Learning attentional templates for value-based decision-making.

Attention filters sensory inputs to enhance task-relevant information. It is guided by an "attentional template" that represents the stimulus features that are currently relevant. To understand how the brain learns and uses templates, we trained monkeys to perform a visual search task that required them to repeatedly learn new attentional templates. Neural recordings found that templates were represented across the prefrontal and parietal cortex in a structured manner, such that perceptually neighboring templates had similar neural representations. When the task changed, a new attentional template was learned by incrementally shifting the template toward rewarded features. Finally, we found that attentional templates transformed stimulus features into a common value representation that allowed the same decision-making mechanisms to deploy attention, regardless of the identity of the template. Altogether, our results provide insight into the neural mechanisms by which the brain learns to control attention and how attention can be flexibly deployed across tasks.

Shared mechanisms underlie the control of working memory and attention.

Cognitive control guides behaviour by controlling what, when, and how information is represented in the brain1. For example, attention controls sensory processing; top-down signals from prefrontal and parietal cortex strengthen the representation of task-relevant stimuli2-4. A similar 'selection' mechanism is thought to control the representations held 'in mind'-in working memory5-10. Here we show that shared neural mechanisms underlie the selection of items from working memory and attention to sensory stimuli. We trained rhesus monkeys to switch between two tasks, either selecting one item from a set of items held in working memory or attending to one stimulus from a set of visual stimuli. Neural recordings showed that similar representations in prefrontal cortex encoded the control of both selection and attention, suggesting that prefrontal cortex acts as a domain-general controller. By contrast, both attention and selection were represented independently in parietal and visual cortex. Both selection and attention facilitated behaviour by enhancing and transforming the representation of the selected memory or attended stimulus. Specifically, during the selection task, memory items were initially represented in independent subspaces of neural activity in prefrontal cortex. Selecting an item caused its representation to transform from its own subspace to a new subspace used to guide behaviour. A similar transformation occurred for attention. Our results suggest that prefrontal cortex controls cognition by dynamically transforming representations to control what and when cognitive computations are engaged.

Neural signature of flexible coding in prefrontal cortex.

The ability of prefrontal cortex to quickly encode novel associations is crucial for adaptive behavior and central to working memory. Fast Hebbian changes in synaptic strength permit forming new associations, but neuronal signatures of this have been elusive. We devised a trialwise index of pattern similarity to look for rapid changes in population codes. Based on a computational model of working memory, we hypothesized that synaptic strength-and consequently, the tuning of neurons-could change if features of a subsequent stimulus need to be "reassociated," i.e., if bindings between features need to be broken to encode the new item. As a result, identical stimuli might elicit different neural responses. As predicted, neural response similarity dropped following rebinding, but only in prefrontal cortex. The history-dependent changes were expressed on top of traditional, fixed selectivity and were not explainable by carryover of previous firing into the current trial or by neural adaptation.

Delay-period activity in frontal, parietal, and occipital cortex tracks noise and biases in visual working memory.

Working memory is imprecise, and these imprecisions can be explained by the combined influences of random diffusive error and systematic drift toward a set of stable states ("attractors"). However, the neural correlates of diffusion and drift remain unknown. Here, we investigated how delay-period activity in frontal and parietal cortex, which is known to correlate with the decline in behavioral memory precision observed with increasing memory load, might relate to diffusion and drift. We analyzed data from an existing experiment in which subjects performed delayed recall for line orientation, at different loads, during functional magnetic resonance imaging (fMRI) scanning. To quantify the influence of drift and diffusion, we modeled subjects' behavior using a discrete attractor model and calculated within-subject correlation between frontal and parietal delay-period activity and whole-trial estimates of drift and diffusion. We found that although increases in frontal and parietal activity were associated with increases in both diffusion and drift, diffusion explained the most variance in frontal and parietal delay-period activity. In comparison, a subsequent whole-brain regression analysis showed that drift, rather than diffusion, explained the most variance in delay-period activity in lateral occipital cortex. These results are consistent with a model of the differential recruitment of general frontoparietal mechanisms in response to diffusive noise and of stimulus-specific biases in occipital cortex.

Drifting codes within a stable coding scheme for working memory.

