Both humans and animals are capable of identifying statistical regularities in their environments, allowing them to navigate, adapt, and survive. Yet, little is known about how humans infer and respond to complex distributional properties beyond basic statistics like the mean and variance. Motivated by real-world situations such as increasingly frequent extreme weather events, we examined how people respond to bimodal reward distributions compared to unimodal ones. During the fMRI session, participants performed an arrow task in which a correct response to the arrow’s direction triggered the presentation of either a unimodal distribution (25% chance of 0 points, 50% chance of 500 points, 25% chance of 1000 points) or a bimodal distribution (40% chance of 0 points, 20% chance of 500 points, 40% chance of 1000 points), followed by feedback. In the post-fMRI session, participants chose between two options per trial. The choice pairs included bimodal versus unimodal distributions, bimodal distributions versus sure gains (ranging from 0 to 1000 in steps of 100), and unimodal distributions versus sure gains. All distributions were explicitly shown to participants, and feedback was provided after each choice. At the neural level (N = 23), activity in value-related regions reflected sensitivity to the shape of reward distributions; however, only the mean activity within the putamen predicted participants’ choice probabilities. At the behavioral level (N = 37), group-level preferences between bimodal and unimodal distributions appeared similar; however, substantial individual variability emerged, some participants consistently preferred bimodal lotteries, others favored unimodal ones, while some showed mixed tendencies. Additionally, participants preferred to select the riskier option after receiving 1000 points from a bimodal distribution than after the same outcome from a unimodal one. Together, these results indicate that humans are sensitive to the shape of reward distributions, both in their neural representations and in their choice behavior.

There is well-established evidence in decision making under risk that humans distort information about probability systematically—people are more sensitive to changes in probability at the extremes (e.g., a shift from 0 to 1% chance of occurrence) than at the mediocre values (e.g., a change from 33% to 34%). However, the uncertainty we face in real life, such as the potential of winning a lottery, seldom gets resolved immediately. As a result, the timing of uncertainty resolution can potentially impact probability distortion. To examine this question, we used a classic common-ratio design known to elicit probability distortion. In a lottery decision task, on each trial participants (n=20) chose between two lotteries different in the timing of uncertainty resolution. One lottery was resolved immediately, while the other was resolved after a delay (7–120 days). To rule out temporal discounting of delayed rewards, regardless of choice, subjects received the reward (in case of winning) at the later date. To investigate the possible connection with temporal discounting, subjects performed a standard intertemporal choice task on a separate day, choosing between an immediate monetary reward and a larger reward that would be received at a later date. At the group level, we found that (1) delaying the resolution of uncertainty of an option significantly reduced its subjective value, (2) probability distortion was consistently observed across different delays, and (3) individual subjects’ tendency to devalue later uncertainty resolution did not correlate with how they discount delayed rewards in the standard intertemporal choice task. Together, these findings suggest that the timing of uncertainty resolution does not interact with how we distort probability information and may even be dissociable from how we discount delayed rewards.

青春期是形塑自我概念的重要時期，除了早年依附經驗，也受到他人給予自我評價之社會回饋的影響。本研究以事件關聯電位（Event-Related Potential, ERP）來探討青少年與成年人在評價自我後的社會回饋之神經處理歷程，並檢視與依附關係的關聯性。研究紀錄27名大學生及34名高中生在社會評價典範（Social Judgment Paradigm, SJP）的表現。受試者須判斷人格特質詞彙（正向、負向）是否符合自我特質，接續會收到電腦認同性回饋（認同或否定），並紀錄回饋是否一致（電腦認同性是否與受試者的認同一致）。受試者也需要填寫行為量表以測量依附關係。結果顯示，青少年和成人皆對認同回饋所引發的RewP和P300振幅比否定回饋更正向，而不一致回饋引發LPP振幅比一致回饋更正向，尤其在電腦給予否定回饋時更為明顯。相關方面，在成人組中發現依附迴避與LPP回饋一致性效果呈現正相關。本研究結果顯示，社會回饋影響個體的自我認知與情感反應，認同回饋更能引發個體的注意力，而後續歷程反映社會回饋與自我認知間的衝突。而迴避依附型態似乎更為影響個體的回饋一致性的處理歷程。

Understanding speech in noisy environments is a significant perceptual challenge, but prior exposure to a clean version of that speech can substantially improve subsequent comprehension. This study investigated the neural mechanisms underlying this perceptual benefit using fMRI. We examined how clean exposure altered behavior, fMRI multivoxel patterns, and functional connectivity for Mandarin sentences masked at −12 dB SNR. In the scanner, each sentence appeared sequentially as Pre-Noisy, Clean, and then Post-Noisy across six runs with 24 sentences. 21 participants reported comprehension on every trial.

Behavior improved significantly, with Clean higher than Pre and Post higher than Pre. Univariate fMRI analysis identified broad effects across the language network, including the superior and middle temporal gyri (STG/MTG), supramarginal gyrus (SMG), angular gyrus (AG), inferior frontal gyrus (IFG), and dorsolateral prefrontal cortex (DLPFC). Support vector machine–based multivoxel pattern analysis (MVPA-SVM) decoded Pre, Clean, and Post above chance across STG, MTG, SMG, AG, IFG, precentral gyrus (PreCG), and DLPFC. Representational similarity analysis (RSA) showed that cross-condition alignment shifted after clean exposure, with Post-to-Clean exceeding Pre-to-Clean in STG, MTG, SMG, and IFG. A t-distributed stochastic neighbor embedding (t-SNE) revealed that, in STG/MTG, the Clean condition clustered at the center, Post-Noisy surrounded that center, and Pre-Noisy was farther out; in IFG, each sentence formed its own cluster, with Clean and Post-Noisy close together and Pre-Noisy more distant. Functional connectivity, assessed with generalized psychophysiological interaction (gPPI), revealed the network dynamics. Hearing Clean speech reduced connectivity between temporal and sensorimotor regions. Conversely, when noise returned (Post-Noisy), connectivity increased between the IFG and parietal/sensorimotor regions. Together, these findings suggest that clean exposure creates a stable reference representation in the temporal cortex, and when noise returns, a lateral IFG–parietal–sensorimotor circuit is then recruited to actively reconstruct the degraded input.

Sung speech, which combines musical and linguistic cues, is often thought to enhance verbal memory—a phenomenon known as the song memory advantage. According to the temporal scaffolding hypothesis, melody provides a structural framework that facilitates recall of sung materials. However, existing evidence remains inconsistent: some studies report memory benefits, whereas others show null or even negative effects. This study investigated how melodic and rhythmic structures in sung mnemonics influence sentence memory. A 2 × 2 design manipulated pitch (fixed/varied) and rhythm (fixed/varied) across four sung conditions and a spoken baseline (96 ten-syllable sentences in total). Participants first performed an immediate recall task, typing each sentence they heard, followed by an embedded melody recognition task to ensure attention to contour. To assess long-term retention, the same participants returned one week later for a delayed recognition test (old/new judgments), with sensitivity indexed by d′. Results show that recall accuracy was higher for spoken than sung speech. Within sung speech conditions, fixed-rhythm sung speech enhanced recall regardless of pitch variations. In contrast, varied-rhythm sung speech impaired recall, particularly when paired with variable-pitch contours. Melody recognition accuracy exceeded 75%, confirming attention to melodic features. At the one-week recognition test, recognition sensitivity (d′) remained above chance, again favoring spoken over sung sentences. Delayed recognition was highest for varied-pitch, fixed-rhythm sung speech, and lowest for fixed-pitch, varied rhythm sung speech, suggesting that rhythmic regularity supports long-term memory consolidation. Overall, findings suggest that rhythm plays a more critical role than pitch in supporting memory for sung speech, with implications for optimizing sung mnemonics for memory enhancement in clinical and non-clinical settings.

