Generative replay underlies compositional inference in the hippocampal-prefrontal circuit.

Human reasoning depends on reusing pieces of information by putting them together in new ways. However, very little is known about how compositional computation is implemented in the brain. Here, we ask participants to solve a series of problems that each require constructing a whole from a set of elements. With fMRI, we find that representations of novel constructed objects in the frontal cortex and hippocampus are relational and compositional. With MEG, we find that replay assembles elements into compounds, with each replay sequence constituting a hypothesis about a possible configuration of elements. The content of sequences evolves as participants solve each puzzle, progressing from predictable to uncertain elements and gradually converging on the correct configuration. Together, these results suggest a computational bridge between apparently distinct functions of hippocampal-prefrontal circuitry and a role for generative replay in compositional inference and hypothesis testing.

A virtual rodent predicts the structure of neural activity across behaviours.

Animals have exquisite control of their bodies, allowing them to perform a diverse range of behaviours. How such control is implemented by the brain, however, remains unclear. Advancing our understanding requires models that can relate principles of control to the structure of neural activity in behaving animals. Here, to facilitate this, we built a 'virtual rodent', in which an artificial neural network actuates a biomechanically realistic model of the rat1 in a physics simulator2. We used deep reinforcement learning3-5 to train the virtual agent to imitate the behaviour of freely moving rats, thus allowing us to compare neural activity recorded in real rats to the network activity of a virtual rodent mimicking their behaviour. We found that neural activity in the sensorimotor striatum and motor cortex was better predicted by the virtual rodent's network activity than by any features of the real rat's movements, consistent with both regions implementing inverse dynamics6. Furthermore, the network's latent variability predicted the structure of neural variability across behaviours and afforded robustness in a way consistent with the minimal intervention principle of optimal feedback control7. These results demonstrate how physical simulation of biomechanically realistic virtual animals can help interpret the structure of neural activity across behaviour and relate it to theoretical principles of motor control.

A distributional code for value in dopamine-based reinforcement learning.

Since its introduction, the reward prediction error theory of dopamine has explained a wealth of empirical phenomena, providing a unifying framework for understanding the representation of reward and value in the brain1-3. According to the now canonical theory, reward predictions are represented as a single scalar quantity, which supports learning about the expectation, or mean, of stochastic outcomes. Here we propose an account of dopamine-based reinforcement learning inspired by recent artificial intelligence research on distributional reinforcement learning4-6. We hypothesized that the brain represents possible future rewards not as a single mean, but instead as a probability distribution, effectively representing multiple future outcomes simultaneously and in parallel. This idea implies a set of empirical predictions, which we tested using single-unit recordings from mouse ventral tegmental area. Our findings provide strong evidence for a neural realization of distributional reinforcement learning.

Toward a Rational and Mechanistic Account of Mental Effort.

In spite of its familiar phenomenology, the mechanistic basis for mental effort remains poorly understood. Although most researchers agree that mental effort is aversive and stems from limitations in our capacity to exercise cognitive control, it is unclear what gives rise to those limitations and why they result in an experience of control as costly. The presence of these control costs also raises further questions regarding how best to allocate mental effort to minimize those costs and maximize the attendant benefits. This review explores recent advances in computational modeling and empirical research aimed at addressing these questions at the level of psychological process and neural mechanism, examining both the limitations to mental effort exertion and how we manage those limited cognitive resources. We conclude by identifying remaining challenges for theoretical accounts of mental effort as well as possible applications of the available findings to understanding the causes of and potential solutions for apparent failures to exert the mental effort required of us.

Meta-Learned Models of Cognition.

Psychologists and neuroscientists extensively rely on computational models for studying and analyzing the human mind. Traditionally, such computational models have been hand-designed by expert researchers. Two prominent examples are cognitive architectures and Bayesian models of cognition. While the former requires the specification of a fixed set of computational structures and a definition of how these structures interact with each other, the latter necessitates the commitment to a particular prior and a likelihood function which - in combination with Bayes' rule - determine the model's behavior. In recent years, a new framework has established itself as a promising tool for building models of human cognition: the framework of meta-learning. In contrast to the previously mentioned model classes, meta-learned models acquire their inductive biases from experience, i.e., by repeatedly interacting with an environment. However, a coherent research program around meta-learned models of cognition is still missing to this day. The purpose of this article is to synthesize previous work in this field and establish such a research program. We accomplish this by pointing out that meta-learning can be used to construct Bayes-optimal learning algorithms, allowing us to draw strong connections to the rational analysis of cognition. We then discuss several advantages of the meta-learning framework over traditional methods and reexamine prior work in the context of these new insights.

