Explaining distortions in metacognition with an attractor network model of decision uncertainty

Metacognition is the ability to reflect on, and evaluate, our cognition and behaviour. Distortions in metacognition are common in mental health disorders, though the neural underpinnings of such dysfunction are unknown. One reason for this is that models of key components of metacognition, such as decision confidence, are generally specified at an algorithmic or process level. While such models can be used to relate brain function to psychopathology, they are difficult to map to a neurobiological mechanism. Here, we develop a biologically-plausible model of decision uncertainty in an attempt to bridge this gap. We first relate the model’s uncertainty in perceptual decisions to standard metrics of metacognition, namely mean confidence level (bias) and the accuracy of metacognitive judgments (sensitivity). We show that dissociable shifts in metacognition are associated with isolated disturbances at higher-order levels of a circuit associated with self-monitoring, akin to neuropsychological findings that highlight the detrimental effect of prefrontal brain lesions on metacognitive performance. Notably, we are able to account for empirical confidence judgements by fitting the parameters of our biophysical model to first-order performance data, specifically choice and response times. Lastly, in a reanalysis of existing data we show that self-reported mental health symptoms relate to disturbances in an uncertainty-monitoring component of the network. By bridging a gap between a biologically-plausible model of confidence formation and observed disturbances of metacognition in mental health disorders we provide a first step towards mapping theoretical constructs of metacognition onto dynamical models of decision uncertainty. In doing so, we provide a computational framework for modelling metacognitive performance in settings where access to explicit confidence reports is not possible.     

Systematic mistakes are likely in bounded optimal decision-making systems

Systematic mistakes can be distinguished from other types of mistakes in that they are repeatable and predictable within a given organism, and are not due to uncertainty or lack of information. Here we provide a mathematical definition for the concept of systematic mistakes, which captures the way this concept has been used in the behavioral sciences. We also provide an analytical model of information processing networks that are made of large numbers of components, in analogy to the brain being made of a large number of neurons. We show that, for almost all behavioral tasks, and for a wide range of limitations on the computational complexity of the decision-making network, the best possible decision-makers will make systematic mistakes. This result, together with available empirical evidence, suggests that violations of rationality in humans and animals are consistent with natural selection, as the latter operates under constraints.     

The neural basis of metacognitive ability

Ability in various cognitive domains is often assessed by measuring task performance, such as the accuracy of a perceptual categorization. A similar analysis can be applied to metacognitive reports about a task to quantify the degree to which an individual is aware of his or her success or failure. Here, we review the psychological and neural underpinnings of metacognitive accuracy, drawing on research in memory and decision-making. These data show that metacognitive accuracy is dissociable from task performance and varies across individuals. Convergent evidence indicates that the function of the rostral and dorsal aspect of the lateral prefrontal cortex (PFC) is important for the accuracy of retrospective judgements of performance. In contrast, prospective judgements of performance may depend upon medial PFC. We close with a discussion of how metacognitive processes relate to concepts of cognitive control, and propose a neural synthesis in which dorsolateral and anterior prefrontal cortical subregions interact with interoceptive cortices (cingulate and insula) to promote accurate judgements of performance.     

Dynamic Excitatory and Inhibitory Gain Modulation Can Produce Flexible,Robust and Optimal Decision-making

Behavioural and neurophysiological studies in primates have increasingly shown the involvement of urgency signals during the temporal integration of sensory evidence in perceptual decision-making. Neuronal correlates of such signals have been found in the parietal cortex, and in separate studies, demonstrated attention-induced gain modulation of both excitatory and inhibitory neurons. Although previous computational models of decision-making have incorporated gain modulation, their abstract forms do not permit an understanding of the contribution of inhibitory gain modulation. Thus, the effects of co-modulating both excitatory and inhibitory neuronal gains on decision-making dynamics and behavioural performance remain unclear. In this work, we incorporate time-dependent co-modulation of the gains of both excitatory and inhibitory neurons into our previous biologically based decision circuit model. We base our computational study in the context of two classic motion-discrimination tasks performed in animals. Our model shows that by simultaneously increasing the gains of both excitatory and inhibitory neurons, a variety of the observed dynamic neuronal firing activities can be replicated. In particular, the model can exhibit winner-take-all decision-making behaviour with higher firing rates and within a significantly more robust model parameter range. It also exhibits short-tailed reaction time distributions even when operating near a dynamical bifurcation point. The model further shows that neuronal gain modulation can compensate for weaker recurrent excitation in a decision neural circuit, and support decision formation and storage. Higher neuronal gain is also suggested in the more cognitively demanding reaction time than in the fixed delay version of the task. Using the exact temporal delays from the animal experiments, fast recruitment of gain co-modulation is shown to maximize reward rate, with a timescale that is surprisingly near the experimentally fitted value. Our work provides insights into the simultaneous and rapid modulation of excitatory and inhibitory neuronal gains, which enables flexible, robust, and optimal decision-making.     

