LECTURES
Ole Jensen, University of Birmingham
Title: Gating by alpha band inhibition revised: a case for an indirect control mechanism and involvement of the basal ganglia
In recent years, there has been a paradigm shift in the understanding of the functional role of alpha oscillations (8-13 Hz). These oscillations were previously thought to reflect a state of rest but are now known to be causally involved in resource allocation in the working brain by selectively inhibiting task-irrelevant regions. However, it is still debated whether alpha oscillations are under direct top-down control to suppress visual distractions. We have proposed a revised mechanism where alpha oscillations are indirectly regulated by the perceptual load of goal-relevant information. Our mechanism aligns with perceptual load theory and is supported by recent studies using MEG where we measured the modulation of alpha oscillations when manipulating the perceptual load of targets or distractors. Our research also suggests that the basal ganglia are involved in modulating alpha oscillations in spatial attention tasks. By combining MRI and MEG data, we found that volumetric asymmetries of the caudate nucleus and Globus pallidus predict the ability to modulate alpha oscillations by spatial attention. In the future when intend to apply the revised framework when instigating resource allocations in real-life tasks that require spatial attention, such as reading and visual exploration.
Jacquie Gottlieb, Columbia University, New York
Title: Allocating attention for information gain: the roles of knowledge, information value and cognitive costs
Attention has long been both an indispensable and a highly controversial construct in cognitive science. Much of the controversy surrounding attention, I argued, stems from a lack of theoretical frameworks describing “top-down” attention control. How do individuals select sensory stimuli to serve behavioral goals? I will discuss a novel theory addressing this question, which proposes that attention maximizes expected information gains (EIG) – i.e., targets information channels (e.g., peripheral locations or features) that are expected to reduce an individual’s uncertainty about future actions or states. I will present evidence that EIG-based control is implemented through interactions between fronto-parietal and executive networks. Specifically, visual and fronto-parietal networks implement competitive based on diagnosticity -- the accuracy of the predictions made by stimuli appearing in a channel. Executive networks, in turn, provide dynamic estimates of uncertainty that determine the immediate value of additional information and control the strength of competitive interactions. Thus, when uncertainty is low and future states are predictable in advance, attention can be more broadly distributed including to some task-irrelevant stimuli. However, when uncertainty and EIG become higher, attention is sharply focused on the most diagnostic predictors, which become more valuable for achieving one’s goal. I will discuss evidence from humans and non-human primates supporting this view, and its broader implications of the theory for topics such as curiosity and the role of previous knowledge in EIG estimation.
Birte Forstmann, University of Amsterdam
Title: Multi-study fMRI outlooks on subcortical BOLD responses in the stop-signal paradigm
Functional magnetic resonance imaging (fMRI) approaches have frequently used the stop-signal task (SST) to examine cortico-basal ganglia networks. Here, we merged five such datasets, using a novel aggregatory method allowing the unification of raw fMRI data across different scanner sites. In this talk, I will present a meta-analysis, along with other recent aggregatory fMRI studies, showing that there is no evidence for the innervation of the hyperdirect or indirect cortico-basal-ganglia pathways in successful response inhibition. Instead large subcortical activity profiles for failed stop trials were found. I will discuss possible explanations for the mismatch of findings between the fMRI results presented here and results from other research modalities that have implicated nodes of the basal ganglia in successful inhibition.
Michael Frank, Brown University
Title: Adaptive Gating to Optimise Working Memory Chunking in Frontostriatal Neural Networks
How and why is working memory (WM) capacity limited? Traditional accounts focus either on limitations on the number or items that can be stored (slots models), or degradations in their precision with increasing load (resource models). I will argue that WM limitations arise as a result of difficulties in management of WM information content, requiring control processes to guide the updating and read-out of information to and from distinct addresses within memory. In neural networks, these processes can be mediated by striatal input and output gating of distinct clusters of prefrontal populations. Such strategies cannot be explicitly supervised (i.e,, there is no homunculus), but they can be adapted via reinforcement learning, leading to a challenge to credit assignment in WM tasks with increasing load. I will show that instead, networks can adapt their gating policies to "chunk" information, reusing the same prefrontal populations to store multiple items, leading to resource-like constraints within a slot-like system, and inducing a tradeoff between quantity and precision of information. Networks adapt chunking gating policies as a function of task demands, mimicking human performance and normative models.
