Target-guided dual clustering active representation learning assisting numerical investigation of plasma-catalytic CO2 hydrogenation to methanol

Effective goal-directed decision making relies on memory and planning-processes that are known to decline with age. We tested the hypothesis that these declines stem from a common mechanism by focusing on mnemonic discrimination, a measure of memory precision that shows unique vulnerability to age-related decline. We used a latent learning task that measures the ability to learn and make judgments about associations among interconnected stimuli, assessing performance across the adult life span. This task allows us to measure multistep inference judgments that reflect how individuals organize relational structure, previously shown to capture the internal model-construction processes that support model-based planning. In Experiment 1, we examined relationships between judgment performance and memory precision. In Experiment 2, we tested whether a "blocked" learning schedule designed to reduce memory interference by separating overlapping objects could improve performance for individuals with weaker memory abilities. Across the life span, both younger and older adults showed evidence of successful latent learning and inference, but variability in judgment performance was explained by mnemonic discrimination ability. In Experiment 2, mnemonic discrimination interacted with training condition: intermixed training benefited those with high memory precision, whereas blocked training benefited those with low memory precision. We also implemented artificial neural network simulations, which reproduced these qualitative patterns. These findings suggest that age-related declines in complex judgments stem from declines in mnemonic discrimination and demonstrate that individualized, memory-based training interventions can improve learning and reasoning processes that support goal-directed planning, offering a promising approach to preserving decision-making abilities across the life span.

Latent learning experiments were critical in shaping Tolman's cognitive map theory. In a spatial navigation task, latent learning means that animals acquire knowledge of their environment through exploration, such that pre-exposed animals learn faster on a subsequent learning task than naive ones. This enhancement has been shown to depend on the design of the pre-exposure phase. Here, we hypothesize that the deep successor representation (DSR), a recent computational model for cognitive map formation, can account for the modulation of latent learning because it is sensitive to the statistics of behavior during exploration. In our model, exploration aligned with the future reward location significantly improves reward learning compared to random, misdirected, or no exploration, as reported by experiments. This effect generalizes across different action selection strategies. We show that these performance differences follow from the spatial information encoded in the structure of the DSR acquired in the pre-exposure phase. In summary, this study sheds light on the mechanisms underlying latent learning and how such learning shapes cognitive maps, impacting their effectiveness in goal-directed spatial tasks.

Edward Tolman, an experimental psychologist renowned for his research on the "cognitive map" and "latent learning" in rats, pursued his career within the tradition of behaviorisms. In the history of psychology, he is positioned as a precursor to cognitivism. This is because the concepts he introduced as intervening variables were later interpreted as having a representational nature. On the other hand, the philosophy of Maurice Merleau-Ponty was also influenced by Tolman's concepts, particularly his theory of sign-Gestalt. Merleau-Ponty's ideas, especially as developed in The Structure of Behavior, have had a profound impact on the notion of embodiment within enactivism. Enactivism, as a school of thought opposing representationalism, sharply contrasts with cognitivism, which places Tolman at the intersection of two opposing intellectual currents. This paper, thus, reexamines Tolman's ambivalent nature, exploring how it arose and reassessing his ideas as a precursor not only to cognitivism but also indirectly to enactivism. We argue that Merleau-Ponty inherited Tolman's concept of sign-Gestalt as having a non-representational nature and utilized it within his relational account of behavior. While Tolman's own theoretical framework remained ontologically ambiguous, Merleau-Ponty recast his insights into a conception of behavior as a structured, embodied interaction between organism and environment. These reformulated insights subsequently provided a conceptual foundation for enactivist cognitive science, particularly in its emphasis on the co-constitution of perception and action, and the dynamic coupling of agent and environment. Tolman maintained a behaviorist outlook throughout his early career, but his writing often remained ambiguous; as a result, that ambiguity enabled his work to exert influence on two opposing intellectual traditions at once.

Animals often balance asocial and social information strategically, adjusting when and from whom they copy based on context. Yet the cognition driving this dynamic-and its broader implications-remains poorly understood. We tested whether zebra finches use a copy-if-dissatisfied strategy by manipulating the quality of their initial nest-building or reproductive experience, showing them a conspecific nest-builder and tracking subsequent material choices. Builder-males were more likely to choose the demonstrated 'social' material-particularly at first choice-if they had previously used low-quality material. Using cognitive modelling, we estimated how latent learning mechanisms shaped decisions, identifying two asocial and two social parameters. These estimates provide the first formal evidence for the cognitive mechanisms of nest building. Forward simulations informed-but not predetermined-by these parameters approximated observed behaviour, supporting their causal role. We then used these parameters in exploratory simulations to test how choices might shift under novel payoff contexts. We found that payoff structure-not (dis)satisfaction-was the primary driver of social material use, though higher rewards did not proportionally increase copying. These exploratory simulation results offer preliminary insight into mechanisms underlying material-use variation. Our study illustrates how computational modelling can robustly link behaviour to underlying learning mechanisms and probe the generalizability of animal cognition-a rarity in this field.

