Distinct prediction errors in mesostriatal circuits of the human brain mediate learning about the values of both states and actions: evidence from high-resolution fMRI

Active reinforcement learning versus action bias and hysteresis: control with a mixture of experts and nonexperts

Active reinforcement learning enables dynamic prediction and control, where one should not only maximize rewards but also minimize costs such as of inference, decisions, actions, and time. For an embodied agent such as a human, decisions are also shaped by physical aspects of actions. Beyond the effects of reward outcomes on learning processes, to what extent can modeling of behavior in a reinforcement-learning task be complicated by other sources of variance in sequential action choices? What of the effects of action bias (for actions per se) and action hysteresis determined by the history of actions chosen previously? The present study addressed these questions with incremental assembly of models for the sequential choice data from a task with hierarchical structure for additional complexity in learning. With systematic comparison and falsification of computational models, human choices were tested for signatures of parallel modules representing not only an enhanced form of generalized reinforcement learning but also action bias and hysteresis. We found evidence for substantial differences in bias and hysteresis across participants-even comparable in magnitude to the individual differences in learning. Individuals who did not learn well revealed the greatest biases, but those who did learn accurately were also significantly biased. The direction of hysteresis varied among individuals as repetition or, more commonly, alternation biases persisting from multiple previous actions. Considering that these actions were button presses with trivial motor demands, the idiosyncratic forces biasing sequences of action choices were robust enough to suggest ubiquity across individuals and across tasks requiring various actions. In light of how bias and hysteresis function as a heuristic for efficient control that adapts to uncertainty or low motivation by minimizing the cost of effort, these phenomena broaden the consilient theory of a mixture of experts to encompass a mixture of expert and nonexpert controllers of behavior.

State-specific alterations in the neural computations underlying inhibitory control in women remitted from bulimia nervosa

The neurocomputational processes underlying bulimia nervosa and its primary symptoms, out-of-control overeating and purging, are poorly understood. Research suggests that the brains of healthy individuals form a dynamic internal model to predict whether control is needed in each moment. This study tested the hypothesis that this computational process of inhibitory control is abnormally affected by metabolic state (being fasted or fed) in bulimia nervosa. A Bayesian ideal observer model was fit to behavioral data acquired from 22 women remitted from bulimia nervosa and 20 group-matched controls who completed a stop-signal task during two counterbalanced functional MRI sessions, one after a 16 h fast and one after a meal. This model estimates participants' trial-by-trial updating of the probability of a stop signal based on their experienced trial history. Neural analyses focused on control-related Bayesian prediction errors, which quantify the direction and degree of "surprise" an individual experiences on any given trial. Regardless of group, metabolic state did not affect behavioral performance on the task. However, metabolic state modulated group differences in neural activation. In the fed state, women remitted from bulimia nervosa had attenuated prediction-error-dependent activation in the left dorsal caudate. This fed-state activation was lower among women with more frequent past binge eating and self-induced vomiting. When they are in a fed state, individuals with bulimia nervosa may not effectively process unexpected information needed to engage inhibitory control. This may explain the difficulties these individuals have stopping eating after it begins.

Decision heuristics in contexts integrating action selection and execution

Heuristics can inform human decision making in complex environments through a reduction of computational requirements (accuracy-resource trade-off) and a robustness to overparameterisation (less-is-more). However, tasks capturing the efficiency of heuristics typically ignore action proficiency in determining rewards. The requisite movement parameterisation in sensorimotor control questions whether heuristics preserve efficiency when actions are nontrivial. We developed a novel action selection-execution task requiring joint optimisation of action selection and spatio-temporal skillful execution. State-appropriate choices could be determined by a simple spatial heuristic, or by more complex planning. Computational models of action selection parsimoniously distinguished human participants who adopted the heuristic from those using a more complex planning strategy. Broader comparative analyses then revealed that participants using the heuristic showed combined decisional (selection) and skill (execution) advantages, consistent with a less-is-more framework. In addition, the skill advantage of the heuristic group was predominantly in the core spatial features that also shaped their decision policy, evidence that the dimensions of information guiding action selection might be yoked to salient features in skill learning.