Blitz Abstracts
Blitz 1
Dr Simon Little, Ucsf
Title: Assessing goal-directed decision making in Parkinson’s disease using dopamine and intracranial recordings from human fronto-basal ganglia circuits
Adaptive behavior requires balancing easy, fast habitual choices with slow, effortful goal-directed control over decision making. The basal ganglia (BG) is hypothesized to integrate habitual input from premotor cortices (PMC) with goal-directed control signals from prefrontal regions under the influence of dopamine. Dysfunction in these circuits leads to disabling motivation deficits in neuropsychiatric disorders, including apathy and impulsivity in Parkinson’s disease (PD). Recent work in computational psychiatry has linked severity of these symptoms to decreases in goal-directed decision making, and dopaminergic medications have been found to boost goal-directed decision making in healthy controls and people with PD. Thus, identifying neural signals underlying these decision-making strategies could lead to potential biomarkers to tailor precision therapies for motivation deficits. Here, we use novel sensing-enabled deep brain stimulation devices to record intracranial data from human PMC and BG while people with PD perform a modified two-step reward learning paradigm that separates goal-directed and habitual decisions (n=9 patients, both ON/OFF dopamine). Contrary to prior reports, reinforcement learning modeling of choice behavior revealed that dopamine increased habitual decision making at the group level (p=0.08), but with substantial individual differences. Single-trial theta power in PMC and BG was predicted by reward prediction errors (RPEs), and model comparisons revealed that PMC theta was best explained by habitual RPEs, while BG theta reflected a hybrid of habitual and goal-directed RPE signals. Collectively, these findings demonstrate the power of modern neurotechnology for studying mechanisms of human reward circuitry that can inform future treatments of motivation deficits in neuropsychiatric conditions.
Dr. Jan Wessel, University Of Iowa
Title: The human subthalamic nucleus transiently inhibits active attentional processes
The subthalamic nucleus (STN) of the basal ganglia is key to the inhibitory control of movement. Accordingly, it is a primary target for the neurosurgical treatment of movement disorders like Parkinson’s Disease, where modulating the STN via deep-brain stimulation (DBS) can release excess inhibition of thalamo-cortical motor circuits. However, the STN is also anatomically connected to other thalamo-cortical circuits, including those underlying cognitive processes like attention5,6. This suggests that the STN may also contribute to the inhibition of those processes. We here tested this hypothesis in humans. We used a novel, wireless outpatient method to record intracranial local field potentials from STN DBS implants during a visual attention task9. We also modulated STN via DBS. In both cases, we simultaneously recorded high-density EEG to extract the steady-state visual evoked potential (SSVEP), a neural measure of visual attentional engagement. Introducing unexpected, distracting sounds lead to a momentary reduction of this SSVEP. This suppression was preceded by sound-related γ-frequency (>60Hz) activity in STN, which scaled with each sound’s surprisal value. The STN activity statistically mediated the suppressive effect of surprisal on the SSVEP. Finally, modulating STN activity via DBS reduced the sound-related SSVEP-suppression. Together, these findings provide the first causal evidence that the human STN contributes to the inhibition of attention, a non-motor process. Beyond their support for a domain-general inhibitory role of the STN, these findings also suggest a mechanism underlying known cognitive side-effects of STN DBS.
Dr Jacob Russin, Brown University
Title: Human Curriculum Effects Emerge with In-Context Learning in Neural Networks
Human learning is sensitive to rule-like structure and the curriculum of examples used for training. In tasks governed by succinct rules, learning is more robust when related examples are blocked across trials, but in the absence of such rules, interleaving is more effective. To date, no neural model has simultaneously captured these seemingly contradictory effects. Here we show that this same tradeoff spontaneously emerges with "in-context learning" (ICL) both in neural networks trained with metalearning and in large language models (LLMs). ICL is the ability to learn new tasks "in context" --- without weight changes --- via an inner-loop algorithm implemented in activation dynamics. Experiments with pretrained LLMs and metalearning transformers show that ICL exhibits the blocking advantage demonstrated in humans on a task involving rule-like structure, and conversely, that concurrent in-weight learning reproduces the interleaving advantage observed in humans on tasks lacking such structure.
Blitz 2
Dr Romy Froemer, University of Birmingham
Title: Neural underpinnings of the evaluation of control to determine how much mental effort is worth investing
Dr. Ross Otto, McGill University
Title: Goal Gradient Effects in Effortful Control Exertion
The Goal-Gradient hypothesis posits that organisms increase effort expenditure as a function of their proximity to a goal. Despite nearly a century having passed since its original formulation, goal gradient-like behaviour in human cognitive performance remains poorly understood: are we more willing to engage in costly cognitive processing when we are near, versus far from a goal state? Moreover, the computational mechanisms underpinning these potential goal gradient effects—for example, whether goal proximity affects fidelity of stimulus encoding, response caution, or other identifiable mechanisms governing speed and accuracy—are unclear. Here, in two experiments (N=82), we examine the effect of goal proximity, operationalized as progress towards completion of a rewarded task block, upon task performance in an attentionally demanding oddball task. Supporting the goal gradient hypothesis, we found that participants responded more quickly, but not less accurately, when rewards were proximal than when they were distal. Critically, this effect was only observed when participants were given information about goal proximity. Using hierarchical Drift Diffusion Models (DDMs), we found that these apparent goal gradient performance effects were best explained by a collapsing-bounds DDM, in which proximity to a goal simultaneously 1) reduced response caution and 2) increased efficiency of information processing. Taken together, these results suggest that goal gradients could help explain the frequently observed fluctuations in engagement of cognitively effortful processing, extending the scope of the goal-gradient hypothesis to the domain of cognitive tasks.