The aim of this study is to address data sparsity, popularity bias, and insufficient diversity in knowledge point recommendation algorithms within intelligent tutoring systems. This study proposes a new model that dynamically adjusts negative sample probability and weighting based on global frequency and pedagogical difficulty to enhance recommendation precision and equity, especially for underrepresented or complex knowledge points. The study employs a publicly available dataset comprising 6,607 student records and 20 distinct variables. Without explicit curricular tags, knowledge points were operationalized as latent learning units identified through unsupervised clustering. Specifically, seven key performance and behavioral indicators, Hours_Studied, Attendance, Previous_Scores, Sleep_Hours, Tutoring_Sessions, Physical_Activity, and Exam_Score were selected as features for clustering. These features were first standardized using Z-score normalization. Subsequently, the standardized features were partitioned into ten clusters using the K-Means algorithm with a random_state of 42 to ensure reproducibility. The frequency of a knowledge point was defined as its relative prevalence within the dataset. The difficulty of each knowledge point was inferred from aggregated student performance within the corresponding cluster. Results certify achieving NDCG of 0.95, HRA10 of 1.00, and AUC of 0.96.

We propose a flexible Bayesian approach for sparse Gaussian graphical modeling of multivariate time series. We account for temporal correlation in the data by assuming that observations are characterized by an underlying and unobserved hidden discrete autoregressive process. We assume multivariate Gaussian emission distributions and capture spatial dependencies by modeling the state-specific precision matrices via graphical horseshoe priors. We characterize the mixing probabilities of the hidden process via a cumulative shrinkage prior that accommodates zero-inflated parameters for non-active components, and further incorporate a sparsity-inducing Dirichlet prior to estimate the effective number of states from the data. For posterior inference, we develop a sampling procedure that allows estimation of the number of discrete autoregressive lags and the number of states, and that cleverly avoids having to deal with the changing dimensions of the parameter space. We thoroughly investigate performance of our proposed methodology through several simulation studies. We further illustrate the use of our approach for the estimation of dynamic brain connectivity based on fMRI data collected on a subject performing a task-based experiment on latent learning. Supplementary materials for this article are available online.

Statistical learning (SL) enables the extraction of regularities from sensory input, yet the neural dynamics supporting this process—particularly in the visual modality—remain incompletely understood. Sixty-seven adults were familiarized with a continuous stream of shape sequences containing statistical structure that defined shape triplets. We recorded EEG to familiar sequences (presented in isolation) and unfamiliar foils. Both early (N100) and late (N400) event related potential (ERP) components were significantly more negative for unfamiliar than familiar sequences, reflecting robust neural sensitivity to learned structure. Notably, these familiarity effects were evident in both high- and low-performing participants and were not predicted by overall behavioral sensitivity, suggesting that neural indices of learning can emerge independently of explicit recognition. Follow-up analyses incorporating trial-level accuracy revealed a striking crossover interaction: for sensitive participants, ERP familiarity effects were stronger on correct trials, whereas for insensitive participants, effects were larger on incorrect trials. These findings highlight a dissociation between neural and behavioral measures of statistical learning and underscore the value of ERPs in capturing latent learning processes that may elude conscious awareness.

Dopamine is classically thought to drive learning based on errors in the prediction of rewards and punishments. However, animals also learn to predict cues with no intrinsic value or biological relevance to ongoing behavior, and it is unclear whether such latent learning also relies on dopaminergic prediction errors. Here, we tested this by recording dopamine release in the nucleus accumbens and dorsomedial striatum while rats executed a sensory preconditioning task that incorporated both types of learning. We found that dopamine release in both regions correlated with errors in predicting value-neutral cues during latent learning and with errors in predicting reward during reward-based conditioning. Moreover, dopamine in the nucleus accumbens reflected inferred value in the probe test, supported by orbitofrontal cortex activity. Our findings show that dopamine signals prediction errors about both valued and neutral stimuli, consistent with its operation as a general teaching signal that supports learning across different informational domains.