Reinforcement learning with associative or discriminative generalization across states and actions: fMRI at 3 T and 7 T

The model-free algorithms of "reinforcement learning" (RL) have gained clout across disciplines, but so too have model-based alternatives. The present study emphasizes other dimensions of this model space in consideration of associative or discriminative generalization across states and actions. This "generalized reinforcement learning" (GRL) model, a frugal extension of RL, parsimoniously retains the single reward-prediction error (RPE), but the scope of learning goes beyond the experienced state and action. Instead, the generalized RPE is efficiently relayed for bidirectional counterfactual updating of value estimates for other representations. Aided by structural information but as an implicit rather than explicit cognitive map, GRL provided the most precise account of human behavior and individual differences in a reversal-learning task with hierarchical structure that encouraged inverse generalization across both states and actions. Reflecting inference that could be true, false (i.e., overgeneralization), or absent (i.e., undergeneralization), state generalization distinguished those who learned well more so than action generalization. With high-resolution high-field fMRI targeting the dopaminergic midbrain, the GRL model's RPE signals (alongside value and decision signals) were localized within not only the striatum but also the substantia nigra and the ventral tegmental area, including specific effects of generalization that also extend to the hippocampus. Factoring in generalization as a multidimensional process in value-based learning, these findings shed light on complexities that, while challenging classic RL, can still be resolved within the bounds of its core computations.

Reinforcement-learning in fronto-striatal circuits

We review the current state of knowledge on the computational and neural mechanisms of reinforcement-learning with a particular focus on fronto-striatal circuits. We divide the literature in this area into five broad research themes: the target of the learning-whether it be learning about the value of stimuli or about the value of actions; the nature and complexity of the algorithm used to drive the learning and inference process; how learned values get converted into choices and associated actions; the nature of state representations, and of other cognitive machinery that support the implementation of various reinforcement-learning operations. An emerging fifth area focuses on how the brain allocates or arbitrates control over different reinforcement-learning sub-systems or "experts". We will outline what is known about the role of the prefrontal cortex and striatum in implementing each of these functions. We then conclude by arguing that it will be necessary to build bridges from algorithmic level descriptions of computational reinforcement-learning to implementational level models to better understand how reinforcement-learning emerges from multiple distributed neural networks in the brain.

Reward and fictive prediction error signals in ventral striatum: asymmetry between factual and counterfactual processing

Reward prediction error, the difference between the expected and obtained reward, is known to act as a reinforcement learning neural signal. In the current study, we propose a model fitting approach that combines behavioral and neural data to fit computational models of reinforcement learning. Briefly, we penalized subject-specific fitted parameters that moved away too far from the group median, except when that deviation led to an improvement in the model's fit to neural responses. By means of a probabilistic monetary learning task and fMRI, we compared our approach with standard model fitting methods. Q-learning outperformed actor-critic at both behavioral and neural level, although the inclusion of neuroimaging data into model fitting improved the fit of actor-critic models. We observed both action-value and state-value prediction error signals in the striatum, while standard model fitting approaches failed to capture state-value signals. Finally, left ventral striatum correlated with reward prediction error while right ventral striatum with fictive prediction error, suggesting a functional hemispheric asymmetry regarding prediction-error driven learning.

Hypothalamic Interactions with Large-Scale Neural Circuits Underlying Reinforcement Learning and Motivated Behavior

Biological agents adapt behavior to support the survival needs of the individual and the species. In this review we outline the anatomical, physiological, and computational processes that support reinforcement learning (RL). We describe two circuits in the primate brain that are linked to specific aspects of learning and goal-directed behavior. The ventral circuit, that includes the amygdala, ventral medial prefrontal cortex, and ventral striatum, has substantial connectivity with the hypothalamus. The dorsal circuit, that includes inferior parietal cortex, dorsal lateral prefrontal cortex, and the dorsal striatum, has minimal connectivity with the hypothalamus. The hypothalamic connectivity suggests distinct roles for these circuits. We propose that the ventral circuit defines behavioral goals, and the dorsal circuit orchestrates behavior to achieve those goals.

