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.

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.

Disrupting dorsal hippocampus impairs category learning in rats.

Categorization requires a balance of mechanisms that can generalize across common features and discriminate against specific details. A growing literature suggests that the hippocampus may accomplish these mechanisms by using fundamental mechanisms like pattern separation, pattern completion, and memory integration. Here, we assessed the role of the rodent dorsal hippocampus (HPC) in category learning by combining inhibitory DREADDs (Designer Receptors Exclusively Activated by Designer Drugs) and simulations using a neural network model. Using touchscreens, we trained rats to categorize distributions of visual stimuli containing black and white gratings that varied along two continuous dimensions. Inactivating the dorsal HPC impaired category learning and generalization, suggesting that the rodent HPC plays an important role during categorization. Hippocampal inactivation had no effect on a control discrimination task that used identical trial procedures as the categorization tasks, suggesting that the impairments were specific to categorization. Model simulations were conducted with variants of a neural network to assess the impact of selective deficits on category learning. The hippocampal inactivation groups were best explained by a model that injected random noise into the computation that compared the similarity between category stimuli and existing memory representations. This model is akin to a deficit in mechanisms of pattern completion, which retrieves similar memory representations using partial information.

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.

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.

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.

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.

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.

Structured, uncertainty-driven exploration in real-world consumer choice.

Making good decisions requires people to appropriately explore their available options and generalize what they have learned. While computational models can explain exploratory behavior in constrained laboratory tasks, it is unclear to what extent these models generalize to real-world choice problems. We investigate the factors guiding exploratory behavior in a dataset consisting of 195,333 customers placing 1,613,967 orders from a large online food delivery service. We find important hallmarks of adaptive exploration and generalization, which we analyze using computational models. In particular, customers seem to engage in uncertainty-directed exploration and use feature-based generalization to guide their exploration. Our results provide evidence that people use sophisticated strategies to explore complex, real-world environments.

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.

Selective attention in rat visual category learning.

A prominent theory of category learning, COVIS, posits that new categories are learned with either a declarative or procedural system, depending on the task. The declarative system uses the prefrontal cortex (PFC) to learn rule-based (RB) category tasks in which there is one relevant sensory dimension that can be used to establish a rule for solving the task, whereas the procedural system uses corticostriatal circuits for information integration (II) tasks in which there are multiple relevant dimensions, precluding use of explicit rules. Previous studies have found faster learning of RB versus II tasks in humans and monkeys but not in pigeons. The absence of a learning rate difference in pigeons has been attributed to their lacking a PFC. A major gap in this comparative analysis, however, is the lack of data from a nonprimate mammalian species, such as rats, that have a PFC but a less differentiated PFC than primates. Here, we investigated RB and II category learning in rats. Similar to pigeons, RB and II tasks were learned at the same rate. After reaching a learning criterion, wider distributions of stimuli were presented to examine generalization. A second experiment found equivalent RB and II learning with wider category distributions. Computational modeling revealed that rats extract and selectively attend to category-relevant information but do not consistently use rules to solve the RB task. These findings suggest rats are on a continuum of PFC function between birds and primates, with selective attention but limited ability to utilize rules relative to primates.

Dynamic updating of hippocampal object representations reflects new conceptual knowledge.

Concepts organize the relationship among individual stimuli or events by highlighting shared features. Often, new goals require updating conceptual knowledge to reflect relationships based on different goal-relevant features. Here, our aim is to determine how hippocampal (HPC) object representations are organized and updated to reflect changing conceptual knowledge. Participants learned two classification tasks in which successful learning required attention to different stimulus features, thus providing a means to index how representations of individual stimuli are reorganized according to changing task goals. We used a computational learning model to capture how people attended to goal-relevant features and organized object representations based on those features during learning. Using representational similarity analyses of functional magnetic resonance imaging data, we demonstrate that neural representations in left anterior HPC correspond with model predictions of concept organization. Moreover, we show that during early learning, when concept updating is most consequential, HPC is functionally coupled with prefrontal regions. Based on these findings, we propose that when task goals change, object representations in HPC can be organized in new ways, resulting in updated concepts that highlight the features most critical to the new goal.