Decision making often depends on a reference point, traditionally thought to be fixed, reflecting a decision maker’s status quo. Recent theories proposed a stochastic reference point shaped by a decision maker’s expectations. A key distinction between these two proposals lies in how varying degrees of risk a decision maker expects would impact choice behavior. While the fixed-point model predicts no impact, the stochastic model predicts less risk aversion when more risk is expected. To formally compare these models, we designed a between-subject lottery decision task. Subjects chose between a ‘reference’ lottery (same throughout the experiment) and a varying lottery (162 trials). Participants were randomly assigned to one of two groups, each with a distinct reference lottery: a sure gain (100%, $1,000; no-risk reference group) or a risky lottery (10%, $10,000; high-risk reference group). The varying lotteries were generated from a 4-by-4 lottery matrix comprising outcome ($5000, $2500, $1450, $1150) and probability (0.2, 0.4, 0.7, 0.9). A subset of trials included an identical choice pair, (10%, $10,000) vs. (100%, $1,000), across both groups. Behaviorally, the high-risk reference group chose the risky option more frequently than the no-risk reference group, supporting the stochastic reference point model. Moreover, risk preference varied systematically with the risk of the varying lottery in opposite directions between groups: in the high-risk reference group, preference for risk decreased as varying lotteries became less risky, whereas in the no-risk reference group, preference for risk increased. While such ‘risk-preference gradients’ can be captured by both models, the stochastic model predictions better matched participants’ choice behavior in the no-risk reference group. Together, these results suggest humans form probabilistic representations of reference points based on recent experience and use them to guide decision making.

Developmental prosopagnosia (DP), is a neurodevelopmental condition characterized by persistent difficulties in recognizing faces, despite normal vision and intelligence. In childhood, DP often goes unrecognized, in spite of profound social, emotional, and academic consequences. Although research on adult DP has advanced rapidly in recent years, systematic investigations of childhood DP remain scarce and methodologically heterogeneous. This review synthesizes the literature on childhood DP, outlines conceptual and methodological challenges, and evaluates available screening and assessment approaches. A recent review identified 224 studies on DP, yet only 14 examined children, and just 10 reported confirmed cases since the first documentation in 1976. The review further highlights: (1) the unique social, safety, and self-identity challenges in childhood DP; (2) factors limiting progress in child-focused research; (3) a summary of existing findings; (4) an outline of the current assessment tools; and (5) a proposal for a child-adapted measure, the PI-25 Child, extending the adult PI-20 (Shah et al., 2015). Additionally, it is misleading to use “face blindness” as a synonym of DP, as it reflects difficulty recognizing faces rather than a total loss of face perception. Finally, we conclude by highlighting the need for standardized, early identification of childhood DP and call for stronger integration between research and clinical practice to promote early screening and support.

Belief polarization pervades socio-scientific debates, yet the neurocognitive bases of how beliefs change remain unclear. We investigated belief updating about nuclear power using fMRI, testing how individual differences shape responses to supporting versus opposing evidence. 87 participants viewed 80 minutes of naturalistic video segments on nuclear energy (17 supportive, 18 opposing) with presentation order counterbalanced across groups. We used inter-subject representational similarity analysis (IS-RSA) to quantify neural reconfiguration. Structural equation models (SEMs) estimated contributions from six questionnaire-measured traits: Big Five Inventory (BFI-44), Intolerance of Uncertainty Scale (IUS-12), Need for Closure (NFC), Interpersonal Reactivity Index (IRI), Positive and Negative Affect Schedule (PANAS), and Short Schwartz Value Survey (SVS). 

Behaviorally, both perspectives shifted beliefs in the expected directions. IS-RSA revealed an updating network spanning the dorsomedial prefrontal cortex (DMPFC) and posterior cingulate cortex (PCC). Belief change for both perspectives was associated with neural similarity changes in the anterior ventrolateral prefrontal cortex (VLPFC). Notably, changes under opposing information related to the ventromedial prefrontal cortex (VMPFC) and PCC, whereas changes under supporting information related to the nucleus accumbens (NAcc). In SEMs, only NFC (β = −0.357, p = .0089) and PANAS Negative Affect (β = 0.352, p = .0029) reliably predicted belief change under opposing evidence; supportive exposure effects were non-significant (NFC β = 0.275, p = .066; NA β = −0.116, p = .382).

These findings reveal that belief updating about controversial issues is context-sensitive. Avoidance of ambiguity and negative affect jointly predict how individuals respond to belief-challenging perspectives, providing a neurocognitive model for motivated reasoning.

Recently, it became clear that bacteria in the gut form extensive mutual relationships with the brain. Nevertheless, research on gut-brain interactions in Humans is relatively limited, and almost nothing is known about the relationship between the gut microbiome and perception, particularly visual perception.

We have used two versions of the Continuous Flash Suppression (CFS) paradigm as a mode to assess visual conscious perception. In the first case (N=47), we used a rapid serial visual presentation of Mondrian shapes shown to the dominant eye and a Gabor pattern shown to the non-dominant eye. In the second case (N=46), we used shallower suppression stimuli in the form of natural scene images shown to the dominant eye and a bullseye pattern shown to the non-dominant eye.

In parallel, we collected stool samples from our participants and performed 16S rRNA V3-V4 amplicon sequencing. After standard quality-control filtering steps and using QIIME2 pipeline, we obtained an amplicon sequence variant (ASV) table. The ASVs were collapsed to the species level for further analysis. We then used an in-house script and commonly used R packages for microbiome analysis to calculate alpha diversity for each individual's gut microbiome.

We obtained a bimodal histogram of alpha diversity values, with a leading group of high alpha diversity and a smaller group of lower alpha diversity. After filtering out individuals with low alpha diversity (N=6), we obtained a statistically significant correlation between alpha diversity values and the total switching period in the CSF paradigm. Bayes Factor also provided support for the hypothesis that the alpha diversity and CFS total switching duration are correlated.

This result suggests that the gut microbiome contributes to some of the CFS-related individual differences. Furthermore, this is the first time a possible relationship between the gut microbiome and visual perception has been demonstrated.

