Reactivation of prior responses drives serial dependence and stabilizes working memory representations.

Serial dependence has shown seemingly contradictory effects on visual perception and working memory. While serial dependence promotes perpetual and mnemonic stability, it biases behavioral reports towards prior information. The neural mechanisms that drive both biasing and adaptive stabilizing effects are not well understood. We proposed and tested a reactivation and integration mechanism that can account for these contradictory effects. We used multivariate pattern analyses of EEG data (26 human participants, 17 females, 9 males) to examine the reactivation of prior reported orientation during the delay period of a visual working memory task. The reactivation strength of prior reports, but not prior sensory items, was predictive of the magnitude of serial-dependency biases. These reactivated representations integrated with the representation of the current memory item and improved the ability to decode the current contents of memory. Overall, our data provide convergent evidence suggesting that prior reports in a visual working memory task are reactivated on the subsequent trial and become integrated with current memory representations. This similarity-dependent reactivation mechanism drives both report biasing and stabilization effects attributed to serial-dependence in working memory.Significance Statement A primary objective of working memory is to differentiate the present from the past. However, in a continuous visual experience, recent past information often remains relevant for processing subsequent information. How can working memory leverage relevant past information to aid perception and memory while avoiding unwanted influence from irrelevant past information? Here, we show that maintaining new information in working memory can selectively reactivate and assimilate similar information from the recent past. This reactivation biases memory for present items towards the past, which may not always be desirable, but it also enhances the fidelity and precision of these memories, which is certainly an adaptive feature.

Neural Systems Underlying the Implementation of Working Memory Removal Operations.

Recently, multi-voxel pattern analysis has verified that information can be removed from working memory (WM) via three distinct operations replacement, suppression, or clearing compared to information being maintained ( Kim et al., 2020). While univariate analyses and classifier importance maps in Kim et al. (2020) identified brain regions that contribute to these operations, they did not elucidate whether these regions represent the operations similarly or uniquely. Using Leiden-community-detection on a sample of 55 humans (17 male), we identified four brain networks, each of which has a unique configuration of multi-voxel activity patterns by which it represents these WM operations. The visual network (VN) shows similar multi-voxel patterns for maintain and replace, which are highly dissimilar from suppress and clear, suggesting this network differentiates whether an item is held in WM or not. The somatomotor network (SMN) shows a distinct multi-voxel pattern for clear relative to the other operations, indicating the uniqueness of this operation. The default mode network (DMN) has distinct patterns for suppress and clear, but these two operations are more similar to each other than to maintain and replace, a pattern intermediate to that of the VN and SMN. The frontoparietal control network (FPCN) displays distinct multi-voxel patterns for each of the four operations, suggesting that this network likely plays an important role in implementing these WM operations. These results indicate that the operations involved in removing information from WM can be performed in parallel by distinct brain networks, each of which has a particular configuration by which they represent these operations.

Suppressing the Maintenance of Information in Working Memory Alters Long-term Memory Traces.

The sensory recruitment hypothesis conceptualizes information in working memory as being activated representations of information in long-term memory. Accordingly, changes made to an item in working memory would be expected to influence its subsequent retention. Here, we tested the hypothesis that suppressing information from working memory, which can reduce short-term access to that information, may also alter its long-term neural representation. We obtained fMRI data (n = 25; 13 female / 12 male participants) while participants completed a working memory removal task with scene images as stimuli, followed by a final surprise recognition test of the examined items. We applied a multivariate pattern analysis to the data to quantify the engagement of suppression on each trial, to track the contents of working memory during suppression, and to assess representational changes afterward. Our analysis confirms previous reports that suppression of information in working memory involves focused attention to target and remove unwanted information. Furthermore, our findings provide new evidence that even a single dose of suppression of an item in working memory can (if engaged with sufficient strength) produce lasting changes in its neural representation, particularly weakening the unique, item-specific features, which leads to forgetting. Our study sheds light on the underlying mechanisms that contribute to the suppression of unwanted thoughts and highlights the dynamic interplay between working memory and long-term memory.

Author Correction: Closed-loop brain training: the science of neurofeedback.