Dr Andrea Pisauro, University Of Plymouth
Title: Neural and Computational mechanisms of effort based decisions under the pressure of a deadline
Exerting effort is fundamental for survival. Yet, animals typically avoid exerting effort unless associated with large immediate rewards, a behaviour linked to effort cost processing in the anterior cingulate cortex (ACC) and the putamen. However, there are situations when exerting effort is valuable, as it leads to progress towards medium and long-term goals. Here, using a novel task and computational modelling, we showed that effort can be valued, and preferred, when there is pressure to reach a goal before a deadline. Using ultra-high-field fMRI, we identified a spectrum of computations within functionally connected putamen and ACC sub-regions that signalled and updated estimates of deadline pressure. Separate putamen and ACC sub-regions processed the costs or added value of effort, with variation in these signals between people associated with differences in deadline pressure sensitivity. This work provides a novel account of the neurocomputational mechanisms underlying effort-based decisions under pressure, and how deadlines fundamentally shift the value and the motivation to exert effort.
The ability to guide decisions based on context is a key building block of intelligent behaviour. Previous research has examined how context shapes instantaneous decisions, where stimuli evoke independent action-outcome mappings, revealing that top-down control signals originating in prefrontal cortex can change the encoding of sensory information. Here, we study the control mechanisms supporting sequential context-based decisions, where multiple steps are taken to achieve a given goal. We asked participants to navigate an avatar through a grid world to find rewards in two of four potential goal locations, whilst undergoing fMRI. Participants learned to perform the task in two distinct contexts, each of which defined the location of the second reward contingent on the first. By leveraging computational modelling and multivariate analyses of BOLD signals, we considered how the representation of space might vary with goal states. In the hippocampus and orbitofrontal cortex, we observed spatial maps that were “compressed,” with goals sharing prospective pathways clustering together in neural state space to emphasise the dimension predictive of reward. This compression was only observed once subjects could infer goal locations, it tracked individual differences in task performance, and it markedly differed from spatial signals observed elsewhere in the brain. A computational model in which place fields represent both current and prospective locations can account for these results, providing a mechanistic theory of how the brain implements sequential context-dependent decisions.
Blitz 3
Dr. Andrew Westbrook, Rutgers University
Title: What Criticality Tells Us About Brain Maturation, Excitation-Inhibition Balance, and Cognitive Effort
The human brain is a complex dynamical system with emergent properties implying that it operates near a critical point – at the boundary between dramatically different dynamical regimes. A key control parameter determining proximity to criticality is the balance between excitatory and inhibitory neurotransmission (or “E-I balance”). While hyper-inhibited systems are non-responsive, hyper-excited systems have runaway dynamics. In contrast, operating near criticality, with excitation and inhibition in balance, affords desirable functionality, from the standpoint of cognition. Critical systems have maximal susceptibility and dynamic range. They also maximize entropy and exhibit long-range spatial and temporal correlations. Such properties are essential for demanding cognitive control and working-memory operations that require both flexibility for updating with new information and stability for maintenance over retention intervals. These properties also appear to diminish when people engage in demanding tasks suggesting a normative explanation for why we treat thinking as effortful. Brain development provides a tests of these hypotheses as prior work shows both more tightly regulated E-I balance and improved working memory performance with maturation through adulthood. Examining electroencephalography data through the lens of critical dynamics, we find evidence of increasing E-I balance and closer proximity to criticality in adulthood. We further find that working memory ability depends on proximity to criticality and, intriguingly, that people whose brains veer farther from criticality experience more subjective cognitive effort during demanding working memory tasks.