Many studies have shown that individuals with autism spectrum disorder (ASD) often have impairments in emotion recognition and atypical expression imitation, which can lead to misunderstandings when interpreting social situations or expressing internal emotional states. This study aims to evaluate the effects of an eight-week home-based intervention program for adults with ASD, using a pre- and post-test design with an experimental and a waiting-list control group. We recruited 32 neurotypical adults and 32 adults with ASD, matched for age and education. All neurotypical participants completed two computerized Dynamic Emotion Recognition and Expression Imitation tasks and three standardized measures: the Autism-Spectrum Quotient (AQ), the 20-Item Prosopagnosia Index (PI-20), and the online Cambridge Face Memory Test (CFMT). The performance of neurotypical adults will serve as the benchmark. The ASD group is divided into two subgroups. The experimental group completes a pre-test, the 8-week intervention, and a post-test. We developed a home-based intervention website that provides weekly links to computerized games and quizzes designed to enhance discrimination among facial expressions. At the pre-test, the ASD group received the same tasks as the neurotypical group, with two additional measures: the Reading the Mind in the Eyes Test (RMET) and the Toronto Alexithymia Scale (TAS-20). The waiting-list group receives the intervention after the post-test. Preliminary results show that at baseline, the ASD group scored higher on the AQ and PI-20, lower on the CFMT, and showed significantly lower accuracy in emotion recognition than the neurotypical group. At the post-test, the ASD group’s accuracy in emotion recognition improved significantly after intervention, while expression imitation, CFMT, RMET, and TAS-20 scores showed no significant change. Our study suggests a promising web-based intervention that may be particularly helpful for improving the ability to recognize emotional expressions.

Title: Computational Thematic Analysis of Gender Identity Narratives Using Large Language Models

Background: Traditional qualitative analysis of gender identity experiences is labor-intensive and may be limited by researcher subjectivity. Large language models (LLMs) offer potential for systematic and reproducible thematic analysis of interview data while maintaining interpretive depth.

​Methods: We analyzed semi-structured interview transcripts from 15 trans individuals in Turkey by using a keyword-based retrieval system integrated with LLaMA 3.1-8B Instruct Transcripts were processed through iterative query-based analysis where relevant text segments were extracted using case-insensitive keyword matching with contextual windowing. Retrieved segments were ranked by keyword frequency, and top-ranked results were provided as context to the LLM for generating thematic interpretations using Turkish-language prompts.

​Results: Preliminary thematic analysis identified five recurring themes: (1) early childhood identity recognition and self-awareness, (2) navigation of societal gender role expectations and norms, (3) internal psychological struggles related to gender identity, (4) community visibility and organizing through LGBT+ organizations, and (5) family relationships and gender expression dynamics. Themes were identified through iterative querying focusing on identity-related concepts, societal interactions, and personal experiences.​

​Conclusions: Keyword-based retrieval combined with LLM-assisted analysis provides an efficient approach to exploring gender identity narratives. Preliminary thematic analysis identified patterns related to identity recognition, societal navigation, and community experiences. The methodological framework demonstrates potential for computational approaches to qualitative analysis in understudied populations, though findings require validation through traditional coding methods. Future research could systematically examine linguistic patterns distinguishing identity awareness from aspirational desire, building on this methodological foundation.

Generalized Anxiety Disorder (GAD) is characterized by excessive and uncontrollable worry that interferes with emotional regulation, cognitive control, and executive functioning. Previous neuroimaging studies have emphasized the critical roles of the prefrontal cortex (PFC) and anterior cingulate cortex (ACC) in attention and emotion regulation in GAD. However, most have relied on fMRI, which provides high spatial but limited temporal resolution, making it insufficient to capture rapid neural dynamics. To address this limitation, the present study integrated event-related potentials (ERP) with Dynamic Causal Modeling (DCM) to investigate time-resolved cortical connectivity in GAD during a color-word Stroop task.

Twenty patients with GAD and twenty age- and gender-matched healthy controls performed the task while EEG was recorded. ERP results revealed significantly reduced N100, P200, N200, and N450 amplitudes in GAD, indicating impairments in sensory encoding, attentional allocation, and conflict monitoring. Although the P300 reduction did not reach significance, it suggested inefficiency in cognitive updating and working memory.

Bayesian model selection identified Model 3 as optimal for both groups, supporting a dominant top-down modulation mechanism. DCM further showed that within the intrinsic A matrix, the temporo-parietal junction (TPJ) → right medial superior parietal cortex (RMSPC) pathway was weakened in GAD, reflecting impaired baseline attentional regulation. In the task-modulatory B matrix, GAD patients exhibited decreased TPJ→RMSPC modulation, reversed ACC→TPJ directionality, and enhanced left-to-right frontal coupling (LFO→RFO), suggesting compensatory frontal control.

Together, these findings reveal disrupted top-down regulation and weakened PFC–ACC communication in GAD, highlighting dysfunctional dynamic connectivity underlying cognitive-affective inefficiency in anxiety disorders.

A central goal of functional MRI (fMRI) research is to compare activation patterns across subjects and conditions, often visualized as group-level statistical maps. However, intersubject variability in cortical anatomy poses challenges for accurate alignment. Surface-based analysis offers improved spatial precision over volume-based methods by aligning data according to the two-dimensional geometry of the cortical sheet, enhancing anatomical correspondence and reducing signal contamination between adjacent gyri and sulci. To perform surface-based analysis, functional data must be mapped from voxel (volume) to vertex (surface) space, which requires cortical surface reconstruction from structural MRI. FreeSurfer is widely used for this purpose but is computationally expensive, leading to high processing time and energy consumption. To explore more efficient alternatives, we compared several open-source surface reconstruction tools—including FreeSurfer, FastSurfer, FastCSR, and Vox2Cortex—using a publicly available task-based fMRI dataset of 76 subjects performing motor tasks (tongue, fingers, and feet movements). Among the tested methods, FreeSurfer and FastSurfer successfully generated high-quality cortical surfaces suitable for surface-based fMRI analysis. We registered each subject’s functional data to these surfaces and conducted group-level GLM analyses to compare cortical activation patterns across the two pipelines. FastSurfer produced surfaces with accuracy comparable to FreeSurfer while requiring substantially less processing time, energy, and carbon emissions, being approximately 4.5 times faster and generating about 74% less carbon emissions. Additionally, a performance evaluation on 20 subjects using different numbers of CPU threads and GPU configurations showed that the 8-thread GPU—the fastest among the eight tested profiles—reduced computation time by about 61% compared to the slowest run, and reduced carbon emissions by 60% compared to the highest-emission profile. This project demonstrates that using FastSurfer, along with an appropriate computational profile for surface reconstruction can significantly reduce both runtime and environmental impact.

Tickertape synaesthesia (TTS) is a rare phenomenon in which individuals “see” words in their mind when hearing someone speak (Francis, 1883). People with TTS are thought to have a unique connection between visual imagery and speech, giving them advantages in tasks that require mentally visualizing spoken words. However, the identification of TTS still relies heavily on subjective measures such as self-reports (Chun & Hupé, 2013; Holm, Eilertsen, & Price, 2015), and existing tasks are inadequate for non-native English speakers (Hauw, Soudany, & Cohen, 2023).

This study aimed to develop novel Mandarin word-shape tasks to provide objective criteria for identifying TTS. Participants completed three auditory–visual tasks in which they heard two-character Mandarin words and were asked to count the number of (1) horizontal lines, (2) vertical lines, and (3) quadrilateral components (closed, four-sided orthographic enclosures (e.g., 口, 日, 田)) in the corresponding traditional Han characters. After the tasks, participants completed a four-item Mandarin questionnaire translated from an existing English measure (Holm, Eilertsen, & Price, 2015).

We recruited 27 participants and applied the Crawford and Howell modified single-case t-test to compare one tickertape participant with six controls. The participant with TTS responded with equal accuracy but significantly faster in the horizontal and vertical tasks than controls [t(5) = –3.61, p = .02; t(5) = –2.82, p = .048], whereas controls showed higher accuracy in the quadrilateral task [t(5) = –2.99, p = .040]. We suspect that the quadrilateral task may have been too easy, allowing participants without strong mental visualization of spoken words to perform equally well, thus failing to distinguish between the two groups.

