Distinct hippocampal mechanisms support concept formation and updating.

Learning systems must constantly decide whether to create new representations or update existing ones. For example, a child learning that a bat is a mammal and not a bird would be best served by creating a new representation, whereas updating may be best when encountering a second similar bat. Characterizing the neural dynamics that underlie these complementary memory operations requires identifying the exact moments when each operation occurs. We address this challenge by interrogating fMRI brain activation with a computational learning model that predicts trial-by-trial when memories are created versus updated. We found distinct neural engagement in anterior hippocampus and ventral striatum for model-predicted memory create and update events during early learning. Notably, the degree of this effect in hippocampus, but not ventral striatum, significantly related to learning outcome. Hippocampus additionally showed distinct patterns of functional coactivation with ventromedial prefrontal cortex and angular gyrus during memory creation and premotor cortex during memory updating. These findings suggest that complementary memory functions, as formalized in computational learning models, underlie the rapid formation of novel conceptual knowledge, with the hippocampus and its interactions with frontoparietal circuits playing a crucial role in successful learning. Significance statement: How do we reconcile new experiences with existing knowledge? Prominent theories suggest that novel information is either captured by creating new memories or leveraged to update existing memories, yet empirical support of how these distinct memory operations unfold during learning is limited. Here, we combine computational modeling of human learning behaviour with functional neuroimaging to identify moments of memory formation and updating and characterize their neural signatures. We find that both hippocampus and ventral striatum are distinctly engaged when memories are created versus updated; however, it is only hippocampus activation that is associated with learning outcomes. Our findings motivate a key theoretical revision that positions hippocampus is a key player in building organized memories from the earliest moments of learning.

Modeling Similarity and Psychological Space.

Similarity and categorization are fundamental processes in human cognition that help complex organisms make sense of the cacophony of information in their environment. These processes are critical for tasks such as recognizing objects, making decisions, and forming memories. In this review, we provide an overview of the current state of knowledge on similarity and psychological spaces, discussing the theories, methods, and empirical findings that have been generated over the years. Although the concept of similarity has important limitations, it plays a key role in cognitive modeling. The review surfaces three key themes. First, similarity and mental representations are merely two sides of the same coin, existing as a similarity-representation duality that defines a psychological space. Second, both the brain's mental representations and the study of mental representations are made possible by exploiting second-order isomorphism. Third, similarity analysis has near-universal applicability across all levels of cognition, providing a common research language.

You can't play 20 questions with nature and win redux.

An incomplete science begets imperfect models. Nevertheless, the target article advocates for jettisoning deep-learning models with some competency in object recognition for toy models evaluated against a checklist of laboratory findings; an approach which evokes Alan Newell's 20 questions critique. We believe their approach risks incoherency and neglects the most basic test; can the model perform its intended task.

A too-good-to-be-true prior to reduce shortcut reliance.

Despite their impressive performance in object recognition and other tasks under standard testing conditions, deep networks often fail to generalize to out-of-distribution (o.o.d.) samples. One cause for this shortcoming is that modern architectures tend to rely on ǣshortcutsǥ superficial features that correlate with categories without capturing deeper invariants that hold across contexts. Real-world concepts often possess a complex structure that can vary superficially across contexts, which can make the most intuitive and promising solutions in one context not generalize to others. One potential way to improve o.o.d. generalization is to assume simple solutions are unlikely to be valid across contexts and avoid them, which we refer to as the too-good-to-be-true prior. A low-capacity network (LCN) with a shallow architecture should only be able to learn surface relationships, including shortcuts. We find that LCNs can serve as shortcut detectors. Furthermore, an LCN's predictions can be used in a two-stage approach to encourage a high-capacity network (HCN) to rely on deeper invariant features that should generalize broadly. In particular, items that the LCN can master are downweighted when training the HCN. Using a modified version of the CIFAR-10 dataset in which we introduced shortcuts, we found that the two-stage LCN-HCN approach reduced reliance on shortcuts and facilitated o.o.d. generalization.

Signatures of cross-modal alignment in children's early concepts.