A link between neuroscience and informatics: large-scale modeling of memory processes

Utilizing advances in functional neuroimaging and computational neural modeling, neuroscientists have increasingly sought to investigate how distributed networks, composed of functionally defined subregions, combine to produce cognition. Large-scale, biologically realistic neural models, which integrate data from cellular, regional, whole brain, and behavioral sources, delineate specific hypotheses about how these interacting neural populations might carry out high-level cognitive tasks. In this review, we discuss neuroimaging, neural modeling, and the utility of large-scale biologically realistic models using modeling of short-term memory as an example. We present a sketch of the data regarding the neural basis of short-term memory from non-human electrophysiological, computational and neuroimaging perspectives, highlighting the multiple interacting brain regions believed to be involved. Through a review of several efforts, including our own, to combine neural modeling and neuroimaging data, we argue that large scale neural models provide specific advantages in understanding the distributed networks underlying cognition and behavior.     

Speed encoding in correlation motion detectors as a consequence of spatial structure

For animals to carry out a wide range of detection, recognition and navigation tasks, visual motion signals are crucial. The encoding of motion information has therefore, attracted much attention in the experimental and computational study of brain function. Two main alternative mechanisms have been proposed on the basis of behavioural and physiological experiments. On one hand, correlation-type and motion energy detectors are simple and efficient in the design of their basic mechanism but are tuned to temporal frequency rather than to speed. On other hand, gradient-type motion detectors directly represent an estimate of speed, but may require more demanding processing mechanisms. We demonstrate here how the temporal frequency dependence observed for sine-wave gratings can disappear for less constrained stimuli, to be replaced by responses reflecting speed for stimuli like square waves when a phase-sensitive detection mechanism is employed. We conclude from these observations that temporal frequency tuning is not necessarily a limitation for motion vision based on correlation detectors, and more generally demonstrate in view of the typical Fourier composition of natural scenes, that correlation detectors operating in such environments can encode image speed. In the context of our results, we discuss the implications of the loss of phase sensitivity inherent in using a linear system approach to describe neural processing.     

Training Excitatory-Inhibitory Recurrent Neural Networks for Cognitive Tasks: A Simple and Flexible Framework

The ability to simultaneously record from large numbers of neurons in behaving animals has ushered in a new era for the study of the neural circuit mechanisms underlying cognitive functions. One promising approach to uncovering the dynamical and computational principles governing population responses is to analyze model recurrent neural networks (RNNs) that have been optimized to perform the same tasks as behaving animals. Because the optimization of network parameters specifies the desired output but not the manner in which to achieve this output, “trained” networks serve as a source of mechanistic hypotheses and a testing ground for data analyses that link neural computation to behavior. Complete access to the activity and connectivity of the circuit, and the ability to manipulate them arbitrarily, make trained networks a convenient proxy for biological circuits and a valuable platform for theoretical investigation. However, existing RNNs lack basic biological features such as the distinction between excitatory and inhibitory units (Dale’s principle), which are essential if RNNs are to provide insights into the operation of biological circuits. Moreover, trained networks can achieve the same behavioral performance but differ substantially in their structure and dynamics, highlighting the need for a simple and flexible framework for the exploratory training of RNNs. Here, we describe a framework for gradient descent-based training of excitatory-inhibitory RNNs that can incorporate a variety of biological knowledge. We provide an implementation based on the machine learning library Theano, whose automatic differentiation capabilities facilitate modifications and extensions. We validate this framework by applying it to well-known experimental paradigms such as perceptual decision-making, context-dependent integration, multisensory integration, parametric working memory, and motor sequence generation. Our results demonstrate the wide range of neural activity patterns and behavior that can be modeled, and suggest a unified setting in which diverse cognitive computations and mechanisms can be studied.     

