling required NMDA, but not AMPA, currents, which were in turn related to a triphasic STN response characterized by spiking, silence, and bursting periods. Finally, theta band resonance was also strongly modulated by architectural connectivity, with maximal theta arising when multiple cortical populations project to individual STN "conflict detector" units because of an NMDA-dependent supralinear response. Our results provide insights into the biophysical principles and architectural constraints that give rise to STN dynamics during response conflict, and how their disruption can lead to impulsivity and compulsivity.SIGNIFICANCE STATEMENT The subthalamic nucleus exhibits theta band power modulation related to cognitive control over motor actions during conditions of response conflict. However, the mechanisms of such dynamics are not understood. Here we developed a novel biophysically detailed and data-constrained large-scale model of the subthalamopallidal network, and examined the impacts of cellular and network architectural properties that give rise to theta dynamics. Our investigations implicate an important role for NMDA receptors and cortico-subthalamic nucleus topographical connectivities in theta power modulation.

Self-judgment dissected: A computational modeling analysis of self-referential processing and its relationship to trait mindfulness facets and depression symptoms.

Cognitive theories of depression, and mindfulness theories of well-being, converge on the notion that self-judgment plays a critical role in mental health. However, these theories have rarely been tested via tasks and computational modeling analyses that can disentangle the information processes operative in self-judgments. We applied a drift-diffusion computational model to the self-referential encoding task (SRET) collected before and after an 8-week mindfulness intervention (n = 96). A drift-rate regression parameter representing positive-relative to negative-self-referential judgment strength positively related to mindful awareness and inversely related to depression, both at baseline and over time; however, this parameter did not significantly relate to the interaction between mindful awareness and nonjudgmentalness. At the level of individual depression symptoms, at baseline, a spectrum of symptoms (inversely) correlated with the drift-rate regression parameter, suggesting that many distinct depression symptoms relate to valenced self-judgment between subjects. By contrast, over the intervention, changes in only a smaller subset of anhedonia-related depression symptoms showed substantial relationships with this parameter. Both behavioral and model-derived measures showed modest split-half and test-retest correlations. Results support cognitive theories that implicate self-judgment in depression and mindfulness theories, which imply that mindful awareness should lead to more positive self-views.

Thunderstruck: The ACDC model of flexible sequences and rhythms in recurrent neural circuits.

Adaptive sequential behavior is a hallmark of human cognition. In particular, humans can learn to produce precise spatiotemporal sequences given a certain context. For instance, musicians can not only reproduce learned action sequences in a context-dependent manner, they can also quickly and flexibly reapply them in any desired tempo or rhythm without overwriting previous learning. Existing neural network models fail to account for these properties. We argue that this limitation emerges from the fact that sequence information (i.e., the position of the action) and timing (i.e., the moment of response execution) are typically stored in the same neural network weights. Here, we augment a biologically plausible recurrent neural network of cortical dynamics to include a basal ganglia-thalamic module which uses reinforcement learning to dynamically modulate action. This "associative cluster-dependent chain" (ACDC) model modularly stores sequence and timing information in distinct loci of the network. This feature increases computational power and allows ACDC to display a wide range of temporal properties (e.g., multiple sequences, temporal shifting, rescaling, and compositionality), while still accounting for several behavioral and neurophysiological empirical observations. Finally, we apply this ACDC network to show how it can learn the famous "Thunderstruck" song intro and then flexibly play it in a "bossa nova" rhythm without further training.

Computational Psychiatry Needs Time and Context.

Why has computational psychiatry yet to influence routine clinical practice? One reason may be that it has neglected context and temporal dynamics in the models of certain mental health problems. We develop three heuristics for estimating whether time and context are important to a mental health problem: Is it characterized by a core neurobiological mechanism? Does it follow a straightforward natural trajectory? And is intentional mental content peripheral to the problem? For many problems the answers are no, suggesting that modeling time and context is critical. We review computational psychiatry advances toward this end, including modeling state variation, using domain-specific stimuli, and interpreting differences in context. We discuss complementary network and complex systems approaches. Novel methods and unification with adjacent fields may inspire a new generation of computational psychiatry.

Beyond Drift Diffusion Models: Fitting a Broad Class of Decision and Reinforcement Learning Models with HDDM.