Prof. Maria Ruz, University of Granada
Title: Neural flexible coding of control variables through novel verbal instructions
Humans excel at following instructed commands, often guided by sentences that convey details about the relevance of pieces of information and the rules associating them with the required actions. This ability, tightly linked to cognitive control mechanisms, is most useful in novel changing scenarios. In our lab we use variations of a paradigm where complex novel verbal instructions, composed by orthogonal control-related variables, are combined with neuroimaging (fMRI, EEG) and multivariate analyses to study how different task dimensions shape the format in which the human brain represents the instructed information and the underlying dynamics. Results show that in such context of hierarchical combinatorial reuse and ample task space, the control operations to be applied (e.g. either to select or to integrate information) generate separable geometries that dominate neural coding during preparatory and task implementation epochs, sustained by compositional (vs. conjunctive) representational formats that generalize across time and stimulus items. More concrete, nested variables (e.g. relevant features such as color and shape, stimulus categories or motor responses) generate activations traceable during shorter periods of time, with lower levels of generalization across time and task contexts. Overall, our investigations shed light on how knowledge is selectively represented and flexibly reconfigured to meet instructed demands
Senne Braem, Ghent University
Title: Learning where to be flexible takes time: Evidence from task switching, decision making, and neural networks
Humans are remarkably efficient at adapting to different contexts by exerting varying levels of cognitive control. However, while it is often assumed that such context-sensitive flexibility can be learned, the dynamics and representations underlying this form of learning are mostly unclear. We developed a task switching paradigm and a decision making paradigm, where participants were each time exposed to two environments with different task-switching probabilities. Across both paradigms, our results show that people can learn different control strategies in the low- versus high-switch environment, but only after several days of training. To further understand how this slow learning contrasts with earlier observations in the literature suggesting faster learning, we developed a recurrent neural network model to simulate and study context-sensitive flexibility. Our model learned to successfully adjust its flexibility in response to different control needs. Importantly, however, we demonstrate that this context-specific flexibility can come about in two ways: through fast activity-based adaptations or slow weight-based learning. We further show that earlier demonstrations of item-, task-, or environment-sensitive flexibility, likely depend on different dynamics and representations, and may not always reflect or require (weight-based) learning. Together, these findings provide novel insights in how humans can learn to manage the flexibility-stability trade-off.
Prof Roshan Cools, Donders Institute For Brain, Cognition And Behaviour
Title: Controllability-based control of motivational bias during learning
The ability to exert cognitive control is computationally expensive. Given our limited resources, we need to recruit such an expensive control strategy only when it is needed, for example, when the current instrumental context requires actions (e.g., nogo-to-win) that are in direct conflict with our hardwired Pavlovian biases (e.g., go-to-win). How do we decide when to recruit cognitive control? We posit a crucial role for our ability to estimate the controllability of the environment. To test the hypothesis we conducted a behavioral go/nogo-to-win or -avoid learning study in which outcome controllability is manipulated by varying the degree to which outcomes depend on the participant’s action choice. Preliminary results show that cognitive control over Pavlovian biases varies as a function of estimated controllability, so that Pavlovian go-to-win and nogo-to-avoid biases are greater in low- than high-controllability blocks, in a way that cannot be explained by differences in outcome predictability. We also plan to present data from a fast fMRI study that builds on the recent discovery that wave-like dopamine release patterns in the rodent dorsal striatum depend on outcome controllability (Hamid et al., 2021), thus suggesting a key role for striatal dopamine in deciding whether to recruit cognitive control. We aim to uncover analogous waves of striatal activity during the performance of our paradigm, depending on fluctuations of outcome controllability. We hypothesize that higher controllability estimates are associated with greater dorsal-to-ventral striatal waves that in turn predict greater cognitive control of maladaptive behavioural Pavlovian biases.
Blitz 4
Dr. Yaara Erez , Bar-Ilan University
Title: Electrophysiological characteristics of frontal control regions – an ECOG study
Robust and consistent fMRI evidence show that cognitive control processes are supported by a distributed frontoaparietal network. Yet, the fine-grained electrophysiological characteristics of control regions at the meso-scale are not yet well understood. Here we used electrocorticography (ECOG) in patients with brain tumours undergoing awake surgery. During surgery, patients performed a task with increased demand, similarly to tasks used in fMRI studies showing the recruitment of control regions. Task-related high gamma activity converged with the frontoparietal network and distinguished frontal control regions from adjacent cortical networks. In contrast, decreases in activity in the beta range were widespread and non-specific. We further asked whether functional connectivity was modulated by task demand. Phase-amplitude coupling of beta and high gamma activity was larger within control regions during task performance compared to coupling with electrodes outside these regions. Recent evidence demonstrated that glioma tumours integrate within their surroundings, affecting local neural processing. Combining ECOG data with pre-surgery resting-state fMRI from the same patients, we found indications for participation of tumour-infiltrated cortex in whole-brain large-scale circuits. Our results reveal patterns of activity and connectivity in control regions at multiple scales, with implications for both basic and translational neuroscience.