Exposure to arsenic is known to impair learning and memory functions in animal models and humans. However, the transgenerational inheritance of these cognitive deficits and the underlying epigenetic mechanisms remain poorly understood. The present study investigated the inter- and transgenerational effects of ancestral arsenic exposure on the cognitive performance (latent learning) of zebrafish via maternal and paternal lineages and the underlying biochemical and molecular alterations in the brain, including the DNA methylation patterns of cognition-related genes. Adult male zebrafish exposed to dietary arsenic [30, 60, or 100 μg/g as arsenite for 90 days; F0 generation] were crossed with unexposed (control) females and vice versa to generate F1 progeny of maternal and paternal arsenic exposure, respectively. Subsequently, F1 males and females of the same treatments were crossed to generate the F2 progeny of the respective maternal and paternal lineages of ancestral arsenic exposure. It was found that ancestral arsenic exposure induced cognitive dysfunction in F1 and F2 generations of both maternal and paternal lineages. However, the effects occurred at relatively lower levels of ancestral arsenic exposure (30 and 60 μg/g) in the former treatment relative to those (100 μg/g) in the latter. Inter (F1) and transgenerational (F2) cognitive effects of arsenic were associated with concomitant elevated oxidative stress and dopaminergic dysregulation, including repressed expression of cognition-related genes such as genes involved in dopamine signaling and metabolism (Drd1 and MAO) and the brain-derived neurotrophic factor (BDNF). Furthermore, DNA methylation analyses revealed that the downregulation of these genes across three generations (F0 to F2) resulted from the hypermethylation in their promotor regions (Drd1, MAO, BDNF). Collectively, these observations provide novel insights into the epigenetic mechanisms of the transgenerational inheritance of arsenic neurotoxicity.

In real-world settings, stimulus and outcome associations often depend on situational factors, such as Pavlovian occasion setters (OSs), which disambiguate whether a conditional stimulus (CS) will predict an outcome (unconditional stimulus; US). Whereas previous studies show that OSs are often lower in salience than CSs, no study has examined how low-salience OSs affect learning. In two conditioning experiments, we investigated this from the premise that inconsistently reinforced CSs prompt searching for additional stimuli (OSs) that indicate whether the CS will be followed by the US. Occasion setting learning was assessed using extinction rate-as partial reinforcement slows extinction relative to continuous reinforcement-and self-reported latent learning of stimuli. We hypothesized that a high-salience OS would result in faster extinction rates and occasion setting learning, whereas a low-salience OS would result in slower extinction rates and CS partial reinforcement learning. The results of Experiment 1 were mixed; there was no effect of OS salience on extinction rate, but the results for latent learning supported the hypothesis. We conducted Experiment 2 to specifically test extinction rate, and the results supported our hypothesis. The findings suggest that if a salient OS is found, occasion setting is learned; otherwise, CS partial reinforcement is learned.

Generative adversarial networks (GANs) prioritize pixel-level attributes over capturing the entire image distribution, which is critical in image synthesis. To address this challenge, we propose a dual-stream contrastive latent projection generative adversarial network (DSCLPGAN) for the robust augmentation of MRI images. The dual-stream generator in our architecture incorporates two specialized processing pathways: one is dedicated to local feature variation modeling, while the other captures global structural transformations, ensuring a more comprehensive synthesis of medical images. We used a transformer-based encoder-decoder framework for contextual coherence and the contrastive learning projection (CLP) module integrates contrastive loss into the latent space for generating diverse image samples. The generated images undergo adversarial refinement using an ensemble of specialized discriminators, where discriminator 1 (D1) ensures classification consistency with real MRI images, discriminator 2 (D2) produces a probability map of localized variations, and discriminator 3 (D3) preserves structural consistency. For validation, we utilized a publicly available MRI dataset which contains 3064 T1-weighted contrast-enhanced images with three types of brain tumors: meningioma (708 slices), glioma (1426 slices), and pituitary tumor (930 slices). The experimental results demonstrate state-of-the-art performance, achieving an SSIM of 0.99, classification accuracy of 99.4% for an augmentation diversity level of 5, and a PSNR of 34.6 dB. Our approach has the potential of generating high-fidelity augmentations for reliable AI-driven clinical decision support systems.