Computational modeling has become a central aspect of research in the cognitive neurosciences. As the field matures, it is increasingly important to move beyond standard models to quantitatively assess models with richer dynamics that may better reflect underlying cognitive and neural processes. For example, sequential sampling models (SSMs) are a general class of models of decision-making intended to capture processes jointly giving rise to RT distributions and choice data in n-alternative choice paradigms. A number of model variations are of theoretical interest, but empirical data analysis has historically been tied to a small subset for which likelihood functions are analytically tractable. Advances in methods designed for likelihood-free inference have recently made it computationally feasible to consider a much larger spectrum of SSMs. In addition, recent work has motivated the combination of SSMs with reinforcement learning models, which had historically been considered in separate literatures. Here, we provide a significant addition to the widely used HDDM Python toolbox and include a tutorial for how users can easily fit and assess a (user-extensible) wide variety of SSMs and how they can be combined with reinforcement learning models. The extension comes batteries included, including model visualization tools, posterior predictive checks, and ability to link trial-wise neural signals with model parameters via hierarchical Bayesian regression.

Analogous computations in working memory input, output and motor gating: Electrophysiological and computational modeling evidence.

Adaptive cognitive-control involves a hierarchical cortico-striatal gating system that supports selective updating, maintenance, and retrieval of useful cognitive and motor information. Here, we developed a task that independently manipulates selective gating operations into working-memory (input gating), from working-memory (output gating), and of responses (motor gating) and tested the neural dynamics and computational principles that support them. Increases in gating demands, captured by gate switches, were expressed by distinct EEG correlates at each gating level that evolved dynamically in partially overlapping time windows. Further, categorical representations of specific maintained items and of motor responses could be decoded from EEG when the corresponding gate was switching, thereby linking gating operations to prioritization. Finally, gate switching at all levels was related to increases in the motor decision threshold as quantified by the drift diffusion model. Together these results support the notion that cognitive gating operations scaffold on top of mechanisms involved in motor gating.

Computational phenotyping of brain-behavior dynamics underlying approach-avoidance conflict in major depressive disorder.

Adaptive behavior requires balancing approach and avoidance based on the rewarding and aversive consequences of actions. Imbalances in this evaluation are thought to characterize mood disorders such as major depressive disorder (MDD). We present a novel application of the drift diffusion model (DDM) suited to quantify how offers of reward and aversiveness, and neural correlates thereof, are dynamically integrated to form decisions, and how such processes are altered in MDD. Hierarchical parameter estimation from the DDM demonstrated that the MDD group differed in three distinct reward-related parameters driving approach-based decision making. First, MDD was associated with reduced reward sensitivity, measured as the impact of offered reward on evidence accumulation. Notably, this effect was replicated in a follow-up study. Second, the MDD group showed lower starting point bias towards approaching offers. Third, this starting point was influenced in opposite directions by Pavlovian effects and by nucleus accumbens activity across the groups: greater accumbens activity was related to approach bias in controls but avoid bias in MDD. Cross-validation revealed that the combination of these computational biomarkers were diagnostic of patient status, with accumbens influences being particularly diagnostic. Finally, within the MDD group, reward sensitivity and nucleus accumbens parameters were differentially related to symptoms of perceived stress and depression. Collectively, these findings establish the promise of computational psychiatry approaches to dissecting approach-avoidance decision dynamics relevant for affective disorders.

All or nothing belief updating in patients with schizophrenia reduces precision and flexibility of beliefs.

Schizophrenia is characterized by abnormal perceptions and beliefs, but the computational mechanisms through which these abnormalities emerge remain unclear. One prominent hypothesis asserts that such abnormalities result from overly precise representations of prior knowledge, which in turn lead beliefs to become insensitive to feedback. In contrast, another prominent hypothesis asserts that such abnormalities result from a tendency to interpret prediction errors as indicating meaningful change, leading to the assignment of aberrant salience to noisy or misleading information. Here we examine behaviour of patients and control subjects in a behavioural paradigm capable of adjudicating between these competing hypotheses and characterizing belief updates directly on individual trials. We show that patients are more prone to completely ignoring new information and perseverating on previous responses, but when they do update, tend to do so completely. This updating strategy limits the integration of information over time, reducing both the flexibility and precision of beliefs and provides a potential explanation for how patients could simultaneously show over-sensitivity and under-sensitivity to feedback in different paradigms.

Hierarchical clustering optimizes the tradeoff between compositionality and expressivity of task structures for flexible reinforcement learning.