We suggest that individuals with TTS demonstrate behavioral advantages in visualizing spoken words. Future studies will refine task difficulty and validation procedures to enhance the reliability of objective identification methods for TTS.

What and how unconsciousness is initiated and affects the brain are important questions waiting to be answered. One way to explore and gain control of a subject’s state is to administer sedative medicine to induce unconsciousness. With this, we seek to investigate the dynamic changes underlying the state of unconsciousness. Thus, we used openly available fMRI data of seventeen participants. Each of them is scanned with an external auditory stimulus and during an internally oriented resting state under 4 levels of sedation: before, light, deep, and recovery. The dynamic state of the brain is acquired through the Kuramoto order parameter, to measure the level of synchronization and metastability, which can be indicative of the information flow. The current analyses show changes in metastability during deep sedation and a cross-effect on the auditory input and the state of the subject. However, further investigation is still needed for the metastability of each participant who changes variously at different stages of the sedation.

Time persistence is a fundamental property of complex systems that has recently been used as a framework for quantifying dynamic variance in brain activity (Guo et al., 2025). This system-level property demonstrates that the brain, as a whole network, retains certain network-connected configurations while seemingly gradually changing over time. As a potential stable network-level measure quantifying the high-variance dynamic measure, such as fMRI resting-state, this network persistence has been examined only in a public dataset, not in a smaller dataset with specific individual behavior trait measures. To this end, we conducted a replication fMRI resting-state study with 60 subjects. Using a different preprocessing method of Guo and colleagues, the same noise threshold and network time-persistence estimation were applied, along with link- and node-level analyses of functional connectivity ROIs across different time windows. To our surprise, with noise thresholding, neither the link nor the node-level survived and didn't demonstrate the robust pattern of persistence. We conducted a simple control analysis by correlating functional connectivity across time windows, and moderate persistence was observed, with an average Pearson's r of 0.406. Lastly, a lower noise threshold was applied to observe the pattern of consistency, and individual differences in the link persistency measure were extracted and correlated with trait rumination and the depression index; no relationship was found between the behavior index and the link persistency pattern. In this replication study, we observed a moderate time-consistency pattern, but the network-level persistence effect at the node and link levels was not found. This finding points to the potential that the current time persistence measure is not robust across different preprocessing pipelines.

Migraine is a common neurovascular disorder with a global prevalence of approximately 14%, predominantly affecting individuals aged 10–30 years. It is characterized by unilateral, pulsating, and recurrent headaches, often accompanied by nausea, vomiting, photophobia, and phonophobia. Current treatments mainly rely on analgesics such as acetaminophen and nonsteroidal anti-inflammatory drugs (NSAIDs); however, long-term use may lead to side effects or medication overuse. Therefore, developing novel therapies with fewer side effects and neuroprotective potential is of great importance. Curcumin (CUR) possesses anti-inflammatory, antioxidant, and neuroprotective properties, but its lipophilic nature results in poor bioavailability. Piperine (BP) enhances CUR absorption by inhibiting its metabolism. In this study, Neuro-2a cells were used as an in vitro neuronal model, and depolarization was induced with potassium chloride (KCl) to simulate cortical spreading depression (CSD) involved in migraine pathology, evaluate the therapeutic effects of CUR, BP, and their combination. Results showed that KCl significantly reduced Neuro-2a cell viability, whereas the combination of CUR and BP at a 100:1 ratio under 10 mM markedly increased cell survival. These findings suggest that mixture of curcumin and piperine in specific ratios can effectively attenuate KCl-induced neuronal damage, offering potential as a natural therapeutic strategy for migraine.

Alcohol Use Disorder (AUD) remains a major public health and socioeconomic challenge with limited long-term treatment efficacy. While deep brain stimulation (DBS) targeting the nucleus accumbens (NAc) has shown promising effects in reducing alcohol craving and consumption (Coles et al., 2018), its invasiveness restricts clinical application . We investigated a non-invasive neuromodulation approach using stereotactic radiosurgery to deliver low-dose, focal irradiation (Dmax = 30 Gy, 5 mm collimator) bilaterally to the NAc. The study utilized Lee Sung minipigs (LSPs) with a well-established history of voluntary alcohol consumption developed over more than two years of operant training. Three alcohol-exposed LSPs that reached the human binge-drinking equivalent (~0.8 g/kg/day) underwent radiosurgery, while two age-matched animals with comparable drinking histories served as non-surgical controls. Following radiosurgery, the treated animals exhibited a marked reduction in voluntary alcohol intake in both the fixed-ratio (FR = 20; relatively easy) and progressive fixed-ratio (PFR = 10–90, in increments of 10 with 30 trials per FR; relatively difficult) tasks. In contrast, alcohol consumption in the control animals remained unchanged over the same period. Despite reduced alcohol intake, body weight in the irradiated animals increased normally (Köhn, 2012), indicating that appetitive motivation was not adversely affected by the intervention. Moreover, structural and functional magnetic resonance imaging (MRI) was conducted immediately pre-surgery and up to one year post-surgery, focusing on the NAc and dorsal anterior cingulate cortex (dACC). Preliminary findings revealed that alcohol-craving LSPs exhibited smaller NAc volumes and reduced NAc–dACC connectivity compared to healthy controls. Notably, NAc–dACC connectivity was partially restored 12 months after radiosurgery, suggesting a potential neural mechanism for the reduced craving. Overall, these findings demonstrate that stereotactic radiosurgery can non-invasively modulate NAc circuitry and attenuate alcohol-seeking behavior, establishing a promising foundation for future translational and clinical applications in the treatment of AUD.

Background: Despite improved survival rates resulting from advances in acute stroke care, cognitive difficulties remain prevalent during the subacute recovery phase. Social engagement has been recognized as a critical factor supporting cognitive maintenance in aging populations; however, how social network dynamics evolve during stroke rehabilitation and relate to cognitive improvement remains largely unexplored. 

Methods: Thirty-six individuals with subacute stroke (22 males and 14 females; aged 55–85 years) were recruited for this study. Each participant completed a comprehensive neurocognitive assessment battery and social network questionnaires during both the admission phase of hospitalization and at discharge. Standardized change scores were calculated to evaluate variations in cognitive performance, depressive symptoms, and social network indices the two assessment points.

Results: Participants showed significant cognitive improvement at discharge compared with the admission phase of hospitalization, particularly on the digit span, verbal fluency, digit symbol substitution, symbol search, and both parts of the color trail making test. Reductions in depressive symptoms were also observed, reflecting improved psychological status. Social network analyses revealed significant increases in out-degree and betweenness centrality at discharge, indicating greater engagement and integration within the hospital social environment. Furthermore, higher betweenness centrality was positively associated with performance on the digit symbol substitution test, whereas greater network constraint was associated with longer completion times on the color trail making test part B.

Conclusions: This study provides evidence that social networks among subacute stroke patients undergo dynamic changes during hospitalization and that these network properties are closely associated with cognitive recovery. Specifically, lower network constraint and higher social connectivity were related to improved executive functioning and cognitive flexibility, suggesting that more open and diverse social networks may support post-stroke neurocognitive rehabilitation.