Encoding time in neural dynamic regimes with distinct computational tradeoffs

Converging evidence suggests the brain encodes time in dynamic patterns of neural activity, including neural sequences, ramping activity, and complex dynamics. Most temporal tasks, however, require more than just encoding time, and can have distinct computational requirements including the need to exhibit temporal scaling, generalize to novel contexts, or robustness to noise. It is not known how neural circuits can encode time and satisfy distinct computational requirements, nor is it known whether similar patterns of neural activity at the population level can exhibit dramatically different computational or generalization properties. To begin to answer these questions, we trained RNNs on two timing tasks based on behavioral studies. The tasks had different input structures but required producing identically timed output patterns. Using a novel framework we quantified whether RNNs encoded two intervals using either of three different timing strategies: scaling, absolute, or stimulus-specific dynamics. We found that similar neural dynamic patterns at the level of single intervals, could exhibit fundamentally different properties, including, generalization, the connectivity structure of the trained networks, and the contribution of excitatory and inhibitory neurons. Critically, depending on the task structure RNNs were better suited for generalization or robustness to noise. Further analysis revealed different connection patterns underlying the different regimes. Our results predict that apparently similar neural dynamic patterns at the population level (e.g., neural sequences) can exhibit fundamentally different computational properties in regards to their ability to generalize to novel stimuli and their robustness to noise—and that these differences are associated with differences in network connectivity and distinct contributions of excitatory and inhibitory neurons. We also predict that the task structure used in different experimental studies accounts for some of the experimentally observed variability in how networks encode time.     

Paradoxical Interaction between Ocular Activity,Perception, and Decision Confidence at the Threshold of Vision

In humans and some other species perceptual decision-making is complemented by the ability to make confidence judgements about the certainty of sensory evidence. While both forms of decision process have been studied empirically, the precise relationship between them remains poorly understood. We performed an experiment that combined a perceptual decision-making task (identifying the category of a faint visual stimulus) with a confidence-judgement task (wagering on the accuracy of each perceptual decision). The visual stimulation paradigm required steady fixation, so we used eye-tracking to control for stray eye movements. Our data analyses revealed an unexpected and counterintuitive interaction between the steadiness of fixation (prior to and during stimulation), perceptual decision making, and post-decision wagering: greater variability in gaze direction during fixation was associated with significantly increased visual-perceptual sensitivity, but significantly decreased reliability of confidence judgements. The latter effect could not be explained by a simple change in overall confidence (i.e. a criterion artifact), but rather was tied to a change in the degree to which high wagers predicted correct decisions (i.e. the sensitivity of the confidence judgement). We found no evidence of a differential change in pupil diameter that could account for the effect and thus our results are consistent with fixational eye movements being the relevant covariate. However, we note that small changes in pupil diameter can sometimes cause artefactual fluctuations in measured gaze direction and this possibility could not be fully ruled out. In either case, our results suggest that perceptual decisions and confidence judgements can be processed independently and point toward a new avenue of research into the relationship between them.     

Biological and artificial cognition: what can we learn about mechanisms by modelling physical cognition problems using artificial intelligence planning techniques?

Do we fully understand the structure of the problems we present to our subjects in experiments on animal cognition, and the information required to solve them? While we currently have a good understanding of the behavioural and neurobiological mechanisms underlying associative learning processes, we understand much less about the mechanisms underlying more complex forms of cognition in animals. In this study, we present a proposal for a new way of thinking about animal cognition experiments. We describe a process in which a physical cognition task domain can be decomposed into its component parts, and models constructed to represent both the causal events of the domain and the information available to the agent. We then implement a simple set of models, using the planning language MAPL within the MAPSIM simulation environment, and applying it to a puzzle tube task previously presented to orangutans. We discuss the results of the models and compare them with the results from the experiments with orangutans, describing the advantages of this approach, and the ways in which it could be extended.     

Relating Pupil Dilation and Metacognitive Confidence during Auditory Decision-Making

The sources of evidence contributing to metacognitive assessments of confidence in decision-making remain unclear. Previous research has shown that pupil dilation is related to the signaling of uncertainty in a variety of decision tasks. Here we ask whether pupil dilation is also related to metacognitive estimates of confidence. Specifically, we measure the relationship between pupil dilation and confidence during an auditory decision task using a general linear model approach to take into account delays in the pupillary response. We found that pupil dilation responses track the inverse of confidence before but not after a decision is made, even when controlling for stimulus difficulty. In support of an additional post-decisional contribution to the accuracy of confidence judgments, we found that participants with better metacognitive ability – that is, more accurate appraisal of their own decisions – showed a tighter relationship between post-decisional pupil dilation and confidence. Together our findings show that a physiological index of uncertainty, pupil dilation, predicts both confidence and metacognitive accuracy for auditory decisions.     