Compositional Sequence Generation in the Entorhinal-Hippocampal System.

Neurons in the medial entorhinal cortex exhibit multiple, periodically organized, firing fields which collectively appear to form an internal representation of space. Neuroimaging data suggest that this grid coding is also present in other cortical areas such as the prefrontal cortex, indicating that it may be a general principle of neural functionality in the brain. In a recent analysis through the lens of dynamical systems theory, we showed how grid coding can lead to the generation of a diversity of empirically observed sequential reactivations of hippocampal place cells corresponding to traversals of cognitive maps. Here, we extend this sequence generation model by describing how the synthesis of multiple dynamical systems can support compositional cognitive computations. To empirically validate the model, we simulate two experiments demonstrating compositionality in space or in time during sequence generation. Finally, we describe several neural network architectures supporting various types of compositionality based on grid coding and highlight connections to recent work in machine learning leveraging analogous techniques.

Subgoal- and Goal-related Reward Prediction Errors in Medial Prefrontal Cortex.

A longstanding view of the organization of human and animal behavior holds that behavior is hierarchically organized-in other words, directed toward achieving superordinate goals through the achievement of subordinate goals or subgoals. However, most research in neuroscience has focused on tasks without hierarchical structure. In past work, we have shown that negative reward prediction error (RPE) signals in medial prefrontal cortex (mPFC) can be linked not only to superordinate goals but also to subgoals. This suggests that mPFC tracks impediments in the progression toward subgoals. Using fMRI of human participants engaged in a hierarchical navigation task, here we found that mPFC also processes positive prediction errors at the level of subgoals, indicating that this brain region is sensitive to advances in subgoal completion. However, when subgoal RPEs were elicited alongside with goal-related RPEs, mPFC responses reflected only the goal-related RPEs. These findings suggest that information from different levels of hierarchy is processed selectively, depending on the task context.

Predictive representations can link model-based reinforcement learning to model-free mechanisms.

Humans and animals are capable of evaluating actions by considering their long-run future rewards through a process described using model-based reinforcement learning (RL) algorithms. The mechanisms by which neural circuits perform the computations prescribed by model-based RL remain largely unknown; however, multiple lines of evidence suggest that neural circuits supporting model-based behavior are structurally homologous to and overlapping with those thought to carry out model-free temporal difference (TD) learning. Here, we lay out a family of approaches by which model-based computation may be built upon a core of TD learning. The foundation of this framework is the successor representation, a predictive state representation that, when combined with TD learning of value predictions, can produce a subset of the behaviors associated with model-based learning, while requiring less decision-time computation than dynamic programming. Using simulations, we delineate the precise behavioral capabilities enabled by evaluating actions using this approach, and compare them to those demonstrated by biological organisms. We then introduce two new algorithms that build upon the successor representation while progressively mitigating its limitations. Because this framework can account for the full range of observed putatively model-based behaviors while still utilizing a core TD framework, we suggest that it represents a neurally plausible family of mechanisms for model-based evaluation.

Evidence integration in model-based tree search.

Research on the dynamics of reward-based, goal-directed decision making has largely focused on simple choice, where participants decide among a set of unitary, mutually exclusive options. Recent work suggests that the deliberation process underlying simple choice can be understood in terms of evidence integration: Noisy evidence in favor of each option accrues over time, until the evidence in favor of one option is significantly greater than the rest. However, real-life decisions often involve not one, but several steps of action, requiring a consideration of cumulative rewards and a sensitivity to recursive decision structure. We present results from two experiments that leveraged techniques previously applied to simple choice to shed light on the deliberation process underlying multistep choice. We interpret the results from these experiments in terms of a new computational model, which extends the evidence accumulation perspective to multiple steps of action.

Building machines that learn and think for themselves.