Sleep is a fundamental biological process essential for maintaining health and daily functioning. While past research has primarily focused on sleep duration and quality, emerging evidence highlights sleep regularity as an equally critical factor for overall well-being. During childhood, maintaining regular sleep–wake schedules across weekdays and weekends is especially important, as early school start times often conflict with children’s biological sleep preferences. Social jetlag (SJL)—the misalignment between biological and social sleep schedules—represents a common manifestation of irregular sleep and has been linked to poorer academic and cognitive outcomes, yet the neural mechanisms remain unclear. Using actigraphy and structural MRI data from over 3,000 children (ages 10.6-13.7 years) in the Adolescent Brain Cognitive Development (ABCD) Study, this research examined interrelations among objectively measured SJL, cognitive performance, and brain structure. SJL was quantified using the sleep-corrected formula, which controls for differences in total sleep time between weekdays and weekends. Linear mixed-effects models revealed that greater SJL predicted lower performance in language-related domains, including Picture Vocabulary (β = –0.095, p < .001), Oral Reading (β = –0.063, p < .001), and the Crystallized Composite (β = –0.082, p < .001). Higher social jetlag was also associated with reduced cortical volume within the language network (β = –0.0256, p < .01), but not within a motor control network. Mediation analyses confirmed that language-network volume partially accounted for the social jetlag–cognition association. Significant indirect effects were observed for Picture Vocabulary (β = −0.004, p = .010), Oral Reading (β = −0.005, p = .010), and the Crystallized Composite (β = −0.005, p = .007). These findings identify a structural pathway through which irregular sleep timing affects language-based cognitive development, suggesting that sleep regularity may be a more sensitive predictor of neural and cognitive outcomes in childhood.

Perceptual narrowing theory posits that early face perception begins broadly tuned to a range of categories, including nonhuman faces, and progressively specializes for human faces through experience. To test this account, we examined novelty preferences for human, chimpanzee, and monkey faces, as well as for whole-bodied animal stimuli, in infants younger than nine months and those nine months or older. Using a habituation-dishabituation paradigm, we quantified relative looking times toward novel versus familiar exemplars and assessed how these responses changed with age. Younger infants exhibited dishabituation to human faces and to whole-bodied animal stimuli. In contrast, they showed little evidence of discrimination for chimpanzee or monkey faces. Older infants demonstrated an emerging novelty preference for chimpanzee faces, while preferences for human faces and whole-bodied stimuli decreased with age. Crucially, no age-related decline in human-face discrimination was observed, contradicting the expected pattern of perceptual narrowing. These findings support a model of representational growth rather than representational loss. Early in development, attention is drawn to familiar, socially salient, and visually coherent categories that offer stable learning signals. With maturation, the representational space expands to encompass more complex or less familiar configurations, such as chimpanzee faces. This trajectory aligns with curiosity-driven or learning-progress models, in which attention is adaptively directed toward stimuli that maximize informational gain. Overall, the results suggest that early perceptual development reflects an exploratory reallocation of attention rather than a reduction in sensitivity. The first year of life may thus mark a transition from early specialization toward a flexible, learning-based expansion of representational capacity across animate categories.

Emotions rise and fall in daily life, and a key part of this dynamic is emotional valence. Prior work has measured emotional valence based on internal subjective feelings via self-report or external expressions via facial expressions, but it remains unclear whether these two measures depend on the same underlying neural mechanisms during naturalistic video-viewing experiences. We addressed this question using fMRI with a sample size of 70 participants across 18 naturalistic videos, each paired with concurrent traces of self-rated valence and a facial-derived valence index, which we specified as the valence scores derived from facial expressions of emotions. For each video, we trained a separate decoding model that mapped whole-brain activity to either self-rated or facial-derived valence using LASSO-PCR, with a K-fold cross-validation procedure. We then selected a subset of videos that showed reliable predictive performance for both self-rated and facial-derived valence indices, with Pearson correlations ranging from 0.2 to 0.6. We found that the brain patterns for self-rated and facial-derived valence overlapped only partially. Brain models for self-rated valence were also more consistent across videos, with predictive weights frequently observed in the orbitofrontal cortex (OFC) and dorsolateral prefrontal cortex (DLPFC). By contrast, brain models for facial-derived valence varied across videos. These findings suggest that subjective appraisal is supported by relatively stable brain patterns that generalize across diverse contexts, whereas facial-derived valence is more context-dependent. This dissociation indicates that “feeling” and “showing” emotional valence are not interchangeable at the neural level. Our findings indicate that self-rated valence, serving as an internal appraisal, is less impacted by context, whereas facial-derived valence is more directly driven by external cues and is thus more sensitive to what people are attending to and expressing at each moment.

Deception engages cognitive control, but psychopathic traits may shape how control is implemented and whether effects depend on content. We screened an online sample (n = 291) with the Self-Report Psychopathy Scale–III and invited the top/bottom scorers to fMRI. The fMRI sample comprised a high-psychopathy group (HighP; n = 27) and a low-psychopathy group (LowP; n = 30). Using a paradigm adapted from Ofen et al. (2017), participants judged belief statements or episodic prompts under honest or deceptive instructions in a 2 × 2 within-subject design (Content × Honesty) crossed with Group. Lying yielded lower accuracy and shorter response times than truth; within lies, belief deception showed lower accuracy and longer response times than episodic deception. Whole-brain GLMs tested Lie > Truth, Episodic > Belief, their interaction, and group differences, using a cluster-forming threshold of voxel-wise p = .01 with cluster extent k ≥ 106 (uncorrected). Lie > Truth recruited bilateral lateral prefrontal cortex (IFG/MFG/SFG), inferior parietal lobule, and precuneus. Episodic > Belief engaged frontoparietal regions and angular gyrus with posterior cingulate/cingulate cortex, whereas Belief > Episodic was left-lateralized in prefrontal cortex. Group effects were directionally specific: for Lie > Truth, LowP > HighP in bilateral MFG, right IFG, and left MOG, but HighP > LowP in left medial PFC/ACC. For belief lies, LowP > HighP in right IFG/insula and left MFG/MOG; for episodic lies, LowP > HighP in left MFG, whereas HighP > LowP in left caudate. LowP consistently showed Lie > Truth responses across control regions, indicating that deception is costlier and relies on additional lateral PFC and ACC resources. In contrast, HighP exhibited Truth ≥ Lie overall and preferential striatal engagement during episodic lies, consistent with lower-cost, more automated deception via caudate-mediated gating. Content shapes how traits modulate deception, refining trait-sensitive accounts beyond global deficit views.

A fundamental question in visual cognition is whether object recognition depends on interpolation among stored two-dimensional (2D) views or on compositional encoding within a three-dimensional (3D) representational space. When an object is seen from a new angle, does the visual system interpolate between previously experienced image views, or does it encode the object’s structure in a way that allows systematic integration across rotations (yaw, pitch)? This psychophysical study investigates the geometry of viewpoint generalization using  computer-generated unfamiliar objects rotated in 5° increments. Participants are trained on specific viewpoints of one object and subsequently tested on novel interpolated and extrapolated views of the trained object and an untrained object (distractor) in a two-alternative forced-choice (2AFC) task. Recognition performance is modeled under two competing frameworks: a 2D radial basis function (RBF) interpolation model, predicting locally constrained generalization around trained views, and a 3D compositional model, assuming separable and integrative encoding across rotational dimensions. By estimating viewpoint tuning width (σ) and comparing generalization profiles across axes, the study aims to determine whether object representations in the visual system reflect isotropic 2D similarity or factorized 3D structural encoding. The results are expected to clarify the representational architecture underlying viewpoint invariance and to link psychophysical and neurophysiological models of object recognition.