In this article, the affiliation for Mohit Rana was incorrectly listed as the Institute for Biological and Medical Engineering, Department of Psychiatry, and Section of Neuroscience, Pontificia Universidad Católica de Chile, Vicuña Mackenna 4860 Hernán Briones, piso 2, Macul 782-0436, Santiago, Chile. The listed affiliation should have been the following: Departamento de Psiquiatría, Escuela de Medicina, Centro Interdisciplinario de Neurociencias, Pontificia Universidad Católica de Chile, Santiago, Chile; and the Laboratory for Brain-Machine Interfaces and Neuromodulation, Pontificia Universidad Católica de Chile, Santiago, Chile. An acknowledgement to Mohit Rana's funding source was also missing. The following sentence should have been included in the acknowledgments section: M.R. is supported by a Fondecyt postdoctoral fellowship (project no. 3100648).

Rewarded Extinction Increases Amygdalar Connectivity and Stabilizes Long-Term Memory Traces in the vmPFC.

Neurobiological evidence in rodents indicates that threat extinction incorporates reward neurocircuitry. Consequently, incorporating reward associations with an extinction memory may be an effective strategy to persistently attenuate threat responses. Moreover, while there is considerable research on the short-term effects of extinction strategies in humans, the long-term effects of extinction are rarely considered. In a within-subjects fMRI study with both female and male participants, we compared counterconditioning (CC; a form of rewarded-extinction) to standard extinction at recent (24 h) and remote (approximately one month) retrieval tests. Relative to standard extinction, rewarded extinction diminished 24-h relapse of arousal and threat expectancy, and reduced activity in brain regions associated with the appraisal and expression of threat (e.g., thalamus, insula, periaqueductal gray). The retrieval of reward-associated extinction memory was accompanied by functional connectivity between the amygdala and the ventral striatum, whereas the retrieval of standard-extinction memories was associated with connectivity between the amygdala and ventromedial prefrontal cortex (vmPFC). One month later, the retrieval of both standard-extinction and rewarded-extinction was associated with amygdala-vmPFC connectivity. However, only rewarded extinction created a stable memory trace in the vmPFC, identified through overlapping multivariate patterns of fMRI activity from extinction to 24-h and one-month retrieval. These findings provide new evidence that reward may generate a more stable and enduring memory trace of attenuated threat in humans.SIGNIFICANCE STATEMENT Prevalent treatments for pathologic fear and anxiety are based on the principles of Pavlovian extinction. Unfortunately, extinction forms weak memories that only temporarily inhibit the retrieval of threat associations. Thus, to increase the translational relevance of extinction research, it is critical to investigate whether extinction can be augmented to form a more enduring memory, especially after long intervals. Here, we used a multiday fMRI paradigm in humans to compare the short-term and long-term neurobehavioral effects of aversive-to-appetitive counterconditioning (CC), a form of augmented extinction. Our results provide novel evidence that including an appetitive stimulus during extinction can reduce short-term threat relapse and stabilize the memory trace of extinction in the ventromedial prefrontal cortex (vmPFC), for at least one month after learning.

The effect of biomechanical features on classification of dual-task gait.

Early detection of Alzheimer's Disease and Related Disorders (ADRD) has been a focus of research with the hope that early intervention may improve clinical outcomes. The manifestation of motor impairment in early stages of ADRD has led to the inclusion of gait assessments including spatiotemporal parameters in clinical evaluations. This study aims to determine the effect of adding kinetic and kinematic gait features to classification of different levels of cognitive load in healthy individuals. A dual-task paradigm was used to simulate cognitive impairment in 40 healthy adults, with single-task walking trials representing normal, healthy gait. The Paced Auditory Serial Addition Task was administered at two different inter-stimulus intervals representing two levels of cognitive load in dual-task gait. We predicted that a richer dataset would improve classification accuracy relative to spatiotemporal parameters. Repeated Measures ANOVA showed significant changes in 15 different gait features across all three levels of cognitive load. We used three supervised machine learning algorithms to classify data points using a series of different gait feature sets with performance based on the area under the curve (AUC). Classification yielded 0.778 AUC across all three conditions (0.889 AUC Single vs. Dual) using kinematic and spatiotemporal features compared to 0.724 AUC using spatiotemporal features only (0.792 AUC Single vs. Dual). These data suggest that additional kinematic parameters improve classification performance. However, the benefit of measuring a wider set of parameters compared to their cost needs consideration. Further work will lead to a clinically viable ADRD detection classifier.