Working memory (WM) is important to maintain information over short time periods to provide some stability in a constantly changing environment. However, brain activity is inherently dynamic, raising a challenge for maintaining stable mental states. To investigate the relationship between WM stability and neural dynamics, we used electroencephalography to measure the neural response to impulse stimuli during a WM delay. Multivariate pattern analysis revealed representations were both stable and dynamic: there was a clear difference in neural states between time-specific impulse responses, reflecting dynamic changes, yet the coding scheme for memorised orientations was stable. This suggests that a stable subcomponent in WM enables stable maintenance within a dynamic system. A stable coding scheme simplifies readout for WM-guided behaviour, whereas the low-dimensional dynamic component could provide additional temporal information. Despite having a stable subspace, WM is clearly not perfect-memory performance still degrades over time. Indeed, we find that even within the stable coding scheme, memories drift during maintenance. When averaged across trials, such drift contributes to the width of the error distribution.

Working Memory Is Complex and Dynamic, Like Your Thoughts.

Working memory is where thoughts are held and manipulated. For many years, the dominant model was that working memory relied on steady-state neural dynamics. A neural representation was activated and then held in that state. However, as often happens, the more we examine working memory (especially with new technology), the more complex it looks. Recent discoveries show that working memory involves multiple mechanisms, including discontinuous bouts of spiking. Memories are also dynamic, evolving in a task-dependent manner. Cortical rhythms may control those dynamics, thereby endowing top-down "executive" control over our thoughts.

Working Memory Load Modulates Neuronal Coupling.

There is a severe limitation in the number of items that can be held in working memory. However, the neurophysiological limits remain unknown. We asked whether the capacity limit might be explained by differences in neuronal coupling. We developed a theoretical model based on Predictive Coding and used it to analyze Cross Spectral Density data from the prefrontal cortex (PFC), frontal eye fields (FEF), and lateral intraparietal area (LIP). Monkeys performed a change detection task. The number of objects that had to be remembered (memory load) was varied (1-3 objects in the same visual hemifield). Changes in memory load changed the connectivity in the PFC-FEF-LIP network. Feedback (top-down) coupling broke down when the number of objects exceeded cognitive capacity. Thus, impaired behavioral performance coincided with a break-down of Prediction signals. This provides new insights into the neuronal underpinnings of cognitive capacity and how coupling in a distributed working memory network is affected by memory load.

Stimulus Load and Oscillatory Activity in Higher Cortex.

Exploring and exploiting a rich visual environment requires perceiving, attending, and remembering multiple objects simultaneously. Recent studies have suggested that this mental "juggling" of multiple objects may depend on oscillatory neural dynamics. We recorded local field potentials from the lateral intraparietal area, frontal eye fields, and lateral prefrontal cortex while monkeys maintained variable numbers of visual stimuli in working memory. Behavior suggested independent processing of stimuli in each hemifield. During stimulus presentation, higher-frequency power (50-100 Hz) increased with the number of stimuli (load) in the contralateral hemifield, whereas lower-frequency power (8-50 Hz) decreased with the total number of stimuli in both hemifields. During the memory delay, lower-frequency power increased with contralateral load. Load effects on higher frequencies during stimulus encoding and lower frequencies during the memory delay were stronger when neural activity also signaled the location of the stimuli. Like power, higher-frequency synchrony increased with load, but beta synchrony (16-30 Hz) showed the opposite effect, increasing when power decreased (stimulus presentation) and decreasing when power increased (memory delay). Our results suggest roles for lower-frequency oscillations in top-down processing and higher-frequency oscillations in bottom-up processing.

PFC neurons reflect categorical decisions about ambiguous stimuli.

We examined whether PFC neuron activity reflects categorical decisions in monkeys categorizing ambiguous stimuli. A morphing system was used to systematically vary stimulus shape and precisely define category boundaries. Ambiguous stimuli were centered on a category boundary, that is, they were a mix of 50% of two prototypes and therefore had no category information, so monkeys guessed at their category membership. We found that the monkeys' trial-by-trial decision about the category membership of an ambiguous image was reflected in PFC activity. Activity to the same ambiguous image differed significantly, depending on which category the monkey had assigned it to. This effect only occurred when that scheme was behaviorally relevant. These indicate that PFC activity reflects categorical decisions.

Neural substrates of cognitive capacity limitations.

Cognition has a severely limited capacity: Adult humans can retain only about four items "in mind". This limitation is fundamental to human brain function: Individual capacity is highly correlated with intelligence measures and capacity is reduced in neuropsychiatric diseases. Although human capacity limitations are well studied, their mechanisms have not been investigated at the single-neuron level. Simultaneous recordings from monkey parietal and frontal cortex revealed that visual capacity limitations occurred immediately upon stimulus encoding and in a bottom-up manner. Capacity limitations were found to reflect a dual model of working memory. The left and right halves of visual space had independent capacities and thus are discrete resources. However, within each hemifield, neural information about successfully remembered objects was reduced by adding further objects, indicating that resources are shared. Together, these results suggest visual capacity limitation is due to discrete, slot-like, resources, each containing limited pools of neural information that can be divided among objects.