Social Information Is Integrated into Value and Confidence Judgments According to Its Reliability.

How much we like something, whether it be a bottle of wine or a new film, is affected by the opinions of others. However, the social information that we receive can be contradictory and vary in its reliability. Here, we tested whether the brain incorporates these statistics when judging value and confidence. Participants provided value judgments about consumer goods in the presence of online reviews. We found that participants updated their initial value and confidence judgments in a Bayesian fashion, taking into account both the uncertainty of their initial beliefs and the reliability of the social information. Activity in dorsomedial prefrontal cortex tracked the degree of belief update. Analogous to how lower-level perceptual information is integrated, we found that the human brain integrates social information according to its reliability when judging value and confidence.SIGNIFICANCE STATEMENT The field of perceptual decision making has shown that the sensory system integrates different sources of information according to their respective reliability, as predicted by a Bayesian inference scheme. In this work, we hypothesized that a similar coding scheme is implemented by the human brain to process social signals and guide complex, value-based decisions. We provide experimental evidence that the human prefrontal cortex's activity is consistent with a Bayesian computation that integrates social information that differs in reliability and that this integration affects the neural representation of value and confidence.

Estimating the functional dimensionality of neural representations.

Recent advances in multivariate fMRI analysis stress the importance of information inherent to voxel patterns. Key to interpreting these patterns is estimating the underlying dimensionality of neural representations. Dimensions may correspond to psychological dimensions, such as length and orientation, or involve other coding schemes. Unfortunately, the noise structure of fMRI data inflates dimensionality estimates and thus makes it difficult to assess the true underlying dimensionality of a pattern. To address this challenge, we developed a novel approach to identify brain regions that carry reliable task-modulated signal and to derive an estimate of the signal's functional dimensionality. We combined singular value decomposition with cross-validation to find the best low-dimensional projection of a pattern of voxel-responses at a single-subject level. Goodness of the low-dimensional reconstruction is measured as Pearson correlation with a test set, which allows to test for significance of the low-dimensional reconstruction across participants. Using hierarchical Bayesian modeling, we derive the best estimate and associated uncertainty of underlying dimensionality across participants. We validated our method on simulated data of varying underlying dimensionality, showing that recovered dimensionalities match closely true dimensionalities. We then applied our method to three published fMRI data sets all involving processing of visual stimuli. The results highlight three possible applications of estimating the functional dimensionality of neural data. Firstly, it can aid evaluation of model-based analyses by revealing which areas express reliable, task-modulated signal that could be missed by specific models. Secondly, it can reveal functional differences across brain regions. Thirdly, knowing the functional dimensionality allows assessing task-related differences in the complexity of neural patterns.

Heuristics as Bayesian inference under extreme priors.

Simple heuristics are often regarded as tractable decision strategies because they ignore a great deal of information in the input data. One puzzle is why heuristics can outperform full-information models, such as linear regression, which make full use of the available information. These "less-is-more" effects, in which a relatively simpler model outperforms a more complex model, are prevalent throughout cognitive science, and are frequently argued to demonstrate an inherent advantage of simplifying computation or ignoring information. In contrast, we show at the computational level (where algorithmic restrictions are set aside) that it is never optimal to discard information. Through a formal Bayesian analysis, we prove that popular heuristics, such as tallying and take-the-best, are formally equivalent to Bayesian inference under the limit of infinitely strong priors. Varying the strength of the prior yields a continuum of Bayesian models with the heuristics at one end and ordinary regression at the other. Critically, intermediate models perform better across all our simulations, suggesting that down-weighting information with the appropriate prior is preferable to entirely ignoring it. Rather than because of their simplicity, our analyses suggest heuristics perform well because they implement strong priors that approximate the actual structure of the environment. We end by considering how new heuristics could be derived by infinitely strengthening the priors of other Bayesian models. These formal results have implications for work in psychology, machine learning and economics.