Whether supervised or unsupervised, human and machine learning is usually characterized as event-based. However, learning may also proceed by systems alignment in which mappings are inferred between entire systems, such as visual and linguistic systems. Systems alignment is possible because items that share similar visual contexts, such as a car and a truck, will also tend to share similar linguistic contexts. Because of the mirrored similarity relationships across systems, the visual and linguistic systems can be aligned at some later time absent either input. In a series of simulation studies, we considered whether children's early concepts support systems alignment. We found that children's early concepts are close to optimal for inferring novel concepts through systems alignment, enabling agents to correctly infer more than 85% of visual-word mappings absent supervision. One possible explanation for why children's early concepts support systems alignment is that they are distinguished structurally by their dense semantic neighborhoods. Artificial agents using these structural features to select concepts proved highly effective, both in environments mirroring children's conceptual world and those that exclude the concepts that children commonly acquire. For children, systems alignment and event-based learning likely complement one another. Likewise, artificial systems can benefit from incorporating these developmental principles.

A multilevel account of hippocampal function in spatial and concept learning: Bridging models of behavior and neural assemblies.

A complete neuroscience requires multilevel theories that address phenomena ranging from higher-level cognitive behaviors to activities within a cell. We propose an extension to the level of mechanism approach where a computational model of cognition sits in between behavior and brain: It explains the higher-level behavior and can be decomposed into lower-level component mechanisms to provide a richer understanding of the system than any level alone. Toward this end, we decomposed a cognitive model into neuron-like units using a neural flocking approach that parallels recurrent hippocampal activity. Neural flocking coordinates units that collectively form higher-level mental constructs. The decomposed model suggested how brain-scale neural populations coordinate to form assemblies encoding concept and spatial representations and why so many neurons are needed for robust performance at the cognitive level. This multilevel explanation provides a way to understand how cognition and symbol-like representations are supported by coordinated neural populations (assemblies) formed through learning.

Noninvasive Continuous Glucose Monitoring With a Novel Wearable Dial Resonating Sensor: A Clinical Proof-of-Concept Study.

BACKGROUND: A noninvasive, wearable continuous glucose monitor would be a major advancement in diabetes therapy. This trial investigated a novel noninvasive glucose monitor which analyzes spectral variations in radio frequency/microwave signals reflected from the wrist. METHODS: A single-arm, open-label, experimental study compared glucose values from a prototype investigational device with laboratory glucose measurements from venous blood samples (Super GL Glucose Analyzer, Dr. Müller Gerätebau GmbH) at varying levels of glycemia. The study included 29 male participants with type 1 diabetes (age range = 19-56 years). The study comprised three stages with the following aims: (1) demonstrate initial proof-of-principle, (2) test an improved device design, and (3) test performance on two consecutive days without device recalibration. The co-primary endpoints in all trial stages were median and mean absolute relative difference (ARD) calculated across all data points. RESULTS: In stage 1, the median and mean ARDs were 30% and 46%, respectively. Stage 2 produced marked performance improvements with a median and mean ARD of 22% and 28%, respectively. Stage 3 showed that, without recalibration, the device performed as well as the initial prototype (stage 1) with a median and mean ARD of 35% and 44%, respectively. CONCLUSION: This proof-of-concept study shows that a novel noninvasive continuous glucose monitor was capable of detecting glucose levels. Furthermore, the ARD results are comparable to first models of commercially available minimally invasive products without the need to insert a needle. The prototype has been further developed and is being tested in subsequent studies. TRIAL REGISTRATION NUMBER: NCT05023798.

Dorsomedial striatum, but not dorsolateral striatum, is necessary for rat category learning.

Categorization is an adaptive cognitive function that allows us to generalize knowledge to novel situations. Converging evidence from neuropsychological, neuroimaging, and neurophysiological studies suggest that categorization is mediated by the basal ganglia; however, there is debate regarding the necessity of each subregion of the basal ganglia and their respective functions. The current experiment examined the roles of the dorsomedial striatum (DMS; homologous to the head of the caudate nucleus) and dorsolateral striatum (DLS; homologous to the body and tail of the caudate nucleus) in category learning by combining selective lesions with computational modeling. Using a touchscreen apparatus, rats were trained to categorize distributions of visual stimuli that varied along two continuous dimensions (i.e., spatial frequency and orientation). The tasks either required attention to one stimulus dimension (spatial frequency or orientation; 1D tasks) or both stimulus dimensions (spatial frequency and orientation; 2D tasks). Rats with NMDA lesions of the DMS were impaired on both the 1D tasks and 2D tasks, whereas rats with DLS lesions showed no impairments. The lesions did not affect performance on a discrimination task that had the same trial structure as the categorization tasks, suggesting that the category impairments effected processes relevant to categorization. Model simulations were conducted using a neural network to assess the effect of the DMS lesions on category learning. Together, the results suggest that the DMS is critical to map category representations to appropriate behavioral responses, whereas the DLS is not necessary for categorization.