Working Memory Swap Errors Have Identifiable Neural Representations.

Working memory is an essential component of cognition that facilitates goal-directed behavior. Famously, it is severely limited and performance suffers when memory load exceeds an individual's capacity. Modeling of visual working memory responses has identified two likely types of errors: guesses and swaps. Swap errors may arise from a misbinding between the features of different items. Alternatively, these errors could arise from memory noise in the feature dimension used for cueing a to-be-tested memory item, resulting in the wrong item being selected. Finally, it is possible that so-called swap errors actually reflect informed guessing, which could occur at the time of a cue, or alternatively, at the time of the response. Here, we combined behavioral response modeling and fMRI pattern analysis to test the hypothesis that swap errors involve the active maintenance of an incorrect memory item. After the encoding of six spatial locations, a retro-cue indicated which location would be tested after memory retention. On accurate trials, we could reconstruct a memory representation of the cued location in both early visual cortex and intraparietal sulcus. On swap error trials identified with mixture modeling, we were able to reconstruct a representation of the swapped location, but not of the cued location, suggesting the maintenance of the incorrect memory item before response. Moreover, participants subjectively responded with some level of confidence, rather than complete guessing, on a majority of swap error trials. Together, these results suggest that swap errors are not mere response-phase guesses, but instead result from failures of selection in working memory, contextual binding errors, or informed guesses, which produce active maintenance of incorrect memory representations.

Closed-loop brain training: the science of neurofeedback.

Neurofeedback is a psychophysiological procedure in which online feedback of neural activation is provided to the participant for the purpose of self-regulation. Learning control over specific neural substrates has been shown to change specific behaviours. As a progenitor of brain-machine interfaces, neurofeedback has provided a novel way to investigate brain function and neuroplasticity. In this Review, we examine the mechanisms underlying neurofeedback, which have started to be uncovered. We also discuss how neurofeedback is being used in novel experimental and clinical paradigms from a multidisciplinary perspective, encompassing neuroscientific, neuroengineering and learning-science viewpoints.

Distinct monitoring strategies underlie costs and performance in prospective memory.

Prospective memory (PM) describes the ability to remember to perform goal-relevant actions at an appropriate time in the future amid concurrent demands. A key contributor to PM performance is thought to be the effortful monitoring of the environment for PM-related cues, a process whose existence is typically inferred from a behavioral interference measure of reaction times. This measure, referred to as "PM costs," is an informative but indirect proxy for monitoring, and it may not be sufficient to understand PM behaviors in all situations. In this study, we asked participants to perform a visual search task with arrows that varied in difficulty while concurrently performing a delayed-recognition PM task with pictures of faces and scenes. To gain a precise measurement of monitoring behavior, we used eye-tracking to record fixations to all task-relevant stimuli and related these fixation measures to both PM costs and PM accuracy. We found that PM costs reflected dissociable monitoring strategies: higher costs were associated with early and frequent monitoring while lower costs were associated with delayed and infrequent monitoring. Moreover, the link between fixations and PM costs varied with cognitive load, and the inclusion of fixation data yielded better predictions of PM accuracy than using PM costs alone. This study demonstrates the benefit of eye-tracking to disentangle the nature of PM costs and more precisely describe strategies involved in prospective remembering.

Emotional learning retroactively enhances item memory but distorts source attribution.