Model comparison, not model falsification.

Systematically comparing models that vary across components can be more informative and explanatory than determining whether behaviour is optimal, however defined. The process of model comparison has a number of benefits, including the possibility of integrating seemingly disparate empirical findings, understanding individual and group differences, and drawing theoretical connections between model proposals.

Anterior Temporal Lobe Tracks the Formation of Prejudice.

Despite advances in understanding the brain structures involved in the expression of stereotypes and prejudice, little is known about the brain structures involved in their acquisition. Here, we combined fMRI, a task involving learning the valence of different social groups, and modeling of the learning process involved in the development of biases in thinking about social groups that support prejudice. Participants read descriptions of valenced behaviors performed by members of novel social groups, with majority groups being more frequently encountered during learning than minority groups. A model-based fMRI analysis revealed that the anterior temporal lobe tracked the trial-by-trial changes in the valence associated with each group encountered in the task. Descriptions of behavior by group members that deviated from the group average (i.e., prediction errors) were associated with activity in the left lateral PFC, dorsomedial PFC, and lateral anterior temporal cortex. Minority social groups were associated with slower acquisition rates and more activity in the ventral striatum and ACC/dorsomedial PFC compared with majority groups. These findings provide new insights into the brain regions that (a) support the acquisition of prejudice and (b) detect situations in which an individual's behavior deviates from the prejudicial attitude held toward their group.

Limits in decision making arise from limits in memory retrieval.

Some decisions, such as predicting the winner of a baseball game, are challenging in part because outcomes are probabilistic. When making such decisions, one view is that humans stochastically and selectively retrieve a small set of relevant memories that provides evidence for competing options. We show that optimal performance at test is impossible when retrieving information in this fashion, no matter how extensive training is, because limited retrieval introduces noise into the decision process that cannot be overcome. One implication is that people should be more accurate in predicting future events when trained on idealized rather than on the actual distributions of items. In other words, we predict the best way to convey information to people is to present it in a distorted, idealized form. Idealization of training distributions is predicted to reduce the harmful noise induced by immutable bottlenecks in people's memory retrieval processes. In contrast, machine learning systems that selectively weight (i.e., retrieve) all training examples at test should not benefit from idealization. These conjectures are strongly supported by several studies and supporting analyses. Unlike machine systems, people's test performance on a target distribution is higher when they are trained on an idealized version of the distribution rather than on the actual target distribution. Optimal machine classifiers modified to selectively and stochastically sample from memory match the pattern of human performance. These results suggest firm limits on human rationality and have broad implications for how to train humans tasked with important classification decisions, such as radiologists, baggage screeners, intelligence analysts, and gamblers.

Global neural pattern similarity as a common basis for categorization and recognition memory.

Familiarity, or memory strength, is a central construct in models of cognition. In previous categorization and long-term memory research, correlations have been found between psychological measures of memory strength and activation in the medial temporal lobes (MTLs), which suggests a common neural locus for memory strength. However, activation alone is insufficient for determining whether the same mechanisms underlie neural function across domains. Guided by mathematical models of categorization and long-term memory, we develop a theory and a method to test whether memory strength arises from the global similarity among neural representations. In human subjects, we find significant correlations between global similarity among activation patterns in the MTLs and both subsequent memory confidence in a recognition memory task and model-based measures of memory strength in a category learning task. Our work bridges formal cognitive theories and neuroscientific models by illustrating that the same global similarity computations underlie processing in multiple cognitive domains. Moreover, by establishing a link between neural similarity and psychological memory strength, our findings suggest that there may be an isomorphism between psychological and neural representational spaces that can be exploited to test cognitive theories at both the neural and behavioral levels.