Reassessing hierarchical correspondences between brain and deep networks through direct interface.

Functional correspondences between deep convolutional neural networks (DCNNs) and the mammalian visual system support a hierarchical account in which successive stages of processing contain ever higher-level information. However, these correspondences between brain and model activity involve shared, not task-relevant, variance. We propose a stricter account of correspondence: If a DCNN layer corresponds to a brain region, then replacing model activity with brain activity should successfully drive the DCNN's object recognition decision. Using this approach on three datasets, we found that all regions along the ventral visual stream best corresponded with later model layers, indicating that all stages of processing contained higher-level information about object category. Time course analyses suggest that long-range recurrent connections transmit object class information from late to early visual areas.

System alignment supports cross-domain learning and zero-shot generalisation.

Recent findings suggest conceptual relationships hold across modalities. For instance, if two concepts occur in similar linguistic contexts, they also likely occur in similar visual contexts. These similarity structures may provide a valuable signal for alignment when learning to map between domains, such as when learning the names of objects. To assess this possibility, we conducted a paired-associate learning experiment in which participants mapped objects that varied on two visual features to locations that varied along two spatial dimensions. We manipulated whether the featural and spatial systems were aligned or misaligned. Although system alignment was not required to complete this supervised learning task, we found that participants learned more efficiently when systems aligned and that aligned systems facilitated zero-shot generalisation. We fit a variety of models to individuals' responses and found that models which included an offline unsupervised alignment mechanism best accounted for human performance. Our results provide empirical evidence that people align entire representation systems to accelerate learning, even when learning seemingly arbitrary associations between two domains.

Sequential consumer choice as multi-cued retrieval.

Whether adding songs to a playlist or groceries during an online shop, how do we decide what to choose next? We develop a model that predicts such open-ended, sequential choices using a process of cued retrieval from long-term memory. Using the past choice to cue subsequent retrievals, this model predicts the sequential purchases and response times of nearly 5 million grocery purchases made by more than 100,000 online shoppers. Products can be associated in different ways, such as by their episodic association or semantic overlap, and we find that consumers query multiple forms of associative knowledge when retrieving options. Attending to certain knowledge sources, as estimated by our model, predicts important retrieval errors, such as the propensity to forget or add unwanted products. Our results demonstrate how basic memory retrieval mechanisms shape choices in real-world, goal-directed tasks.

A neural network account of memory replay and knowledge consolidation.

Replay can consolidate memories through offline neural reactivation related to past experiences. Category knowledge is learned across multiple experiences, and its subsequent generalization is promoted by consolidation and replay during rest and sleep. However, aspects of replay are difficult to determine from neuroimaging studies. We provided insights into category knowledge replay by simulating these processes in a neural network which approximated the roles of the human ventral visual stream and hippocampus. Generative replay, akin to imagining new category instances, facilitated generalization to new experiences. Consolidation-related replay may therefore help to prepare us for the future as much as remember the past. Generative replay was more effective in later network layers functionally similar to the lateral occipital cortex than layers corresponding to early visual cortex, drawing a distinction between neural replay and its relevance to consolidation. Category replay was most beneficial for newly acquired knowledge, suggesting replay helps us adapt to changes in our environment. Finally, we present a novel mechanism for the observation that the brain selectively consolidates weaker information, namely a reinforcement learning process in which categories were replayed according to their contribution to network performance. This reinforces the idea of consolidation-related replay as an active rather than passive process.

Confound-leakage: confound removal in machine learning leads to leakage.

BACKGROUND: Machine learning (ML) approaches are a crucial component of modern data analysis in many fields, including epidemiology and medicine. Nonlinear ML methods often achieve accurate predictions, for instance, in personalized medicine, as they are capable of modeling complex relationships between features and the target. Problematically, ML models and their predictions can be biased by confounding information present in the features. To remove this spurious signal, researchers often employ featurewise linear confound regression (CR). While this is considered a standard approach for dealing with confounding, possible pitfalls of using CR in ML pipelines are not fully understood. RESULTS: We provide new evidence that, contrary to general expectations, linear confound regression can increase the risk of confounding when combined with nonlinear ML approaches. Using a simple framework that uses the target as a confound, we show that information leaked via CR can increase null or moderate effects to near-perfect prediction. By shuffling the features, we provide evidence that this increase is indeed due to confound-leakage and not due to revealing of information. We then demonstrate the danger of confound-leakage in a real-world clinical application where the accuracy of predicting attention-deficit/hyperactivity disorder is overestimated using speech-derived features when using depression as a confound. CONCLUSIONS: Mishandling or even amplifying confounding effects when building ML models due to confound-leakage, as shown, can lead to untrustworthy, biased, and unfair predictions. Our expose of the confound-leakage pitfall and provided guidelines for dealing with it can help create more robust and trustworthy ML models.