Dr. Theresa Desrochers, Brown University
Title: Neural dynamics in the frontal cortex during sequential control dissociate participants with obsessive-compulsive disorder from healthy controls
Completing sequences, like cooking a meal, are part of everyday life. Often sequences are dictated by rules or tasks rather than the identity of individual steps (e.g., making spaghetti regardless of pasta brand). The neural basis of such sequential control may differ in disorders with disrupted sequential behavior, like obsessive-compulsive disorder (OCD). Neurobiological models theorize the role of several prefrontal cortical (PFC) regions in OCD pathology. These regions may interact with or encompass regions necessary for sequential task control, such as the rostrolateral PFC (RLPFC). However, OCD models have primarily been tested in non-sequential paradigms, leaving open questions as to whether they can account for sequential behaviors and their neural underpinnings. We used fMRI during task sequences in humans to test the involvement of the RLPFC and other brain regions in OCD compared to healthy controls (HCs). We found no difference in RLPFC activity in OCD. However, we found that a dynamic previously found to be necessary for task sequences, increasing (ramping) activity throughout each sequence, was increased in OCD in other prefrontal areas such as the anterior cingulate, ventromedial and dorsolateral PFC. Strikingly, this ramping cooccurred with decreases non-ramping activity in the same regions, and contrasted with results from non-sequential studies showing hypoactivation in other areas. Together these results suggest that models of OCD function and dysfunction need to account for more complex tasks that evolve through time, and suggest novel dynamics and functions associated with cognitive control and the value of internal, sequential processes that could be incorporated.
Dr. Manuela Ruzzoli, Basque Center on Cognition, Brain and Language
Title: Finding enjoyment through cognitive conflict
Overcoming cognitive conflict requires increased cognitive control and is considered inherently aversive. It is commonly assumed that people try to minimize cognitive effort and, consequently, the amount of control they need to exert. In this study, we examine this assumption by allowing participants to choose their preferred level of cognitive conflict. After a familiarization phase, participants (N = 100) completed 10 blocks (30 trials each) of the Stroop task, choosing the proportion of incongruent trials they faced before each block. Overall, participants showed a preference for moderate to high levels of cognitive conflict over low levels. Furthermore, they performed better and faster in blocks characterized by greater cognitive conflict. After each block, participants' ratings revealed that higher levels of cognitive conflict were experienced as effortful, yet enjoyable. Our findings thus challenge the assumption that cognitive conflict is inherently aversive.
Dr Apoorva Bhandaris, Brown University
Title: Task tailored representations in the human lateral prefrontal cortex
How does the human brain flexibly adapt to tasks of varying structure? The lateral prefrontal cortex (lPFC) flexibly represents task information. However, principles that shape lPFC representational geometry remain unsettled. We use deep sampling, functional MRI and pattern analyses to reveal the fine structure of lPFC representational geometries as human participants perform two distinct categorization tasks– one with flat, conjunctive categories and another with hierarchical, context-dependent categories. We show that lPFC preferentially encodes diverse, task-relevant information with task-tailored manifolds of intermediate dimensionality. These geometries preferentially enhance the separability of task-relevant variables while encoding a few in abstract form. In the flat task, a global axis encodes response-relevant categories in abstract form, while category-specific local geometries are high-dimensional. In the hierarchy task, a global axis abstractly encodes the higher-level context and low-dimensional, context-specific local geometries compress irrelevant stimulus information, encoding the relevant information in abstract form. We identify similarities and differences in the representational geometries of the two tasks to highlight the key principles that shape lPFC coding.
Prof. Jennifer Cook, University of Birmingham
Title: Flexible mechanisms for social and individual learning
The existence in the human brain of neural and/or neurochemical pathways that are specialised for learning from social information is the topic of much debate. Indeed, some theories of human cultural evolution posit that humans have social-specific learning mechanisms that are adaptive specialisations moulded by natural selection. Cognitive neuroscientific studies present mixed evidence: some studies find dissociable mechanisms for social learning and learning from individual experience (individual learning), whereas others find the same brain areas and, dopamine-mediated, computations involved in both. In this talk I will argue that neurochemical mechanisms underpinning learning can be dissociated along a primary-secondary but not a social-individual axis. That is, social learning relies upon the dopamine-rich mechanisms that also underpin individual learning when social information is the primary learning source, but not when it comprises a secondary, additional element. This proposal resolves conflicting literature because in studies which find common mechanisms participants are generally encouraged to learn primarily from social information, whereas in studies which find dissociable mechanisms social information generally comprises a secondary source. Our proposal supports a burgeoning field showing that, rather than being fixedly specialised for particular inputs, neurochemical pathways in the human brain can process both social and non-social cues and flexibly switch between the two depending upon which cue is primarily relevant for the task at hand.