An adaptive memory system should prioritize information surrounding a powerful learning event that may prove useful for predicting future meaningful events. The behavioral tagging hypothesis provides a mechanistic framework to interpret how weak experiences persist as durable memories through temporal association with a strong experience. Memories are composed of multiple elements, and different mnemonic aspects of the same experience may be uniquely affected by mechanisms that retroactively modulate a weakly encoded memory. Here, we investigated how emotional learning affects item and source memory for related events encoded close in time. Participants encoded trial-unique category exemplars before, during, and after Pavlovian fear conditioning. Selective retroactive enhancements in 24-h item memory were accompanied by a bias to misattribute items to the temporal context of fear conditioning. The strength of this source memory bias correlated with participants' retroactive item memory enhancement, and source misattribution to the emotional context predicted whether items were remembered overall. In the framework of behavioral tagging: Memory attribution was biased to the temporal context of the stronger event that provided the putative source of memory stabilization for the weaker event. We additionally found that fear conditioning selectively and retroactively enhanced stimulus typicality ratings for related items, and that stimulus typicality also predicted overall item memory. Collectively, these results provide new evidence that items related to emotional learning are misattributed to the temporal context of the emotional event and judged to be more representative of their semantic category. Both processes may facilitate memory retrieval for related events encoded close in time.

Separation of item and context in item-method directed forgetting.

Contextual information plays a critical role in directed forgetting (DF) of lists of items, whereas DF of individual items has been primarily associated with item-level processing. This study was designed to investigate whether context processing also contributes to the forgetting of individual items. Participants first viewed a series of words, with task-irrelevant scene images (used as "context tags") interspersed between them. Later, these words reappeared without the scenes and were followed by an instruction to remember or forget that word. Multivariate pattern analyses of fMRI data revealed that the reactivation of context information associated with the studied words (i.e., scene-related activity) was greater whereas the item-related information diminished after a forget instruction compared to a remember instruction. Critically, we found the magnitude of the separation between item information and context information predicted successful forgetting. These results suggest that the unbinding of an item from its context may support the intention to forget, and more generally they establish that contextual processing indeed contributes to item-method DF.

Differential neural plasticity of individual fingers revealed by fMRI neurofeedback.

Previous work has shown that functional magnetic resonance imaging (fMRI) activity patterns associated with individual fingers can be shifted by temporary impairment of the hand. Here, we investigated whether these neural activity patterns could be modulated endogenously and whether any behavioral changes result from this modulation. We used decoded neurofeedback in healthy individuals to encourage participants to shift the neural activity pattern in sensorimotor cortex of the middle finger toward the index finger, and the ring finger toward the little finger. We first mapped the neural activity patterns for all fingers of the right hand in an fMRI pattern localizer session. Then, in three subsequent neurofeedback sessions, participants were rewarded after middle/ring finger presses according to their activity pattern overlap during each trial. A force-sensitive keyboard was used to ensure that participants were not altering their physical finger coordination patterns. We found evidence that participants could learn to shift the activity pattern of the ring finger but not of the middle finger. Increased variability of these activity patterns during the localizer session was associated with the ability of participants to modulate them using neurofeedback. Participants also showed an increased preference for the ring finger but not for the middle finger in a postneurofeedback motor task. Our results show that neural activity and behaviors associated with the ring finger are more readily modulated than those associated with the middle finger. These results have broader implications for rehabilitation of individual finger movements, which may be limited or enhanced by individual finger plasticity after neurological injury.NEW & NOTEWORTHY It may be possible to remobilize fingers after neurological injury by altering neural activity patterns. Toward this end, we examined whether finger-related neural activity patterns could be modified in healthy individuals without physical intervention, using fMRI neurofeedback. Our findings show that greater variability of neural patterns at baseline predicted a participant's ability to successfully shift these patterns. Because neural variability is common in individuals poststroke, this illustrates a potential clinical benefit of this procedure.

Consensus on the reporting and experimental design of clinical and cognitive-behavioural neurofeedback studies (CRED-nf checklist).

Neurofeedback has begun to attract the attention and scrutiny of the scientific and medical mainstream. Here, neurofeedback researchers present a consensus-derived checklist that aims to improve the reporting and experimental design standards in the field.

More Is Less: Increased Processing of Unwanted Memories Facilitates Forgetting.