When unsupervised training benefits category learning.

Humans continuously categorise inputs, but only rarely receive explicit feedback as to whether or not they are correct. This implies that they may be integrating unsupervised information together with their sparse supervised data - a form of semi-supervised learning. However, experiments testing semi-supervised learning are rare, and are bedevilled with conflicting results about whether the unsupervised information affords any benefit. Here, we suggest that one important factor that has been paid insufficient attention is the alignment between subjects' internal representations of the stimulus material and the experimenter-defined representations that determine success in the tasks. Subjects' representations are shaped by prior biases and experience, and unsupervised learning can only be successful if the alignment suffices. Otherwise, unsupervised learning might harmfully strengthen incorrect assumptions. To test this hypothesis, we conducted an experiment in which subjects initially categorise items along a salient, but task-irrelevant, dimension, and only recover the correct categories when sufficient feedback draws their attention to the subtle, task-relevant, stimulus dimensions. By withdrawing feedback at different stages along this learning curve, we tested whether unsupervised learning improves or worsens performance when internal stimulus representations and task are sufficiently or insufficiently aligned, respectively. Our results demonstrate that unsupervised learning can indeed have opposing effects on subjects' learning. We also discuss factors limiting the degree to which such effects can be predicted from momentary performance. Our work implies that predicting and understanding human category learning in particular tasks requires assessment and consideration of the representational spaces that subjects entertain for the materials involved in those tasks. These considerations not only apply to studies in the lab, but could also help improve the design of tutoring systems and instruction.

Robust priors for regularized regression.

Induction benefits from useful priors. Penalized regression approaches, like ridge regression, shrink weights toward zero but zero association is usually not a sensible prior. Inspired by simple and robust decision heuristics humans use, we constructed non-zero priors for penalized regression models that provide robust and interpretable solutions across several tasks. Our approach enables estimates from a constrained model to serve as a prior for a more general model, yielding a principled way to interpolate between models of differing complexity. We successfully applied this approach to a number of decision and classification problems, as well as analyzing simulated brain imaging data. Models with robust priors had excellent worst-case performance. Solutions followed from the form of the heuristic that was used to derive the prior. These new algorithms can serve applications in data analysis and machine learning, as well as help in understanding how people transition from novice to expert performance.

Abstract Neural Representations of Category Membership beyond Information Coding Stimulus or Response.

For decades, researchers have debated whether mental representations are symbolic or grounded in sensory inputs and motor programs. Certainly, aspects of mental representations are grounded. However, does the brain also contain abstract concept representations that mediate between perception and action in a flexible manner not tied to the details of sensory inputs and motor programs? Such conceptual pointers would be useful when concepts remain constant despite changes in appearance and associated actions. We evaluated whether human participants acquire such representations using fMRI. Participants completed a probabilistic concept learning task in which sensory, motor, and category variables were not perfectly coupled or entirely independent, making it possible to observe evidence for abstract representations or purely grounded representations. To assess how the learned concept structure is represented in the brain, we examined brain regions implicated in flexible cognition (e.g., pFC and parietal cortex) that are most likely to encode an abstract representation removed from sensory-motor details. We also examined sensory-motor regions that might encode grounded sensory-motor-based representations tuned for categorization. Using a cognitive model to estimate participants' category rule and multivariate pattern analysis of fMRI data, we found the left pFC and human middle temporal visual area (MT)/V5 coded for category in the absence of information coding for stimulus or response. Because category was based on the stimulus, finding an abstract representation of category was not inevitable. Our results suggest that certain brain areas support categorization behavior by constructing concept representations in a format akin to a symbol that differs from stimulus-motor codes.

Bidirectional influences of information sampling and concept learning.