We agree with Lake and colleagues on their list of "key ingredients" for building human-like intelligence, including the idea that model-based reasoning is essential. However, we favor an approach that centers on one additional ingredient: autonomy. In particular, we aim toward agents that can both build and exploit their own internal models, with minimal human hand engineering. We believe an approach centered on autonomous learning has the greatest chance of success as we scale toward real-world complexity, tackling domains for which ready-made formal models are not available. Here, we survey several important examples of the progress that has been made toward building autonomous agents with human-like abilities, and highlight some outstanding challenges.

Statistical learning of temporal community structure in the hippocampus.

The hippocampus is involved in the learning and representation of temporal statistics, but little is understood about the kinds of statistics it can uncover. Prior studies have tested various forms of structure that can be learned by tracking the strength of transition probabilities between adjacent items in a sequence. We test whether the hippocampus can learn higher-order structure using sequences that have no variance in transition probability and instead exhibit temporal community structure. We find that the hippocampus is indeed sensitive to this form of structure, as revealed by its representations, activity dynamics, and connectivity with other regions. These findings suggest that the hippocampus is a sophisticated learner of environmental regularities, able to uncover higher-order structure that requires sensitivity to overlapping associations.

Dorsal anterior cingulate and ventromedial prefrontal cortex have inverse roles in both foraging and economic choice.

Recent research has highlighted a distinction between sequential foraging choices and traditional economic choices between simultaneously presented options. This was partly motivated by observations in Kolling, Behrens, Mars, and Rushworth, Science, 336(6077), 95-98 (2012) (hereafter, KBMR) that these choice types are subserved by different circuits, with dorsal anterior cingulate (dACC) preferentially involved in foraging and ventromedial prefrontal cortex (vmPFC) preferentially involved in economic choice. To support this account, KBMR used fMRI to scan human subjects making either a foraging choice (between exploiting a current offer or swapping for potentially better rewards) or an economic choice (between two reward-probability pairs). This study found that dACC better tracked values pertaining to foraging, whereas vmPFC better tracked values pertaining to economic choice. We recently showed that dACC's role in these foraging choices is better described by the difficulty of choosing than by foraging value, when correcting for choice biases and testing a sufficiently broad set of foraging values (Shenhav, Straccia, Cohen, & Botvinick Nature Neuroscience, 17(9), 1249-1254, 2014). Here, we extend these findings in 3 ways. First, we replicate our original finding with a larger sample and a task modified to address remaining methodological gaps between our previous experiments and that of KBMR. Second, we show that dACC activity is best accounted for by choice difficulty alone (rather than in combination with foraging value) during both foraging and economic choices. Third, we show that patterns of vmPFC activity, inverted relative to dACC, also suggest a common function across both choice types. Overall, we conclude that both regions are similarly engaged by foraging-like and economic choice.

Irrational time allocation in decision-making.

Time is an extremely valuable resource but little is known about the efficiency of time allocation in decision-making. Empirical evidence suggests that in many ecologically relevant situations, decision difficulty and the relative reward from making a correct choice, compared to an incorrect one, are inversely linked, implying that it is optimal to use relatively less time for difficult choice problems. This applies, in particular, to value-based choices, in which the relative reward from choosing the higher valued item shrinks as the values of the other options get closer to the best option and are thus more difficult to discriminate. Here, we experimentally show that people behave sub-optimally in such contexts. They do not respond to incentives that favour the allocation of time to choice problems in which the relative reward for choosing the best option is high; instead they spend too much time on problems in which the reward difference between the options is low. We demonstrate this by showing that it is possible to improve subjects' time allocation with a simple intervention that cuts them off when their decisions take too long. Thus, we provide a novel form of evidence that organisms systematically spend their valuable time in an inefficient way, and simultaneously offer a potential solution to the problem.

A comparative evaluation of off-the-shelf distributed semantic representations for modelling behavioural data.

In this paper we carry out an extensive comparison of many off-the-shelf distributed semantic vectors representations of words, for the purpose of making predictions about behavioural results or human annotations of data. In doing this comparison we also provide a guide for how vector similarity computations can be used to make such predictions, and introduce many resources available both in terms of datasets and of vector representations. Finally, we discuss the shortcomings of this approach and future research directions that might address them.

Motivation and cognitive control: from behavior to neural mechanism.