A hallmark of human intelligence, but challenging for reinforcement learning (RL) agents, is the ability to compositionally generalise, that is, to recompose familiar knowledge components in novel ways to solve new problems. For instance, when navigating in a city, one needs to know the location of the destination and how to operate a vehicle to get there, whether it be pedalling a bike or operating a car. In RL, these correspond to the reward function and transition function, respectively. To compositionally generalize, these two components need to be transferable independently of each other: multiple modes of transport can reach the same goal, and any given mode can be used to reach multiple destinations. Yet there are also instances where it can be helpful to learn and transfer entire structures, jointly representing goals and transitions, particularly whenever these recur in natural tasks (e.g., given a suggestion to get ice cream, one might prefer to bike, even in new towns). Prior theoretical work has explored how, in model-based RL, agents can learn and generalize task components (transition and reward functions). But a satisfactory account for how a single agent can simultaneously satisfy the two competing demands is still lacking. Here, we propose a hierarchical RL agent that learns and transfers individual task components as well as entire structures (particular compositions of components) by inferring both through a non-parametric Bayesian model of the task. It maintains a factorised representation of task components through a hierarchical Dirichlet process, but it also represents different possible covariances between these components through a standard Dirichlet process. We validate our approach on a variety of navigation tasks covering a wide range of statistical correlations between task components and show that it can also improve generalisation and transfer in more complex, hierarchical tasks with goal/subgoal structures. Finally, we end with a discussion of our work including how this clustering algorithm could conceivably be implemented by cortico-striatal gating circuits in the brain.

Social reinforcement learning as a predictor of real-life experiences in individuals with high and low depressive symptomatology.

BACKGROUND: Several studies have reported diminished learning from non-social outcomes in depressed individuals. However, it is not clear how depression impacts learning from social feedback. Notably, mood disorders are commonly associated with deficits in social functioning, which raises the possibility that potential impairments in social learning may negatively affect real-life social experiences in depressed subjects. METHODS: Ninety-two participants with high (HD; N = 40) and low (LD; N = 52) depression scores were recruited. Subjects performed a learning task, during which they received monetary outcomes or social feedback which they were told came from other people. Additionally, participants answered questions about their everyday social experiences. Computational models were fit to the data and model parameters were related to social experience measures. RESULTS: HD subjects reported a reduced quality and quantity of social experiences compared to LD controls, including an increase in the amount of time spent in negative social situations. Moreover, HD participants showed lower learning rates than LD subjects in the social condition of the task. Interestingly, across all participants, reduced social learning rates predicted higher amounts of time spent in negative social situations, even when depression scores were controlled for. CONCLUSION: These findings indicate that deficits in social learning may affect the quality of everyday social experiences. Specifically, the impaired ability to use social feedback to appropriately update future actions, which was observed in HD subjects, may lead to suboptimal interpersonal behavior in real life. This, in turn, may evoke negative feedback from others, thus bringing about more unpleasant social encounters.

Frontal network dynamics reflect neurocomputational mechanisms for reducing maladaptive biases in motivated action.

Motivation exerts control over behavior by eliciting Pavlovian responses, which can either match or conflict with instrumental action. We can overcome maladaptive motivational influences putatively through frontal cognitive control. However, the neurocomputational mechanisms subserving this control are unclear; does control entail up-regulating instrumental systems, down-regulating Pavlovian systems, or both? We combined electroencephalography (EEG) recordings with a motivational Go/NoGo learning task (N = 34), in which multiple Go options enabled us to disentangle selective action learning from nonselective Pavlovian responses. Midfrontal theta-band (4 Hz-8 Hz) activity covaried with the level of Pavlovian conflict and was associated with reduced Pavlovian biases rather than reduced instrumental learning biases. Motor and lateral prefrontal regions synchronized to the midfrontal cortex, and these network dynamics predicted the reduction of Pavlovian biases over and above local, midfrontal theta activity. This work links midfrontal processing to detecting Pavlovian conflict and highlights the importance of network processing in reducing the impact of maladaptive, Pavlovian biases.

Reward-predictive representations generalize across tasks in reinforcement learning.

In computer science, reinforcement learning is a powerful framework with which artificial agents can learn to maximize their performance for any given Markov decision process (MDP). Advances over the last decade, in combination with deep neural networks, have enjoyed performance advantages over humans in many difficult task settings. However, such frameworks perform far less favorably when evaluated in their ability to generalize or transfer representations across different tasks. Existing algorithms that facilitate transfer typically are limited to cases in which the transition function or the optimal policy is portable to new contexts, but achieving "deep transfer" characteristic of human behavior has been elusive. Such transfer typically requires discovery of abstractions that permit analogical reuse of previously learned representations to superficially distinct tasks. Here, we demonstrate that abstractions that minimize error in predictions of reward outcomes generalize across tasks with different transition and reward functions. Such reward-predictive representations compress the state space of a task into a lower dimensional representation by combining states that are equivalent in terms of both the transition and reward functions. Because only state equivalences are considered, the resulting state representation is not tied to the transition and reward functions themselves and thus generalizes across tasks with different reward and transition functions. These results contrast with those using abstractions that myopically maximize reward in any given MDP and motivate further experiments in humans and animals to investigate if neural and cognitive systems involved in state representation perform abstractions that facilitate such equivalence relations.