SYMPOSIA
Prof. Heleen Slagter, Vrije Universiteit Amsterdam
Title: Working memory in action
In recent years, it has become clear that working memory does not simply serve temporary storage of previously seen information, but is fundamentally directed towards maintenance of information in service of future action. Yet, if and how sensory representations in working memory are modulated by planned actions is still unclear. We examined this in four studies. In the first three studies, we found that when participants planned actions on an object held in mind, this strengthened their sensory representation, as reflected in enhanced attentional capture on an intermittent visual search task (shown with eye tracking) and a larger Ppc, an ERP component associated with attentional saliency (shown with EEG). The fourth and last study manipulated whether two oriented bars held in working memory were associated with the same action plan or two different action plans. We found that action similarity affected the extent to which the two bars, when similarly oriented, were reported as more visually distinct. A control experiment confirmed the sensory nature of this action-induced repulsion effect. Together, these findings indicate that when we plan an action on an object in mind, attention is automatically drawn to those (imagined) features that are relevant to the planned action, suggesting a more central role for action in working memory than typically assumed.
Prof. Tom Verguts, Ghent University
Title: Cognitive control and neural synchronization
Cognitive control requires binding packages of information (e.g., stimulus-response bindings) together on the fly, often for just a brief period. However, it remains unclear how humans can actually do this. I propose that neural oscillations at theta frequency (4-8 Hz) are a key factor in this respect. I present a computational model of how theta can synchronize specific neural modules. Thus, the model can functionally construct the required packages of information depending on the task at hand; and quickly dismantle them when the package is no longer needed (e.g., in the next trial). I will describe this model as well as behavioral and electrophysiological (measured via EEG) tests of the model. In a second part of the talk, I consider how universal this synchronization strategy is. In particular, we find that for some tasks, it is more efficient to use simpler strategies, and that subjects then also seem to use those simpler strategies more often. I conclude that synchronization can be useful for cognitive control; but that it is only used when really needed.
Alexandra Woolgar, University of Cambridge
Title: Parietal alpha stimulation causally enhances attentional information coding in evoked and oscillatory activity
Selective attention is a fundamental cognitive mechanism that allows people to prioritise task-relevant information while ignoring irrelevant information. Previous research has suggested key roles of parietal evoked potentials and alpha oscillatory responses in spatial attention tasks. However, the informational content of these signals is less clear, and their causal effects on the coding of multiple task elements are yet unresolved. Here, we used concurrent TMS-EEG to causally manipulate pre-stimulus parietal alpha power and investigate the subsequent oscillatory and evoked coding of multiple task features (where to attend, what to attend to, and visual stimulus) in a selective attention task. We found that all three types of task-relevant information could be decoded from evoked potentials, while both where to attend and what to attend to cold be decoded from alpha power. Compared with control arrhythmic-TMS, pre-stimulus rhythmic-TMS (rTMS) over the right intraparietal sulcus (IPS) causally and selectively improved multivariate decoding of the information about where to attend, from both alpha power and evoked potentials, during task performance. Moreover, these decoding improvements predicted individual-subject improvement in behavioural performance. These findings suggest an intimate relationship between pre-stimulus parietal alpha power, evoked and oscillatory responses during task processing, and behaviour, and imply a specific and causal role of IPS-alpha controlled evoked and oscillatory activity in carrying behaviour-driving information about where to focus attention.
Professor Clare Press, University College London
Title: Different types of surprise: How to direct perception to optimise learning
The world never unfolds precisely as we predict, such that our perceptual streams may be considered continuous error signals. However, only some of these error signals constitute signals that the underlying structure of the world has changed. We can learn to expect variability along particular dimensions. We must also incorporate noise in perceptual signals, driven both by the ambiguous outside world and imperfect internal systems. Scientists have long grappled with understanding how we determine whether an error signal constitutes signal or noise, and therefore when to update models. My talk will consider recent work from our lab that asks about the role of perception in these processes. I will present psychophysical, modelling and neuroimaging (EEG and 7T MRI) work that considers the mechanisms allowing us to perceive particular types of error and how they may serve learning.
Stephen Fleming
Title: Explaining how underconfidence is maintained in the face of reality
Computational neuroscience typically considers how agents develop beliefs about the task at hand. However, humans also maintain rich models of ourselves - our skills, abilities and behavioural characteristics. These “hidden” metacognitive beliefs can exert outsized impact on behaviour. They can also diverge markedly from reality - as when debilitating underconfidence prevents someone from pursuing new endeavours. Measuring confidence in minimal psychophysical tasks provides a useful experimental window onto these metacognitive aspects of decision-making. Recently we sought to characterise interactions between "global" beliefs about task performance, and local confidence in individual decisions. We find that global underconfidence in individuals with anxiety and depression can be explained by distortions in learning from local confidence, together with a tendency for (under)confidence to generalize across task domains. In contrast, learning from explicit feedback remained unaffected. Together our findings help explain the formation and maintenance of confidence biases, and reveal the hidden metacognitive influences that shape human behaviour.