Laminar differences in gamma and alpha coherence in the ventral stream.

Attention to a stimulus enhances both neuronal responses and gamma frequency synchrony in visual area V4, both of which should increase the impact of attended information on downstream neurons. To determine whether gamma synchrony is common throughout the ventral stream, we recorded from neurons in the superficial and deep layers of V1, V2, and V4 in two rhesus monkeys. We found an unexpected striking difference in gamma synchrony in the superficial vs. deep layers. In all three areas, spike-field coherence in the gamma (40-60 Hz) frequency range was largely confined to the superficial layers, whereas the deep layers showed maximal coherence at low frequencies (6-16 Hz), which included the alpha range. In the superficial layers of V2 and V4, gamma synchrony was enhanced by attention, whereas in the deep layers, alpha synchrony was reduced by attention. Unlike these major differences in synchrony, attentional effects on firing rates and noise correlation did not differ substantially between the superficial and deep layers. The results suggest that synchrony plays very different roles in feedback and feedforward projections.

Differences in the expression of cortex-wide neural dynamics are related to behavioral phenotype.

Behavior differs across individuals, ranging from typical to atypical phenotypes.1 Understanding how differences in behavior relate to differences in neural activity is critical for developing treatments of neuropsychiatric and neurodevelopmental disorders. One hypothesis is that differences in behavior reflect individual differences in the dynamics of how information flows through the brain. In support of this, the correlation of neural activity between brain areas, termed "functional connectivity," varies across individuals2 and is disrupted in autism,3 schizophrenia,4 and depression.5 However, the changes in neural activity that underlie altered behavior and functional connectivity remain unclear. Here, we show that individual differences in the expression of different patterns of cortical neural dynamics explain variability in both functional connectivity and behavior. Using mesoscale imaging, we recorded neural activity across the dorsal cortex of behaviorally "typical" and "atypical" mice. All mice shared the same recurring cortex-wide spatiotemporal motifs of neural activity, and these motifs explained the large majority of variance in cortical activity (>75%). However, individuals differed in how frequently different motifs were expressed. These differences in motif expression explained differences in functional connectivity and behavior across both typical and atypical mice. Our results suggest that differences in behavior and functional connectivity are due to changes in the processes that select which pattern of neural activity is expressed at each moment in time.

Building compositional tasks with shared neural subspaces.

Cognition is remarkably flexible; we are able to rapidly learn and perform many different tasks1. Theoretical modeling has shown artificial neural networks trained to perform multiple tasks will re-use representations2 and computational components3 across tasks. By composing tasks from these sub-components, an agent can flexibly switch between tasks and rapidly learn new tasks4. Yet, whether such compositionality is found in the brain is unknown. Here, we show the same subspaces of neural activity represent task-relevant information across multiple tasks, with each task compositionally combining these subspaces in a task-specific manner. We trained monkeys to switch between three compositionally related tasks. Neural recordings found task-relevant information about stimulus features and motor actions were represented in subspaces of neural activity that were shared across tasks. When monkeys performed a task, neural representations in the relevant shared sensory subspace were transformed to the relevant shared motor subspace. Subspaces were flexibly engaged as monkeys discovered the task in effect; their internal belief about the current task predicted the strength of representations in task-relevant subspaces. In sum, our findings suggest that the brain can flexibly perform multiple tasks by compositionally combining task-relevant neural representations across tasks.

Corvids optimize working memory by categorizing continuous stimuli.

Working memory (WM) is a crucial element of the higher cognition of primates and corvid songbirds. Despite its importance, WM has a severely limited capacity and is vulnerable to noise. In primates, attractor dynamics mitigate the effect of noise by discretizing continuous information. Yet, it remains unclear whether similar dynamics are seen in avian brains. Here, we show jackdaws (Corvus monedula) have similar behavioral biases as humans; memories are less precise and more biased as memory demands increase. Model-based analysis reveal discrete attractors are evenly spread across the stimulus space. Altogether, our comparative approach suggests attractor dynamics in primates and corvids mitigate the effect of noise by systematically drifting towards specific attractors. By demonstrating this effect in an evolutionary distant species, our results strengthen attractor dynamics as general, adaptive biological principle to efficiently use WM.