Two leading theories of consciousness propose distinct necessary neural substrates. The global neuronal workspace theory (GNWT) posits that consciousness depends on the broadcasting of information from the prefrontal cortex, with recent research identifying the opercular inferior frontal gyrus (IFG) as a key structure for conscious awareness. In contrast, the integrated information theory (IIT) implies that consciousness is rooted in a posterior cortical ‘hot zone’, such as the posterior cingulate cortex (PCC). To date, no experimental findings have conclusively favored one theory over the other. Lesion studies provide a means to investigate this question by examining cases of dysfunction in theoretically implicated brain regions.

Previously, we presented a case study of a patient who underwent extensive surgical resection of the prefrontal cortex yet remained conscious and capable of complex thought. To assess regional neural activity, we measured cerebral perfusion using arterial spin labeling (ASL), comparing those values to normative data from healthy controls (n = 16). Expanding upon our previous analysis, we compared bilateral cerebral perfusion in the dorsolateral prefrontal cortex (DLPFC), ventrolateral prefrontal cortex (VLPFC), posterior cingulate cortex (PCC), precuneus, primary somatosensory cortex (S1), primary visual cortex (V1), and frontal & parietal lobes. The patient’s bilateral PCC activity was within normal ranges. The left VLPFC exhibited normal ASL values, while the right VLPFC showed perfusion levels indicative of an absence of neural function. These findings challenge the GNWT’s assertion that intact prefrontal regions are necessary for consciousness. However, preserved function in one hemisphere of the VLPFC may support a more restricted interpretation of GNWT. The intact PCC aligns with the IIT’s prediction that consciousness depends on the integrity of this region. This case contributes to the ongoing debate on the neural correlates of consciousness, highlighting the value of broader lesion-based studies in this field.

Physio-cognitive decline syndrome (PCDS) is an at-risk geriatric condition characterized by concurrent subtle declines in physical and cognitive function, often preceding disability or dementia. Early detection is essential for timely intervention and promotion of healthy aging. However, optimal strategies for applying standard cognitive screening tools such as the MMSE to community-based detection remain unclear. Using data from 673 adults aged ≥65 years, our analysis comprised four phases. First, to determine whether PCDS represents global or domain-level cognitive decline, we compared four models for identifying PCDS: (1) MMSE total score, (2) total score plus frailty, (3) domain-level impairment (>1 SD below the mean in any MMSE domain), and (4) domain-level impairment plus frailty. Domain-level models outperformed total-score models, suggesting that PCDS-related cognitive decline may not be captured by global cognition alone. Second, to identify relevant cognitive subtests, we applied machine learning, and SHAP interpretation highlighted the Calculation subtest as the most influential predictor. Third, to evaluate whether poorer Calculation performance corresponds to higher PCDS probability, we tested multiple cutoffs (scores 1–4, out of 5) and found that a threshold of ≤4 achieved the highest accuracy, indicating that even a single error increased the likelihood of PCDS. Finally, to quantify this relationship among frail older adults, we applied a causal inference-based model, which showed that the association between Calculation errors and PCDS probability was significantly influenced by age and education. Our findings highlight the role of working memory and mental arithmetic, as measured by the Calculation subtest, as key cognitive indicators of PCDS. Older adults with physical frailty who make any error on the MMSE Calculation item are at substantially higher risk for PCDS, particularly among those who are older or less educated. These individuals should be prioritized for comprehensive neuropsychological assessments to facilitate early intervention and promote healthy aging.

Anatomical images are often used as a mid-step for normalization of images for group analyses, leading to the exclusion of the data in cases of anatomical images absence. In case of particular diagnosis collecting the anatomical data of required quality is not possible because of movements. It makes research related to such samples drastically challenging. In this study we evaluated a direct pipeline for alignment the ¹⁸F-Fluorodeoxyglucose (FDG-PET) images to a simulated PET standard space template. The template was created with higher values in the grey matter area. This method was compared to typical approach of data alignment with the mid-step of anatomical image. The sample consisted of 40 participants with Alzheimer's disease (Alzheimer’s Disease Neuroimaging Initiative dataset) and 47 healthy controls. Image-template similarity was assessed with cross-correlation and showed that there is no significant difference between the methods. Both direct and mid-step alignment resulted in comparable region-of-interest values. Thus, here we provide effective method of aligning the FDG-PET images directly to the template.

Visual perception allows us to experience our surroundings by relying more on the external contents or stimuli present in the environment, therefore making our experience externally generated. These contents tend to be stable and change gradually across time which may lead to observed similarity in the visual experience across individuals. When considering visual mental imagery, our experience is generated by the internal mental contents and does not require any immediate presence of the related external contents in the environment. Despite this difference, there has been an overlap in the neural mechanisms involved in the two processes [1]. In addition, when focusing on visual mental imagery, individuals show a large variability in their experience - from clear or vivid imagery to no phenomenological experience of this kind. But why this variation occurs and what are its consequences is poorly understood. 

One way to better understand neural processing during perception and imagery across the visual stream and how it may influence our overall visual experience is to investigate the recurring patterns of activity shared across the brain, defined as brain states [2]. Therefore, with this study we utilized a previously established paradigm by Dijkstra [3] (examining the temporal dynamics of processing and patterns of overlapping brain activity during the two processes) and investigated the dynamic organisation of brain states [4] specific to visual perception and imagery. Specifically, we focused on the source level Magnetoencephalography (MEG) data corresponding to the perception and imagery of House/ Face image stimuli and utilised the hidden Markov model approach (HMM) to find the related brain states. Furthermore, we explored how these states may transition between the two forms of visual experience.

Facial identity recognition is a remarkable human cognitive skill that most adults perform rapidly and effortlessly, even across natural variations in pose, expression, and lighting. However, recognizing the same identity in different contexts may be challenging for adults with autism spectrum disorder (ASD) The present study examines within-person face recognition in autistic and neurotypical adults. We also explored the correlations among within-person recognition, detail-oriented facial discrimination, and standardized face perception tests. Each participant received four tasks: The Face Identity Sorting Task (Taiwanese and Pakistani conditions), the Face Discrimination Task, the Cambridge Face Memory Test (CFMT), and the Prosopagnosia Index-20(PI-20).Gender- and age-matched neurotypical adults (Mage = 25.2±4.1) and six individuals with ASD (Mage=25.2±2.9) have completed the study. The neurotypical adults exhibited a significant familiarity effect where they made significantly fewer piles for the familiar Taiwanese stimuli (M=2.5±0.5) than the unfamiliar Pakistani stimuli (M=5.8±2.1), and a more convergent pattern of responses in matrix analysis. Preliminary data with the autistic participants also showed a trend of familiarity effect, whereby they made fewer piles for the Taiwanese (M=6.3±4.3) than for the Pakistani (M=10.8±5.1) stimuli, and with a more divergent pattern in matrix analysis. Additionally, autistic individuals reported a higher PI-20 score (i.e., greater difficulties with faces) and a lower CFMT score (i.e., worse memory for faces) than neurotypical individuals. In general, the ASD participants made more piles and more misidentification errors than neurotypical participants across both sets of stimuli, indicating overall challenges in within-person face recognition, particularly for autistic individuals.