Contemporary models of categorization typically tend to sidestep the problem of how information is initially encoded during decision making. Instead, a focus of this work has been to investigate how, through selective attention, stimulus representations are "contorted" such that behaviorally relevant dimensions are accentuated (or "stretched"), and the representations of irrelevant dimensions are ignored (or "compressed"). In high-dimensional real-world environments, it is computationally infeasible to sample all available information, and human decision makers selectively sample information from sources expected to provide relevant information. To address these and other shortcomings, we develop an active sampling model, Sampling Emergent Attention (SEA), which sequentially and strategically samples information sources until the expected cost of information exceeds the expected benefit. The model specifies the interplay of two components, one involved in determining the expected utility of different information sources and the other in representing knowledge and beliefs about the environment. These two components interact such that knowledge of the world guides information sampling, and what is sampled updates knowledge. Like human decision makers, the model displays strategic sampling behavior, such as terminating information search when sufficient information has been sampled and adaptively adjusting the search path in response to previously sampled information. The model also shows human-like failure modes. For example, when information exploitation is prioritized over exploration, the bidirectional influences between information sampling and learning can lead to the development of beliefs that systematically differ from reality. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

The Costs and Benefits of Goal-Directed Attention in Deep Convolutional Neural Networks.

People deploy top-down, goal-directed attention to accomplish tasks, such as finding lost keys. By tuning the visual system to relevant information sources, object recognition can become more efficient (a benefit) and more biased toward the target (a potential cost). Motivated by selective attention in categorisation models, we developed a goal-directed attention mechanism that can process naturalistic (photographic) stimuli. Our attention mechanism can be incorporated into any existing deep convolutional neural networks (DCNNs). The processing stages in DCNNs have been related to ventral visual stream. In that light, our attentional mechanism incorporates top-down influences from prefrontal cortex (PFC) to support goal-directed behaviour. Akin to how attention weights in categorisation models warp representational spaces, we introduce a layer of attention weights to the mid-level of a DCNN that amplify or attenuate activity to further a goal. We evaluated the attentional mechanism using photographic stimuli, varying the attentional target. We found that increasing goal-directed attention has benefits (increasing hit rates) and costs (increasing false alarm rates). At a moderate level, attention improves sensitivity (i.e. increases d ' ) at only a moderate increase in bias for tasks involving standard images, blended images and natural adversarial images chosen to fool DCNNs. These results suggest that goal-directed attention can reconfigure general-purpose DCNNs to better suit the current task goal, much like PFC modulates activity along the ventral stream. In addition to being more parsimonious and brain consistent, the mid-level attention approach performed better than a standard machine learning approach for transfer learning, namely retraining the final network layer to accommodate the new task.

Prelimbic cortex maintains attention to category-relevant information and flexibly updates category representations.

Category learning groups stimuli according to similarity or function. This involves finding and attending to stimulus features that reliably inform category membership. Although many of the neural mechanisms underlying categorization remain elusive, models of human category learning posit that prefrontal cortex plays a substantial role. Here, we investigated the role of the prelimbic cortex (PL) in rat visual category learning by administering excitotoxic lesions before category training and then evaluating the effects of the lesions with computational modeling. Using a touchscreen apparatus, rats (female and male) learned to categorize distributions of category stimuli that varied along two continuous dimensions. For some rats, categorizing the stimuli encouraged selective attention towards a single stimulus dimension (i.e., 1D tasks). For other rats, categorizing the stimuli required divided attention towards both stimulus dimensions (i.e., 2D tasks). Testing sessions then examined generalization to novel exemplars. PL lesions impaired learning and generalization for the 1D tasks, but not the 2D tasks. Then, a neural network was fit to the behavioral data to examine how the lesions affected categorization. The results suggest that the PL facilitates category learning by maintaining attention to category-relevant information and updating category representations.

Category learning in rodents using touchscreen-based tasks.

Categorization is a fundamental cognitive function that organizes our experiences into meaningful "chunks." This category knowledge can then be generalized to novel stimuli and situations. Multiple clinical populations, including people with Parkinson's disease, amnesia, autism, ADHD and schizophrenia, have impairments in the acquisition and use of categories. Although rodent research is well suited for examining the neural mechanisms underlying cognitive functions, many rodent cognitive tasks have limited translational value. To bridge this gap, we use touchscreens to permit greater flexibility in stimulus presentation and task design, track key dependent measures, and minimize experimenter involvement. Touchscreens offer a valuable tool for creating rodent cognitive tasks that are directly comparable to tasks used with humans. Touchscreen tasks are also readily used with cutting-edge neuroscientific methods that are difficult to do in humans such as optogenetics, chemogenetics, neurophysiology and calcium imaging (using miniscopes). In this review, we show advantages of touchscreen-based tasks for studying category learning in rats. We also address multiple factors for consideration when designing category learning tasks, including the limitations of the rodent visual system, experimental design, and analysis strategies.