Dr. Sam McDougle, Yale University
Title: Structured Visuomotor Representations and Cognitive-motor Information Flow
Humans often structure mental representations, such as multi-step plans of action or mental maps of the environment. How structured representations interface with the motor system during action selection is unclear. Here, we used psychophysics and computational modeling to examine the dynamics of hierarchical action selection. Subjects learned a complex visuomotor mapping with latent structure. In spite of successive training trials being independent, response times between trials revealed strong evidence of a tree-like mental graph of the learned mapping. Modeling and control experiments helped rule out alternative explanations, such as lower-level visual and physiological factors. To get a finer-grained understanding of action selection dynamics we implemented a forced response time paradigm. We discovered that selection of a single finger movement can flow from a latent cognitive sequence whereby clusters of actions are serially potentiated over time in accordance with a learned structure. Our results thus reveal direct coupling between high-level cognitive representations and motor preparation.
Prof. Julie Duque, Louvain University
Title: Studying urgency and arousal during decision making in humans
Humans and other animals make a wide range of decisions throughout their daily lives, with varying degrees of speed and precision. This variability is not only due to the amount of evidence based on which one makes decisions but also to the sense of urgency and the level of arousal that can vary between and within individuals, from one situation to another. My talk will focus on the work we have been doing lately to better understand mechanisms underlying urgency and the contribution of arousal during decision making. It will be divided into two parts. In the first section, I will describe a study in which we investigated the impact of urgency on motor neural activity, studied using transcranial magnetic stimulation (TMS) over primary motor cortex (M1) during decision making in an index finger variant of the Tokens task, originally developed for studies of urgency in non-human primates by D. Thura and P. Cisek. Then, in the second part of my talk, I will turn to a recent investigation of the role of the arousal system in decision making. Interestingly, it is possible to causally address the role of arousal in humans by means of transcutaneous Vagus Nerve Stimulation (tVNS), which employs electrical stimulation targeting the auricular branch of the vagus nerve to stimulate non-invasively the locus coeruleus noradrenergic system, one major source of arousal in the brain. I will present behavioural data collected in the random dot motion task with online trains of tVNS (compared to sham) and pupillometry.
Dr. Katja Kornysheva, University of Birmingham
Title: Dissecting the neural control of dextrous actions
When we learn motor skills, it is widely believed that there is a shift from associative to motor reference frames paralleled by the increased involvement of cortical and subcortical areas with direct projections to brainstem and spinal control centres. This shift is thought to underlie the formation of a motor “repertoire” enabling procedural holistic control of fine-grained kinematic trajectories. However, I will argue that even after prolonged training and motor skill production entirely from memory without external guidance, complex movements like typing or handwriting are not controlled by retrieving holistic trajectories. Instead, they are assembled from smaller components each time they are retrieved from memory and involve areas traditionally associated with declarative memory, e.g. the hippocampus, shortly before being executed. I propose that this hierarchical control mechanism supports ad-hoc behavioural flexibility, a key characteristic of skilled motor control in humans.
Dr. Steve Chang, Yale University
Title:
Studying the neuroscience of cooperation in marmoset dyads
Social interaction is essential to our daily lives, shaping interpersonal communication and the decisions we make. One of the major hallmarks of advanced social cognition is the flexible ability to work together for mutual benefits while competing against one another for limited resources. This ability is grounded in diverse and adaptive strategies. However, studying the neurobiology of cooperation has been challenging. This is largely because the standard animal models of neuroscience do not reliably exhibit cooperative strategies. Therefore, there is a need to investigate the neurobiology of complex social cognition in a species whose social structure strongly depends on both cooperation, which is often accompanied by high social tolerance. This presentation will describe a novel automated cooperation paradigm in freely moving common marmosets that combines markerless behavioral tracking, a dynamic Bayesian network modeling of behavioral dependencies, and wireless neuronal recording. Cooperation in this paradigm was guided by the strategic use of social gaze and was critically dependent on social relationships. Overall, this presentation introduces a robust naturalistic cooperation paradigm suitable for neural investigation.
Dr. Patricia Lockwood, University of Birmingham
Title: Prosocial Learning and Motivation across the Lifespan
Many of our decisions affect other people. Our choices can decelerate climate change, stop the spread of infectious diseases, and directly help or harm others. Prosocial behaviours – decisions that help others – could contribute to reducing the impact of these challenges, yet their computational and neural mechanisms remain poorly understood.