Generative Adversarial Networks (GANs) prioritize pixel-level attributes over capturing the entire image distribution which is critical in image synthesis. To address this challenge, we propose (DSCLPGAN) a dual-stream generator coupled with contrastive latent projection (CLP) for the robust augmentation of MRI images. The dual- stream generator in our architecture incorporates two specialized processing pathways: one dedicated to local feature variation modeling, while the other captures global structural transformations, ensuring a more comprehensive synthesis of medical images. We used a transformer-based encoder-decoder framework for contextual coherence and the contrastive learning projection (CLP) module integrates contrastive loss into the latent space for generating diverse image samples. The generated images undergo adversarial refinement using an ensemble of specialized discriminators where discriminator 1 (D1) ensures classification consistency with real MRI images, discriminator 2 (D2) produces a probability map of localized variations and discriminator (D3) functions for preserving structural consistency. For validation, we utilize a publicly available MRI dataset which contains from 3064 T1-weighted contrast- enhanced images with three types of brain tumor: meningioma (708 slices), glioma (1426 slices), and pituitary tumor (930 slices). Experimental results demonstrate state-of-the-art performance, achieving an SSIM of 0.99, classification accuracy of 99.4% for an at an augmentation diversity level of 5 and PSNR of 34. 6 dB. Our approach has the potential of generating high-fidelity augmentations for reliable AI-driven clinical decision support systems.

Alzheimer's disease (AD), the leading cause of dementia, affects 1 in 9 people aged 65 and older. The disease impacts patients on multiple levels, from memory and problem-solving issues to difficulties with basic functions and personality changes. Unfortunately, there is only a handful of FDA-approved drugs, and none of them offer an effective cure. Therefore, recent strategies have focused on preventing and delaying disease onset, rather than curing already developed pathological changes in the brain. In this study, we investigated the therapeutic potential of carnosine (CAR), a naturally occurring dipeptide known for its multimodal mechanism of action, such as the ability to mitigate neuroinflammation, oxidative stress, and deficiencies in neurotropic factors, all of which are connected with aging-related cognitive decline and an increased risk of developing dementia. For this purpose, we utilized an okadaic acid-induced zebrafish model of AD, which replicates some of the key features of the disease, including hyperphosphorylation of tau protein, changes in Aβ-fragments, and cognitive decline. By employing a latent learning behavioral assay and fast-scan cyclic voltammetry, we evaluated the effect of CAR on the prevention of cognitive decline and neurochemical changes in the AD-like zebrafish brain. Our findings revealed that CAR prevents impaired learning and motor dysfunction in a sex-dependent manner and reduces anxiety-like behavior. Additionally, we found that CAR inhibits dopamine release impairment. Hence, our study demonstrates the potential of CAR as a promising candidate for further investigations focused on identifying molecules that could potentially serve as therapeutics for delaying the onset of AD.

Latent learning is a process that enables the brain to transform experiences into "cognitive maps," a form of implicit memory, without requiring reinforced training. To investigate its neural mechanisms, we record from hippocampal neurons in mice during latent learning of spatial maps and observe that the high-dimensional neural state space gradually transforms into a low-dimensional manifold that closely resembles the physical environment. This transformation process is associated with the neural reactivation of navigational experiences during sleep. Additionally, we identify a subset of hippocampal neurons that, rather than forming place fields in a novel environment, maintain weak spatial tuning but gradually develop correlated activity with other neurons. The elevated correlation introduces redundancy into the ensemble code, transforming the neural state space into a low-dimensional manifold that effectively links discrete place fields of place cells into a map-like structure. These results suggest a potential mechanism for latent learning of spatial maps in the hippocampus.

Low-dose computed tomography (LDCT) offers reduced X-ray radiation exposure but at the cost of compromised image quality, characterized by increased noise and artifacts. Recently, transformer models emerged as a promising avenue to enhance LDCT image quality. However, the success of such models relies on a large amount of paired noisy and clean images, which are often scarce in clinical settings. In computer vision and natural language processing, masked autoencoders (MAE) have been recognized as a powerful self-pretraining method for transformers, due to their exceptional capability to extract representative features. However, the original pretraining and fine-tuning design fails to work in low-level vision tasks like denoising. In response to this challenge, we redesign the classical encoder-decoder learning model and facilitate a simple yet effective streamlined low-level vision MAE, referred to as LoMAE, tailored to address the LDCT denoising problem. Moreover, we introduce an MAE-GradCAM method to shed light on the latent learning mechanisms of the MAE/LoMAE. Additionally, we explore the LoMAE's robustness and generability across a variety of noise levels. Experimental findings show that the proposed LoMAE enhances the denoising capabilities of the transformer and substantially reduce their dependency on high-quality, ground-truth data. It also demonstrates remarkable robustness and generalizability over a spectrum of noise levels. In summary, the proposed LoMAE provides promising solutions to the major issues in LDCT including interpretability, ground truth data dependency, and model robustness/generalizability.