The intention to forget can produce long-lasting effects. This ability has been linked to suppression of both rehearsal and retrieval of unwanted memories, processes mediated by the prefrontal cortex and hippocampus. Here, we describe an alternative account in which the intention to forget is associated with increased engagement with the unwanted information. We used pattern classifiers to decode human functional magnetic resonance imaging data from a task in which male and female participants viewed a series of pictures and were instructed to remember or forget each one. Pictures followed by a forget instruction elicited higher levels of processing in the ventral temporal cortex compared with those followed by a remember instruction. This boost in processing led to more forgetting, particularly for items that showed moderate (vs weak or strong) activation. This result is consistent with the nonmonotonic plasticity hypothesis, which predicts weakening and forgetting of memories that are moderately activated.SIGNIFICANCE STATEMENT The human brain cannot remember everything. Forgetting has a critical role in curating memories and discarding unwanted information. Intentional forgetting has traditionally been linked to passive processes, such as the withdrawal of sustained attention or a stoppage of memory rehearsal. It has also been linked to active suppression of memory processes during encoding and retrieval. Using functional magnetic resonance imaging and machine-learning methods, we show new evidence that intentional forgetting involves an enhancement of memory processing in the sensory cortex to achieve desired forgetting of recent visual experiences. This enhancement temporarily boosts the activation of the memory representation and renders it vulnerable to disruption via homeostatic regulation. Contrary to intuition, deliberate forgetting may involve more rather than less attention to unwanted information.

Predictability Changes What We Remember in Familiar Temporal Contexts.

The human brain constantly anticipates the future based on memories of the past. Encountering a familiar situation reactivates memory of previous encounters, which can trigger a prediction of what comes next to facilitate responsiveness. However, a prediction error can lead to pruning of the offending memory, a process that weakens its representation in the brain and leads to forgetting. Our goal in this study was to evaluate whether memories are spared from such pruning in situations that allow for accurate predictions at the categorical level, despite prediction errors at the item level. Participants viewed a sequence of objects, some of which reappeared multiple times ("cues"), followed always by novel items. Half of the cues were followed by new items from different (unpredictable) categories, while others were followed by new items from a single (predictable) category. Pattern classification of fMRI data was used to identify category-specific predictions after each cue. Pruning was observed only in unpredictable contexts, while encoding of new items was less robust in predictable contexts. These findings demonstrate that how associative memories are updated is influenced by the reliability of abstract-level predictions in familiar contexts.

A simulation-based approach to improve decoded neurofeedback performance.

The neural correlates of specific brain functions such as visual orientation tuning and individual finger movements can be revealed using multivoxel pattern analysis (MVPA) of fMRI data. Neurofeedback based on these distributed patterns of brain activity presents a unique ability for precise neuromodulation. Recent applications of this technique, known as decoded neurofeedback, have manipulated fear conditioning, visual perception, confidence judgements and facial preference. However, there has yet to be an empirical justification of the timing and data processing parameters of these experiments. Suboptimal parameter settings could impact the efficacy of neurofeedback learning and contribute to the 'non-responder' effect. The goal of this study was to investigate how design parameters of decoded neurofeedback experiments affect decoding accuracy and neurofeedback performance. Subjects participated in three fMRI sessions: two 'finger localizer' sessions to identify the fMRI patterns associated with each of the four fingers of the right hand, and one 'finger finding' neurofeedback session to assess neurofeedback performance. Using only the localizer data, we show that real-time decoding can be degraded by poor experiment timing or ROI selection. To set key parameters for the neurofeedback session, we used offline simulations of decoded neurofeedback using data from the localizer sessions to predict neurofeedback performance. We show that these predictions align with real neurofeedback performance at the group level and can also explain individual differences in neurofeedback success. Overall, this work demonstrates the usefulness of offline simulation to improve the success of real-time decoded neurofeedback experiments.

Dissociating refreshing and elaboration and their impacts on memory.