I will present recent work that examines prosocial motivation, how willing we are to incur costs to help others, prosocial learning, how we learn from the outcomes of our choices when they affect other people, and prosocial preferences, our self-reports of willingness to help others. I will show that there are important differences in these processes as we grow older. Next, I will present studies probing the neural basis of prosocial motivation and prosocial learning using computational modelling, functional neuroimaging and voxel-based lesion-symptom mapping. This work shows that different aspects of prosocial motivation and learning can be distinguished by signals in brain areas that are domain general and domain specific for social processing. Finally, I will discuss recent work that has examined prosocial preferences and age on a global scale, and has examined how we can change social preferences in older adults. These studies highlight that differences in prosocial behaviour in older adults are similar around the world and that older adults may be more susceptible to impulsive social influence.
Dr. Lusha Zhu, Peking University
Title: A computational shortcut to coordination: common knowledge and neural alignment
Coordinating actions for mutual benefits is ubiquitous among social animals. Game theory and multi-agent AI research proposes that coordination in the absence of communication often involves complex, iterative reasoning about the mental states of other agents. Such decision-making processes are widely considered to be prohibitively difficult and error-prone, raising questions regarding how social animals achieve effective, flexible everyday coordination in a seemingly effortless manner. Here, building on a long-standing conjecture that shared social understanding can facilitate coordination, we show that the interpersonal alignment of fMRI activity in the posterior cingulate cortex (PCC) in one group of subjects can reliably and specifically predict coordination in novel, one-shot contexts in a separate, large online sample. These results suggest that coordination may be supported by shared world understanding similarly coded in the PCC across individuals, which can be flexibly assembled and compared in service of social behavior.
Dr. Vijay Mohan K Namboodiri, UCSF
Title: Retrospective learning in the brain
A hallmark of intelligence is the ability to learn associations between causes and effects (e.g., environmental cues and associated rewards). The near consensus understanding of the last few decades is that animals learn cause-effect associations from errors in the prediction of the effect (e.g., a reward prediction error or RPE). This theory has been hugely influential in neuroscience as decades of evidence suggested that mesolimbic dopamine (DA)— known to be critical for associative learning—appears to signal RPE. Though some evidence questioned whether DA signals RPE, the RPE hypothesis remained the best explanation of learning because no other normative theory of learning explained experimental observations inconsistent with RPE while also capturing phenomena explained by RPE.However, my lab recently proposed a new theory of associative learning (named ANCCR, read “anchor”) which postulates that animals learn associations by retrospectively identifying causes of meaningful effects such as rewards and that mesolimbic dopamine conveys that a current event is meaningful (Jeong et al., Science, 2022, Burke et al., BioRxiv, 2023, Garr et al., BioRxiv, 2023). The core idea is simple: you can learn to predict the future by retrodicting the past, and you retrodict the past only after meaningful events. Here, I will present the basic formulation of this theory, experimental data focused on distinguishing predictions of ANCCR and RPE, and unpublished experimental results demonstrating that behavioral and dopaminergic learning rates from cue-reward experiences are quantitatively scaled by reward sparsity, providing an effective solution to “few-shot learning” from sparse experiences.
Professor Matthew Rushworth, University of Oxford
Title: Anterior lateral prefrontal cortex and prospective metacognition
Before we decide what to do next, we assess how difficult it will be for us to do it. Recent behavioural analyses suggest that humans and macaques are able to assess how difficult it is likely to be to tackle a particular decision even before the decision is actually made. Using fMRI it is possible to show that arriving at an estimate of whether or not one can tackle a particular decision depends on an accumulating signal in the anterior lateral prefrontal cortex area 47 (alPFC47). The signal is proportional to the probability that the decision will be performed correctly. The faster the neural signal increases, the more likely a person is to take the decision. Different patterns of neural activity are associated with evaluating the probability that you yourself can perform a decision correctly and probabilities of external contingencies that do not depend on you. This process of metacognitive estimation of our own decision-making abilities can also be used to estimate how well some one else might perform a decision task. However, because these estimates are derived from projections of our own performance abilities, they are most helpful when we are estimating how well someone with similar or worse skills will perform as opposed to someone with better skills. A similar, albeit limited, ability is present in macaques and is associated with activity in a similar area. Disrupting the alPFC47 activity does not affect decision making per se but it does compromise the prospective metacognitive evaluation of decision making.
Dr. Carolina Rezaval, University of Birmingham
Title: Is love blind? Mating proximity gates threat perception
When forced to choose between fundamental needs, making the wrong decision could prove fatal. However, it is currently unclear how alternative options are evaluated and appropriate actions are prioritised. To tackle this problem, we developed an experimental system to study the neural circuit mechanisms that integrate the benefit of imminent courtship success with the risk of predation in Drosophila. By combining our novel behavioural assay with neurogenetics, connectomics and live imaging, we identified the neural circuitry that establishes behavioural priority during this ‘life-death’ conflict. Crucially, we found that the probability of mating success defines the decision to reproduce or flee. Our work reveals how the brain weighs up antagonistic advantages and risks, and the probability of success, at a cellular-circuit level.