The neural basis of swap errors in working memory.

When making decisions in a cluttered world, humans and other animals often have to hold multiple items in memory at once - such as the different items on a shopping list. Psychophysical experiments in humans and other animals have shown remembered stimuli can sometimes become confused, with participants reporting chimeric stimuli composed of features from different stimuli. In particular, subjects will often make "swap errors" where they misattribute a feature from one object as belonging to another object. While swap errors have been described behaviorally, their neural mechanisms are unknown. Here, we elucidate these neural mechanisms through trial-by-trial analysis of neural population recordings from posterior and frontal brain regions while monkeys perform two multi-stimulus working memory tasks. In these tasks, monkeys were cued to report the color of an item that either was previously shown at a corresponding location (requiring selection from working memory) or will be shown at the corresponding location (requiring attention to a position). Animals made swap errors in both tasks. In the neural data, we find evidence that the neural correlates of swap errors emerged when correctly remembered information is selected incorrectly from working memory. This led to a representation of the distractor color as if it were the target color, underlying the eventual swap error. We did not find consistent evidence that swap errors arose from misinterpretation of the cue or errors during encoding or storage in working memory. These results suggest an alternative to established views on the neural origins of swap errors, and highlight selection from and manipulation in working memory as crucial - yet surprisingly brittle - neural processes.

Multiplexed Subspaces Route Neural Activity Across Brain-wide Networks.

Cognition is flexible. Behaviors can change on a moment-by-moment basis. Such flexibility is thought to rely on the brain's ability to route information through different networks of brain regions in order to support different cognitive computations. However, the mechanisms that determine which network of brain regions is engaged are unknown. To address this, we combined cortex-wide calcium imaging with high-density electrophysiological recordings in eight cortical and subcortical regions of mice. Different dimensions within the population activity of each brain region were functionally connected with different cortex-wide 'subspace networks' of regions. These subspace networks were multiplexed, allowing a brain region to simultaneously interact with multiple independent, yet overlapping, networks. Alignment of neural activity within a region to a specific subspace network dimension predicted how neural activity propagated between regions. Thus, changing the geometry of the neural representation within a brain region could be a mechanism to selectively engage different brain-wide networks to support cognitive flexibility.

Working memory control dynamics follow principles of spatial computing.

Working memory (WM) allows us to remember and selectively control a limited set of items. Neural evidence suggests it is achieved by interactions between bursts of beta and gamma oscillations. However, it is not clear how oscillations, reflecting coherent activity of millions of neurons, can selectively control individual WM items. Here we propose the novel concept of spatial computing where beta and gamma interactions cause item-specific activity to flow spatially across the network during a task. This way, control-related information such as item order is stored in the spatial activity independent of the detailed recurrent connectivity supporting the item-specific activity itself. The spatial flow is in turn reflected in low-dimensional activity shared by many neurons. We verify these predictions by analyzing local field potentials and neuronal spiking. We hypothesize that spatial computing can facilitate generalization and zero-shot learning by utilizing spatial component as an additional information encoding dimension.

A Goldilocks theory of cognitive control: Balancing precision and efficiency with low-dimensional control states.

Cognitive control orchestrates interactions between brain regions, guiding the transformation of information to support contextually appropriate and goal-directed behaviors. In this review, we propose a hierarchical model of cognitive control where low-dimensional control states direct the flow of high-dimensional representations between regions. This allows cognitive control to flexibly adapt to new environments and maintain the representational capacity to capture the richness of the world.

In V1, attending is not learning to see.

Learning and attention improve perception by increasing information about a stimulus in the neural population. In this issue of Neuron, Poort et al. investigate the circuit mechanisms underlying attention and learning, finding they work through different mechanisms.

Fast rule switching and slow rule updating in a perceptual categorization task.

To adapt to a changing world, we must be able to switch between rules already learned and, at other times, learn rules anew. Often we must do both at the same time, switching between known rules while also constantly re-estimating them. Here, we show these two processes, rule switching and rule learning, rely on distinct but intertwined computations, namely fast inference and slower incremental learning. To this end, we studied how monkeys switched between three rules. Each rule was compositional, requiring the animal to discriminate one of two features of a stimulus and then respond with an associated eye movement along one of two different response axes. By modeling behavior, we found the animals learned the axis of response using fast inference (rule switching) while continuously re-estimating the stimulus-response associations within an axis (rule learning). Our results shed light on the computational interactions between rule switching and rule learning, and make testable neural predictions for these interactions.