Maintenance of information in working memory (WM) is assumed to rely on refreshing and elaboration, but clear mechanistic descriptions of these cognitive processes are lacking, and it is unclear whether they are simply two labels for the same process. This fMRI study investigated the extent to which refreshing, elaboration, and repeating of items in WM are distinct neural processes with dissociable behavioral outcomes in WM and long-term memory (LTM). Multivariate pattern analyses of fMRI data revealed differentiable neural signatures for these processes, which we also replicated in an independent sample of older adults. In some cases, the degree of neural separation within an individual predicted their memory performance. Elaboration improved LTM, but not WM, and this benefit increased as its neural signature became more distinct from repetition. Refreshing had no impact on LTM, but did improve WM, although the neural discrimination of this process was not predictive of the degree of improvement. These results demonstrate that refreshing and elaboration are separate processes that differently contribute to memory performance.

Self-regulation strategy, feedback timing and hemodynamic properties modulate learning in a simulated fMRI neurofeedback environment.

Direct manipulation of brain activity can be used to investigate causal brain-behavior relationships. Current noninvasive neural stimulation techniques are too coarse to manipulate behaviors that correlate with fine-grained spatial patterns recorded by fMRI. However, these activity patterns can be manipulated by having people learn to self-regulate their own recorded neural activity. This technique, known as fMRI neurofeedback, faces challenges as many participants are unable to self-regulate. The causes of this non-responder effect are not well understood due to the cost and complexity of such investigation in the MRI scanner. Here, we investigated the temporal dynamics of the hemodynamic response measured by fMRI as a potential cause of the non-responder effect. Learning to self-regulate the hemodynamic response involves a difficult temporal credit-assignment problem because this signal is both delayed and blurred over time. Two factors critical to this problem are the prescribed self-regulation strategy (cognitive or automatic) and feedback timing (continuous or intermittent). Here, we sought to evaluate how these factors interact with the temporal dynamics of fMRI without using the MRI scanner. We first examined the role of cognitive strategies by having participants learn to regulate a simulated neurofeedback signal using a unidimensional strategy: pressing one of two buttons to rotate a visual grating that stimulates a model of visual cortex. Under these conditions, continuous feedback led to faster regulation compared to intermittent feedback. Yet, since many neurofeedback studies prescribe implicit self-regulation strategies, we created a computational model of automatic reward-based learning to examine whether this result held true for automatic processing. When feedback was delayed and blurred based on the hemodynamics of fMRI, this model learned more reliably from intermittent feedback compared to continuous feedback. These results suggest that different self-regulation mechanisms prefer different feedback timings, and that these factors can be effectively explored and optimized via simulation prior to deployment in the MRI scanner.

Neural mechanisms of cue-approach training.

Biasing choices may prove a useful way to implement behavior change. Previous work has shown that a simple training task (the cue-approach task), which does not rely on external reinforcement, can robustly influence choice behavior by biasing choice toward items that were targeted during training. In the current study, we replicate previous behavioral findings and explore the neural mechanisms underlying the shift in preferences following cue-approach training. Given recent successes in the development and application of machine learning techniques to task-based fMRI data, which have advanced understanding of the neural substrates of cognition, we sought to leverage the power of these techniques to better understand neural changes during cue-approach training that subsequently led to a shift in choice behavior. Contrary to our expectations, we found that machine learning techniques applied to fMRI data during non-reinforced training were unsuccessful in elucidating the neural mechanism underlying the behavioral effect. However, univariate analyses during training revealed that the relationship between BOLD and choices for Go items increases as training progresses compared to choices of NoGo items primarily in lateral prefrontal cortical areas. This new imaging finding suggests that preferences are shifted via differential engagement of task control networks that interact with value networks during cue-approach training.

Pruning of memories by context-based prediction error.

The capacity of long-term memory is thought to be virtually unlimited. However, our memory bank may need to be pruned regularly to ensure that the information most important for behavior can be stored and accessed efficiently. Using functional magnetic resonance imaging of the human brain, we report the discovery of a context-based mechanism for determining which memories to prune. Specifically, when a previously experienced context is reencountered, the brain automatically generates predictions about which items should appear in that context. If an item fails to appear when strongly expected, its representation in memory is weakened, and it is more likely to be forgotten. We find robust support for this mechanism using multivariate pattern classification and pattern similarity analyses. The results are explained by a model in which context-based predictions activate item representations just enough for them to be weakened during a misprediction. These findings reveal an ongoing and adaptive process for pruning unreliable memories.