Evidence for model-based encoding of Pavlovian contingencies in the human brain

Prominent accounts of Pavlovian conditioning successfully approximate the frequency and intensity of conditioned responses under the assumption that learning is exclusively model-free; that animals do not develop a cognitive map of events. However, these model-free approximations fall short of comprehensively capturing learning and behavior in Pavlovian conditioning. We therefore performed multivoxel pattern analysis of high-resolution functional MRI data in human participants to test for the encoding of stimulus-stimulus associations that could support model-based computations during Pavlovian conditioning. We found that dissociable sub-regions of the striatum encode predictions of stimulus-stimulus associations and predictive value, in a manner that is directly related to learning performance. Activity patterns in the orbitofrontal cortex were also found to be related to stimulus-stimulus as well as value encoding. These results suggest that the brain encodes model-based representations during Pavlovian conditioning, and that these representations are utilized in the service of behavior.

Reward Learning over Weeks Versus Minutes Increases the Neural Representation of Value in the Human Brain

Over the past few decades, neuroscience research has illuminated the neural mechanisms supporting learning from reward feedback. Learning paradigms are increasingly being extended to study mood and psychiatric disorders as well as addiction. However, one potentially critical characteristic that this research ignores is the effect of time on learning: human feedback learning paradigms are usually conducted in a single rapidly paced session, whereas learning experiences in ecologically relevant circumstances and in animal research are almost always separated by longer periods of time. In our experiments, we examined reward learning in short condensed sessions distributed across weeks versus learning completed in a single "massed" session in male and female participants. As expected, we found that after equal amounts of training, accuracy was matched between the spaced and massed conditions. However, in a 3-week follow-up, we found that participants exhibited significantly greater memory for the value of spaced-trained stimuli. Supporting a role for short-term memory in massed learning, we found a significant positive correlation between initial learning and working memory capacity. Neurally, we found that patterns of activity in the medial temporal lobe and prefrontal cortex showed stronger discrimination of spaced- versus massed-trained reward values. Further, patterns in the striatum discriminated between spaced- and massed-trained stimuli overall. Our results indicate that single-session learning tasks engage partially distinct learning mechanisms from distributed training. Our studies begin to address a large gap in our knowledge of human learning from reinforcement, with potential implications for our understanding of mood disorders and addiction.SIGNIFICANCE STATEMENT Humans and animals learn to associate predictive value with stimuli and actions, and these values then guide future behavior. Such reinforcement-based learning often happens over long time periods, in contrast to most studies of reward-based learning in humans. In experiments that tested the effect of spacing on learning, we found that associations learned in a single massed session were correlated with short-term memory and significantly decayed over time, whereas associations learned in short massed sessions over weeks were well maintained. Additionally, patterns of activity in the medial temporal lobe and prefrontal cortex discriminated the values of stimuli learned over weeks but not minutes. These results highlight the importance of studying learning over time, with potential applications to drug addiction and psychiatry.

A high-resolution probabilistic in vivo atlas of human subcortical brain nuclei

Recent advances in magnetic resonance imaging methods, including data acquisition, pre-processing and analysis, have benefited research on the contributions of subcortical brain nuclei to human cognition and behavior. At the same time, these developments have led to an increasing need for a high-resolution probabilistic in vivo anatomical atlas of subcortical nuclei. In order to address this need, we constructed high spatial resolution, three-dimensional templates, using high-accuracy diffeomorphic registration of T₁- and T₂- weighted structural images from 168 typical adults between 22 and 35 years old. In these templates, many tissue boundaries are clearly visible, which would otherwise be impossible to delineate in data from individual studies. The resulting delineations of subcortical nuclei complement current histology-based atlases. We further created a companion library of software tools for atlas development, to offer an open and evolving resource for the creation of a crowd-sourced in vivo probabilistic anatomical atlas of the human brain.