Unleashing the BEAST: a brief measure of human social information use

Social information enables individuals to reduce uncertainty and increase decision accuracy across a broad range of domains. Intriguingly, individuals and populations consistently differ in social information use. Understanding the underlying causes of this variation has proven challenging due to the lack of a standardized paradigm to quantify social information use. Here we introduce the BEAST (Berlin Estimate AdjuStment Task); a brief (∼5-min), simple, and incentive-compatible behavioural task to quantify individuals' propensities to use social information. In the task, participants observe an image with a number of animals and estimate the total number. Next, they receive another person's estimate, after which they provide a second estimate. An individual's average adjustment quantifies their propensity to use social information. We found that individuals' propensity to use social information is consistent within the task, has considerable test–retest reliability over 9 months, generalizes to other social learning tasks, and correlates with established self-reported measures of social conformity and social proximity. The BEAST thus reliably captures individual variation in social information use. We conclude by highlighting the BEAST's potential to serve as a flexible framework to assess the determinants of human social information use.     

Motoring ahead with rodents

How neural circuits underlie the acquisition and control of learned motor behaviors has traditionally been explored in monkeys and, more recently, songbirds. The development of genetic tools for functional circuit analysis in rodents, the availability of transgenic animals with well characterized phenotypes, and the relative ease with which rats and mice can be trained to perform various motor tasks, make rodents attractive models for exploring the neural circuit mechanisms underlying the acquisition and production of learned motor skills. Here we discuss the advantages and drawbacks of this approach, review recent trends and results, and outline possible strategies for wider adoption of rodents as a model system for complex motor learning.     

Personality influences the neural responses to viewing facial expressions of emotion

Cognitive research has long been aware of the relationship between individual differences in personality and performance on behavioural tasks. However, within the field of cognitive neuroscience, the way in which such differences manifest at a neural level has received relatively little attention. We review recent research addressing the relationship between personality traits and the neural response to viewing facial signals of emotion. In one section, we discuss work demonstrating the relationship between anxiety and the amygdala response to facial signals of threat. A second section considers research showing that individual differences in reward drive (behavioural activation system), a trait linked to aggression, influence the neural responsivity and connectivity between brain regions implicated in aggression when viewing facial signals of anger. Finally, we address recent criticisms of the correlational approach to fMRI analyses and conclude that when used appropriately, analyses examining the relationship between personality and brain activity provide a useful tool for understanding the neural basis of facial expression processing and emotion processing in general.     

A neurocomputational model for the processing of conflicting information in context-dependent decision tasks

Context-dependent computation is a relevant characteristic of neural systems, endowing them with the capacity of adaptively modifying behavioral responses and flexibly discriminating between relevant and irrelevant information in a stimulus. This ability is particularly highlighted in solving conflicting tasks. A long-standing problem in computational neuroscience, flexible routing of information, is also closely linked with the ability to perform context-dependent associations. Here we present an extension of a context-dependent associative memory model to achieve context-dependent decision-making in the presence of conflicting and noisy multi-attribute stimuli. In these models, the input vectors are multiplied by context vectors via the Kronecker tensor product. To outfit the model with a noisy dynamic, we embedded the context-dependent associative memory in a leaky competing accumulator model, and, finally, we proved the power of the model in the reproduction of a behavioral experiment with monkeys in a context-dependent conflicting decision-making task. At the end, we discuss the neural feasibility of the tensor product and made the suggestive observation that the capacities of tensor context models are surprisingly in alignment with the more recent experimental findings about functional flexibility at different levels of brain organization.