Facial attractiveness judgments reflect both universal perceptual components and culturally shaped experiences. Cross-cultural research has shown higher agreement in attractiveness ratings among within-culture raters than among cross-culture raters, suggesting that perceptual experience with specific face groups modulates attractiveness preferences. To test the mechanisms of this experience-driven modulation, this study aimed to computationally model the process using convolutional neural networks (CNNs). By explicitly controlling the models' "experience" through selective fine-tuning, we investigated how exposure shapes the sensitivity and hierarchical representations underlying attractiveness judgments.

We fine-tuned GoogLeNet models separately with Caucasian or Asian faces from the SCUT-FBP5500 dataset. Care was taken to balance race, gender, and the rating distribution of faces. By controlling models’ “experience” through selective fine-tuning with five-fold cross-validation, we examined how well they predicted attractiveness judgments for unseen within-group and cross-group faces, and how attractiveness was processed across the network hierarchy. Prediction accuracy was evaluated using Pearson’s correlation coefficients between human ratings and CNN predictions. Layer-wise activation decoding was conducted via leave-one-out cross-validated support vector regression, and the strength of attractiveness representation was quantified as the correlation between human ratings and decoding predictions. Analyses using the original GoogLeNet pretrained for object classification served as a control model.

CNN models replicated human perceptual patterns, showing a clear within-group accuracy advantage and a consistent cross-group bias. Crucially, the CNN approach provided mechanistic insights unavailable through behavioral methods alone. Layer-wise decoding revealed the control model already exhibited stronger representations for Caucasian than Asian faces. This difference was amplified in the Caucasian models, while Asian models showed stronger Caucasian representations across most layers before converging at the final two. These findings highlight the value of CNNs for computationally modeling how perceptual experience shapes sensitivity and hierarchical representations, demonstrating how exposure enhances within-group precision while promoting biased evaluations of other-race faces.

Musical mnemonics can enhance memory encoding and retrieval via emotional engagement. According to the arousal and mood hypothesis, music’s positive effect on memory is mediated by changes in arousal or mood and activation in emotional brain regions. However, mixed findings currently exist regarding the effects of emotional valence of sung mnemonics on verbal recall, with some reporting positively-valenced music enhance recall while others found no effect. This study examined how emotional valence (positive, negative, or neutral) of sung mnemonics affects immediate and delayed word recall. Emotional valence was manipulated using consonant (e.g., major third) and dissonant (e.g., tritone) musical intervals, known to evoke pleasant and unpleasant feelings. Participants recalled lists of 5-9 Mandarin bisyllabic words presented either in spoken or sung with positive, neutral (unison pitch), or negative musical mnemonics. Recall was tested both immediately (after 30s) and after a 24-hour delay. Emotional valence ratings of the musical intervals were collected using a 7-point Likert scale. Supplementary measures related to musical emotion processing, including the Barcelona Music Reward Questionnaire and Beck Depression Inventory, were also collected. Results show higher recall for emotionally-valenced sung words than spoken ones, with positive-valenced intervals eliciting higher accuracy than negative-valenced or neutral intervals. The sung advantage of musical mnemonics persisted at the 24-hour delay, with higher recall for positive-valenced sung words. Higher pleasantness ratings of musical intervals correlated with better word recall accuracy, suggesting a positive emotional effect. Furthermore, higher levels of music listening and sensitivity to social reward predicted verbal recall performance, while music-related movement did not. Collectively, these findings support memory benefits of positive emotion in musical mnemonics, with implications for optimizing music-based interventions to mitigate memory decline in aging and clinical populations.  

Background: Visuomotor rotation tasks dissociate explicit strategic aiming from implicit sensorimotor recalibration in error-based motor learning. Prior work demonstrates that stepwise increases in perturbation magnitude enhance implicit aftereffects relative to abrupt or gradual schedules. The reverse trajectory—decrementing perturbations—might introduce explicit strategies in learning but remain unexplored. We hypothesized that incrementing versus decrementing perturbation schedules, despite converging at identical terminal rotation (45°), constitute distinct learning contexts that differentially weight implicit and explicit processes.

Methods: In a between-subjects design, participants completed either an incrementing series (0°→15°→30°→45°→0°) or a decrementing series (0°→75°→60°→45°→0°) of visuomotor rotations. On each trial, participants provided post-movement aim reports to isolate explicit contribution. Implicit adaptation was derived as performance angle minus reported aim. We analyzed two critical blocks: the terminal 45° rotation block and the subsequent 0° washout block. For each block, we quantified performance (mean angle), explicit (mean aim report), and implicit (performance minus aim) components. Aftereffects were measured as signed deviations from baseline during washout.

Preliminary Results: Decrementing schedules yielded higher mean performance angles at the 45° block, whereas incrementing schedules produced larger aftereffects during washout. This dissociation was consistent across all participants in this exploratory sample (n =5).

Conclusion: Learning context biases the explicit-implicit balance in visuomotor adaptation: decrementing schedules prioritize online strategic correction, while incrementing schedules strengthen implicit recalibration and retention. This dissociation predicts that incrementing training should favor delayed retention and transfer, whereas decrementing training should enhance volitional adjustability under instruction—a distinction with direct implications for optimizing motor skill acquisition protocols based on training goals.

Effective reinforcement learning requires balancing exploration of uncertain options with exploitation of known outcomes. In real-world contexts, the same action may yield rewards in some situations and punishments in others, yet how these learning processes influence each other remains unclear. Here, we investigated the neurobiological mechanisms underlying how reward and punishment learning interact to guide adaptive behavior through functional magnetic resonance imaging (fMRI), computational modeling, and pharmacological manipulation across two experiments (N = 68). In the first experiment (N = 30), healthy participants engaged in an instrumental learning task with binary choices. Some options yielded only reward or punishment outcomes, while others were associated with both, allowing direct assessment of their interaction. Results showed that when both reward and punishment learning were concurrently engaged, the performance of reward learning was selectively impaired, while punishment learning remained intact. Computational modeling revealed that this impairment was predicted by increased exploration linked to prior punishment associations in reward-learning choices, accompanied by enhanced blood oxygen level-dependent responses in prefrontal regions related to exploration behavior. In the second experiment (N = 38), we used a double-blind, between-subject design to investigate whether dopaminergic signaling mediates this interference. We found that pharmacological blockade of dopamine D2/3 receptor with amisulpride eliminated both the impairment in reward learning and the accompanying increase in exploration. These findings demonstrate that reward learning can be disrupted by punishment-driven exploration in a dopamine-dependent manner, offering a mechanistic account of how the brain resolves competing value signals and insights into dopamine-related learning disturbances in neuropsychiatric conditions.

This study investigates the relationship between visual perceptual learning and error-based motor learning, using texture discrimination and visuomotor adaptation as operational measures. Previous studies have suggested that perceptual and motor learning may share common underlying mechanisms. The present research aims to clarify these domain-general characteristics and explore their neural bases.