Research on cognitive control and executive function has long recognized the relevance of motivational factors. Recently, however, the topic has come increasingly to center stage, with a surge of new studies examining the interface of motivation and cognitive control. In the present article we survey research situated at this interface, considering work from cognitive and social psychology and behavioral economics, but with a particular focus on neuroscience research. We organize existing findings into three core areas, considering them in the light of currently vying theoretical perspectives. Based on the accumulated evidence, we advocate for a view of control function that treats it as a domain of reward-based decision making. More broadly, we argue that neuroscientific evidence plays a critical role in understanding the mechanisms by which motivation and cognitive control interact. Opportunities for further cross-fertilization between behavioral and neuroscientific research are highlighted.

Cognitive effort avoidance and detection in people with schizophrenia.

Many people with schizophrenia exhibit avolition, a difficulty initiating and maintaining goal-directed behavior, considered to be a key negative symptom of the disorder. Recent evidence indicates that patients with higher levels of negative symptoms differ from healthy controls in showing an exaggerated cost of the physical effort needed to obtain a potential reward. We examined whether patients show an exaggerated avoidance of cognitive effort, using the demand selection task developed by Kool, McGuire, Rosen, and Botvinick (Journal of Experimental Psychology. General, 139, 665-682, 2010). A total of 83 people with schizophrenia or schizoaffective disorder and 71 healthy volunteers participated in three experiments where instructions varied. In the standard task (Experiment 1), neither controls nor patients showed expected cognitive demand avoidance. With enhanced instructions (Experiment 2), controls demonstrated greater demand avoidance than patients. In Experiment 3, patients showed nonsignificant reductions in demand avoidance, relative to controls. In a control experiment, patients showed significantly reduced ability to detect the effort demands associated with different response alternatives. In both groups, the ability to detect effort demands was associated with increased effort avoidance. In both groups, increased cognitive effort avoidance was associated with higher IQ and general neuropsychological ability. No significant correlations between demand avoidance and negative symptom severity were observed. Thus, it appears that individual differences in general intellectual ability and effort detection are related to cognitive effort avoidance and likely account for the subtle reduction in effort avoidance observed in schizophrenia.

Optimal behavioral hierarchy.

Human behavior has long been recognized to display hierarchical structure: actions fit together into subtasks, which cohere into extended goal-directed activities. Arranging actions hierarchically has well established benefits, allowing behaviors to be represented efficiently by the brain, and allowing solutions to new tasks to be discovered easily. However, these payoffs depend on the particular way in which actions are organized into a hierarchy, the specific way in which tasks are carved up into subtasks. We provide a mathematical account for what makes some hierarchies better than others, an account that allows an optimal hierarchy to be identified for any set of tasks. We then present results from four behavioral experiments, suggesting that human learners spontaneously discover optimal action hierarchies.

Hierarchical learning induces two simultaneous, but separable, prediction errors in human basal ganglia.

Studies suggest that dopaminergic neurons report a unitary, global reward prediction error signal. However, learning in complex real-life tasks, in particular tasks that show hierarchical structure, requires multiple prediction errors that may coincide in time. We used functional neuroimaging to measure prediction error signals in humans performing such a hierarchical task involving simultaneous, uncorrelated prediction errors. Analysis of signals in a priori anatomical regions of interest in the ventral striatum and the ventral tegmental area indeed evidenced two simultaneous, but separable, prediction error signals corresponding to the two levels of hierarchy in the task. This result suggests that suitably designed tasks may reveal a more intricate pattern of firing in dopaminergic neurons. Moreover, the need for downstream separation of these signals implies possible limitations on the number of different task levels that we can learn about simultaneously.

Mechanisms of motivation-cognition interaction: challenges and opportunities.

Recent years have seen a rejuvenation of interest in studies of motivation-cognition interactions arising from many different areas of psychology and neuroscience. The present issue of Cognitive, Affective, & Behavioral Neuroscience provides a sampling of some of the latest research from a number of these different areas. In this introductory article, we provide an overview of the current state of the field, in terms of key research developments and candidate neural mechanisms receiving focused investigation as potential sources of motivation-cognition interaction. However, our primary goal is conceptual: to highlight the distinct perspectives taken by different research areas, in terms of how motivation is defined, the relevant dimensions and dissociations that are emphasized, and the theoretical questions being targeted. Together, these distinctions present both challenges and opportunities for efforts aiming toward a more unified and cross-disciplinary approach. We identify a set of pressing research questions calling for this sort of cross-disciplinary approach, with the explicit goal of encouraging integrative and collaborative investigations directed toward them.