     

A Spiking Neural Network Model of Model-Free Reinforcement Learning with High-Dimensional Sensory Input and Perceptual Ambiguity

A theoretical framework of reinforcement learning plays an important role in understanding action selection in animals. Spiking neural networks provide a theoretically grounded means to test computational hypotheses on neurally plausible algorithms of reinforcement learning through numerical simulation. However, most of these models cannot handle observations which are noisy, or occurred in the past, even though these are inevitable and constraining features of learning in real environments. This class of problem is formally known as partially observable reinforcement learning (PORL) problems. It provides a generalization of reinforcement learning to partially observable domains. In addition, observations in the real world tend to be rich and high-dimensional. In this work, we use a spiking neural network model to approximate the free energy of a restricted Boltzmann machine and apply it to the solution of PORL problems with high-dimensional observations. Our spiking network model solves maze tasks with perceptually ambiguous high-dimensional observations without knowledge of the true environment. An extended model with working memory also solves history-dependent tasks. The way spiking neural networks handle PORL problems may provide a glimpse into the underlying laws of neural information processing which can only be discovered through such a top-down approach.     

Emerging rules for the distributions of active dendritic conductances

A key goal in neuroscience is to explain how the operations of a neuron emerge from sets of active channels with specific dendritic distributions. If general principles can be identified for these distributions, dendritic channels should reflect the computational role of a given cell type within its functional neural circuit. Here, we discuss insights from experimental and computational data on the distribution of voltage-gated channels in dendrites, and attempt to derive rules for how their interactions implement different dendritic functions. We propose that this type of analysis will be important for understanding behavioural processes in terms of single-neuron properties, and that it constitutes a step towards a 'functional proteomics' of nerve cells, which will be essential for defining neuronal phenotypes.     

Linking behavioural syndromes and cognition: a behavioural ecology perspective

With the exception of a few model species, individual differences in cognition remain relatively unstudied in non-human animals. One intriguing possibility is that variation in cognition is functionally related to variation in personality. Here, we review some examples and present hypotheses on relationships between personality (or behavioural syndromes) and individual differences in cognitive style. Our hypotheses are based largely on a connection between fast-slow behavioural types (BTs; e.g. boldness, aggressiveness, exploration tendency) and cognitive speed-accuracy trade-offs. We also discuss connections between BTs, cognition and ecologically important aspects of decision-making, including sampling, impulsivity, risk sensitivity and choosiness. Finally, we introduce the notion of cognition syndromes, and apply ideas from theories on adaptive behavioural syndromes to generate predictions on cognition syndromes.     

Incorporating uncertainty in predictive species distribution modelling

Motivated by the need to solve ecological problems (climate change, habitat fragmentation and biological invasions), there has been increasing interest in species distribution models (SDMs). Predictions from these models inform conservation policy, invasive species management and disease-control measures. However, predictions are subject to uncertainty, the degree and source of which is often unrecognized. Here, we review the SDM literature in the context of uncertainty, focusing on three main classes of SDM: niche-based models, demographic models and process-based models. We identify sources of uncertainty for each class and discuss how uncertainty can be minimized or included in the modelling process to give realistic measures of confidence around predictions. Because this has typically not been performed, we conclude that uncertainty in SDMs has often been underestimated and a false precision assigned to predictions of geographical distribution. We identify areas where development of new statistical tools will improve predictions from distribution models, notably the development of hierarchical models that link different types of distribution model and their attendant uncertainties across spatial scales. Finally, we discuss the need to develop more defensible methods for assessing predictive performance, quantifying model goodness-of-fit and for assessing the significance of model covariates.     

Uncertainty-aware and interpretable evaluation of Cas9–gRNA and Cas12a–gRNA specificity for fully matched and partially mismatched targets with Deep Kernel Learning

The choice of guide RNA (gRNA) for CRISPR-based gene targeting is an essential step in gene editing applications, but the prediction of gRNA specificity remains challenging. Lack of transparency and focus on point estimates of efficiency disregarding the information on possible error sources in the model limit the power of existing Deep Learning-based methods. To overcome these problems, we present a new approach, a hybrid of Capsule Networks and Gaussian Processes. Our method predicts the cleavage efficiency of a gRNA with a corresponding confidence interval, which allows the user to incorporate information regarding possible model errors into the experimental design. We provide the first utilization of uncertainty estimation in computational gRNA design, which is a critical step toward accurate decision-making for future CRISPR applications. The proposed solution demonstrates acceptable confidence intervals for most test sets and shows regression quality similar to existing models. We introduce a set of criteria for gRNA selection based on off-target cleavage efficiency and its variance and present a collection of pre-computed gRNAs for human chromosome 22. Using Neural Network Interpretation methods, we show that our model rediscovers an established biological factor underlying cleavage efficiency, the importance of the seed region in gRNA.