Generalizing to generalize: Humans flexibly switch between compositional and conjunctive structures during reinforcement learning.

Humans routinely face novel environments in which they have to generalize in order to act adaptively. However, doing so involves the non-trivial challenge of deciding which aspects of a task domain to generalize. While it is sometimes appropriate to simply re-use a learned behavior, often adaptive generalization entails recombining distinct components of knowledge acquired across multiple contexts. Theoretical work has suggested a computational trade-off in which it can be more or less useful to learn and generalize aspects of task structure jointly or compositionally, depending on previous task statistics, but it is unknown whether humans modulate their generalization strategy accordingly. Here we develop a series of navigation tasks that separately manipulate the statistics of goal values ("what to do") and state transitions ("how to do it") across contexts and assess whether human subjects generalize these task components separately or conjunctively. We find that human generalization is sensitive to the statistics of the previously experienced task domain, favoring compositional or conjunctive generalization when the task statistics are indicative of such structures, and a mixture of the two when they are more ambiguous. These results support a normative "meta-generalization" account and suggests that people not only generalize previous task components but also generalize the statistical structure most likely to support generalization.

Within- and across-trial dynamics of human EEG reveal cooperative interplay between reinforcement learning and working memory.

Learning from rewards and punishments is essential to survival and facilitates flexible human behavior. It is widely appreciated that multiple cognitive and reinforcement learning systems contribute to decision-making, but the nature of their interactions is elusive. Here, we leverage methods for extracting trial-by-trial indices of reinforcement learning (RL) and working memory (WM) in human electro-encephalography to reveal single-trial computations beyond that afforded by behavior alone. Neural dynamics confirmed that increases in neural expectation were predictive of reduced neural surprise in the following feedback period, supporting central tenets of RL models. Within- and cross-trial dynamics revealed a cooperative interplay between systems for learning, in which WM contributes expectations to guide RL, despite competition between systems during choice. Together, these results provide a deeper understanding of how multiple neural systems interact for learning and decision-making and facilitate analysis of their disruption in clinical populations.

Anxiety Impedes Adaptive Social Learning Under Uncertainty.

Very little is known about how individuals learn under uncertainty when other people are involved. We propose that humans are particularly tuned to social uncertainty, which is especially noisy and ambiguous. Individuals exhibiting less tolerance for uncertainty, such as those with anxiety, may have greater difficulty learning in uncertain social contexts and therefore provide an ideal test population to probe learning dynamics under uncertainty. Using a dynamic trust game and a matched nonsocial task, we found that healthy subjects (n = 257) were particularly good at learning under negative social uncertainty, swiftly figuring out when to stop investing in an exploitative social partner. In contrast, subjects with anxiety (n = 97) overinvested in exploitative partners. Computational modeling attributed this pattern to a selective reduction in learning from negative social events and a failure to enhance learning as uncertainty rises-two mechanisms that likely facilitate adaptive social choice.

Dissecting the impact of depression on decision-making.

BACKGROUND: Cognitive deficits in depressed adults may reflect impaired decision-making. To investigate this possibility, we analyzed data from unmedicated adults with Major Depressive Disorder (MDD) and healthy controls as they performed a probabilistic reward task. The Hierarchical Drift Diffusion Model (HDDM) was used to quantify decision-making mechanisms recruited by the task, to determine if any such mechanism was disrupted by depression. METHODS: Data came from two samples (Study 1: 258 MDD, 36 controls; Study 2: 23 MDD, 25 controls). On each trial, participants indicated which of two similar stimuli was presented; correct identifications were rewarded. Quantile-probability plots and the HDDM quantified the impact of MDD on response times (RT), speed of evidence accumulation (drift rate), and the width of decision thresholds, among other parameters. RESULTS: RTs were more positively skewed in depressed v. healthy adults, and the HDDM revealed that drift rates were reduced-and decision thresholds were wider-in the MDD groups. This pattern suggests that depressed adults accumulated the evidence needed to make decisions more slowly than controls did. CONCLUSIONS: Depressed adults responded slower than controls in both studies, and poorer performance led the MDD group to receive fewer rewards than controls in Study 1. These results did not reflect a sensorimotor deficit but were instead due to sluggish evidence accumulation. Thus, slowed decision-making-not slowed perception or response execution-caused the performance deficit in MDD. If these results generalize to other tasks, they may help explain the broad cognitive deficits seen in depression.