Multimodal neuroimaging provides complementary information critical for accurate early diagnosis of Alzheimer's disease (AD). However, the inherent variability between multimodal neuroimages hinders the effective fusion of multimodal features. Moreover, achieving reliable and interpretable diagnoses in the field of multimodal fusion remains challenging. To address them, we propose a novel multimodal diagnosis network based on multi-fusion and disease-induced learning (MDL-Net) to enhance early AD diagnosis by efficiently fusing multimodal data. Specifically, MDL-Net proposes a multi-fusion joint learning (MJL) module, which effectively fuses multimodal features and enhances the feature representation from global, local, and latent learning perspectives. MJL consists of three modules, global-aware learning (GAL), local-aware learning (LAL), and outer latent-space learning (LSL) modules. GAL via a self-adaptive Transformer (SAT) learns the global relationships among the modalities. LAL constructs local-aware convolution to learn the local associations. LSL module introduces latent information through outer product operation to further enhance feature representation. MDL-Net integrates the disease-induced region-aware learning (DRL) module via gradient weight to enhance interpretability, which iteratively learns weight matrices to identify AD-related brain regions. We conduct the extensive experiments on public datasets and the results confirm the superiority of our proposed method. Our code will be available at: https://github.com/qzf0320/MDL-Net.

BackgroundImpulse control disorders (ICDs) in Parkinson’s disease are associated with a heavy burden on patients and caretakers. While recovery can occur, ICDs persist in many patients despite optimal management. The basis for this interindividual variability in recovery is unclear and poses a major challenge to personalized health care. MethodsWe adopted a computational psychiatry approach and leveraged the longitudinal, prospective Personalized Parkinson Project (136 people with Parkinson’s disease, within 5 years of diagnosis) to combine dopaminergic learning theory–informed functional magnetic resonance imaging with machine learning (at baseline) to predict ICD symptom recovery after 2 years of follow-up. We focused on change in Questionnaire for Impulsive-Compulsive Disorders in Parkinson’s Disease Rating Scale scores in the entire sample regardless of an ICD diagnosis. ResultsGreater reinforcement learning signals during gain trials but not loss trials at baseline, including those in the ventral striatum and medial prefrontal cortex, and the behavioral accuracy score measured while on medication were associated with greater recovery from impulse control symptoms 2 years later. These signals accounted for a unique proportion of the relevant variability over and above that explained by other known factors, such as decreases in dopamine agonist use. ConclusionsOur results provide a proof of principle for combining generative model–based inference of latent learning processes with machine learning–based predictive modeling of variability in clinical symptom recovery trajectories. We showed that reinforcement learning modeling parameters predicted recovery from ICD symptoms in Parkinson’s disease.

Once Romania came under the influence of the Soviet Union, the new regime established in 1948 was integrated into the diplomatic system created by the Kremlin. The party-state in Bucharest participated in youth festivals, the world women's movement, intellectuals' conferences, the Christian Peace Movement, and particularly the World Peace Congress. All of these events served as propaganda instruments to discourage a more assertive Western anti-Sovietism. Since 1954, Khrushchev initiated a shift in Soviet policy, one focused on peaceful coexistence rather than confrontation between the two formerly irreconcilable political camps. The Romanian People's Republic, along with other state socialist countries, emulated this novel approach in Moscow. The article explores the impact of the new course on Romania’s cultural diplomacy in the late 1950s and the early 1960s. To establish credible diplomatic efforts, there was a need to rethink socialist culture and revalue the pre-communist heritage. This new culture was based on the Leninist principle of “the new culture, national in form, socialist in content”. It signaled that the regime had moved beyond de-Stalinization, and now possessed internal legitimacy that no longer relied on the presence of Soviet troops in the country. However, this socialist culture drew inspiration from postwar autochthonized Bolshevik (Stalinist) nationalism. The medieval tradition rediscovered by Soviet cultural studies (such as Dmitry Likhachev, Viktor Lazarev, or Boris Rybakov) fueled anti-Soviet sentiments in Bucharest, leading, by 1964, to a significant distancing from Moscow. This article traces the diplomatic efforts of the Romanian People's Republic during this period of latent learning of the new rules of the political game within the socialist camp.

The latent growth curve is a technique of structural equations which propose that psychological and social phenomena can be modeled in relationships between factors and indicators. The objective of this work was to establish the differences that reflect the learning of the use of anti-pandemic devices. A longitudinal study was carried out from 2020 to 2024 in a sample of 100 students from a public university in central Mexico who were selected from high school to university. The results show that there are differences, but these are not significant. Such findings agree with studies related to stigma towards anti-pandemic policies. In relation to this state of the art, it is recommended to extend the study to the stigma derived from confinement and distancing in order to establish the learning curve of self-care and prevention. The implications of the study on treatment adherence suggest that stigma would be a latent factor that would be mediating the relationship between the intercept factor and the latent slope factor.