Preliminary findings show that resting-state EEG activity changes following training in both perceptual and motor tasks. In motor learning, beta power decreased in the motor and parietal regions, while visual perceptual learning was associated with increased alpha power in the occipital and parietal areas. These differential changes in resting-state patterns of brain activities may reflect how participants detect and respond to errors during different domains of learning. Furthermore, during the testing phase, different types of error detection processes seem to be associated with learning performance, depending on the strategies participants adopt

It is expected that both types of learning will lead to measurable changes in resting-state activity, particularly in regions involved in sensory integration and motor control. Moreover, individuals who exhibit more efficient error detection and correction are anticipated to show greater cross-domain transfer effects, supporting the idea of shared learning mechanisms across perceptual and motor domains.

Social observation not only modulates placebo analgesia but may also engage empathic mechanisms. Given that individuals typically exhibit greater empathic responses toward in-group than out-group members, group membership contributes to intergroup empathy bias. However, it remains unclear whether, and how, pain-related group membership influences socially induced placebo analgesia via empathic mechanism. The aim of the current study was to investigate the influence of shared prior pain experiences on placebo analgesia. In this study, we manipulated group membership between healthy participants and confederates (i.e., fake participants) by randomly assigning individuals into either an in-group or out-group group (the participant with or without shared prior pain experiences with the confederate, respectively). An observational learning placebo paradigm was employed during functional magnetic resonance imaging scanning, in which the participant was administered the placebo and control creams and they reported their perceived pain intensity after observing the confederate receiving placebo analgesia (i.e., pain reduction in placebo versus control condition). Preliminary data showed that participants in the in-group group reported significantly stronger placebo analgesia compared to those in the out-group group. Crucially, between-group differences in empathic responses predicted the magnitude of placebo analgesia, suggesting a potential role of empathy in modulating socially induced placebo analgesia. These findings offer valuable insights into how social factors, such as group membership, modulate observational learning and empathic processes to shape placebo analgesia.

心流會帶來享受和滿足感，而沉浸在閱讀時的狀態與心流的性質相似，並且事件劃分是閱讀時主要的神經活動之一。因此本研究探討閱讀時沉浸狀態和對該小說喜好程度間的關聯，以及沉浸程度對事件邊界劃分的神經活動影響。研究以武俠小說為實驗文本，行為方面使用閱讀心流量表測量沉浸程度，並分析量表不同面向之間的關聯，神經活動則使用功能性磁振造影的神經影像進行測量。研究結果發現閱讀時的沉浸程度與對該小說的喜好程度有正相關，且在閱讀武俠小說時，比起情感及整體的投入性，關於閱讀理解的面向之間更有相關性，而沉浸程度與事件邊界劃分時的神經活動有負相關。

Rapid behavioral stopping—the ability to cancel a planned or ongoing action—is essential for adaptive behavior and cognitive control. Deficits in stopping are linked to psychiatric conditions such as ADHD, addiction, and obsessive-compulsive disorder, yet the underlying neural mechanisms remain poorly understood. While most studies have focused on cortico-basal ganglia circuits, our recent findings suggest a novel subcortical mechanism involving the inhibition of a specialized subset of basal forebrain (BF) neurons—termed BF bursting neurons—as one of the most reliable neural correlates of rapid stopping.

Here, we investigate the circuit mechanism underlying this inhibition, focusing on glutamatergic (Vglut2) neurons within the BF. Subcortical Vglut2 neurons are broadly associated with encoding negative valence and aversive motivation, suggesting that BF Vglut2 neurons may functionally oppose the reward-predictive signaling of BF bursting neurons. Moreover, BF Vglut2 neurons send dense projections to the lateral habenula (LHb)—a canonical hub for aversive signaling and behavioral suppression—providing a likely pathway for this inhibitory influence. We therefore hypothesize that the BF Vglut2→LHb projection may suppress BF bursting neuron activity to facilitate rapid stopping.

We first established a stop-signal task (SST) for freely moving mice to quantify stopping behavior and stop signal reaction time (SSRT), providing a platform to dissect the role of the BF Vglut2→LHb pathway in inhibitory control using circuit-specific manipulations and in vivo recordings. We found that optogenetic activation of the BF Vglut2→LHb pathway is sufficient to suppress BF bursting neurons and induce rapid behavioral stopping. Conversely, optogenetic inhibition of this pathway impairs stopping performance and shows a trend toward increasing SSRT. Ongoing experiments are characterizing the activity of BF Vglut2 neurons during stopping using electrophysiology and calcium imaging. Together, these studies reveal a previously unrecognized subcortical mechanism for inhibitory control, with implications for understanding the neural basis of impulse control and executive function.

Many real-world experiences follow a long-tailed, Zipfian-like distribution, where a few events are common while many are rare yet potentially consequential. Previous studies show that deep reinforcement learning (RL) agents perform well on frequent cases but often fail to generalize to rare ones. In the human brain, unexpected and goal-relevant events are thought to engage multiple memory and control systems: mesolimbic dopamine enhances hippocampal encoding, the amygdala tags affective salience, and striatal–prefrontal circuits update action values and rules. However, it remains unclear how rarity and importance jointly shape learning and memory, and how these mechanisms might inform artificial agents.

This project investigates these interactions using a spatial navigation paradigm (Zipf’s Gridworld) that systematically manipulates rarity (encounter frequency) and importance (reward magnitude). Human participants navigate multi-room mazes to find target objects drawn from Zipfian frequency distributions. Behavioral metrics such as accuracy, path efficiency, one-shot improvement, and retention assess whether rare but important objects are learned more efficiently and remembered better than rare but unimportant ones. Planned neural analyses of fMRI and sEEG data will examine whether the hippocampus and ventromedial prefrontal cortex jointly encode rarity–importance interactions and whether first-encounter activity predicts later improvement. Behavior will be modeled using an episodic-gate framework, in which the probability of rapid updating depends on surprise and importance. Parallel simulations will test whether implementing this mechanism improves RL agents’ ability to handle Zipfian environments.

By placing humans and artificial agents within a shared experimental framework, the study provides a basis for comparing how each learns under Zipfian conditions. The expected insights will contribute to both cognitive neuroscience (by clarifying how memory systems prioritize rare, important events) and artificial intelligence (by guiding the design of agents that learn from the rare events that matter most).

The acquisition of movement skills through sequential sequence learning has been broadly studied, with the hierarchical control theory widely accepted as a framework explaining how sequential actions are organized. Within this model, the formation of motor chunk is regarded as a critical mechanism for efficient execution. However, the precise psychological benefit and the differential cognitive reliance of these chunk units versus their boundaries during execution remain ambiguous. This study aimed to behaviorally validate the psychological reality of motor chunks by investigating whether the execution of a highly trained sequence unit is differentially susceptible to simultaneous working memory (WM) load.

Participants were explicitly trained on a specific chunk structure using a 4-key Serial Reaction Time Task (SRTT), where stimuli were presented in various colors irrelevant during training. In the subsequent interference testing phase, an embedded 1-back dual-task was introduced, applied at two critical transition points: Between-Chunk boundaries and Within-Chunk positions. We hypothesized that because Between-Chunk transitions are more reliant on central executive resources, they would lead to a higher Reaction Time (RT) cost on the primary task under WM load.

Preliminary results partially supported the overall cognitive demand hypothesis. As expected, Baseline SRTT RT was consistently higher at Between-Chunk boundaries compared to Within-Chunk positions, while the SRTT RT Cost (RT Disruption – RT Control) was indeed higher for the Between-Chunk position, the difference between the two interference positions was not significantly greater than the existing baseline difference. However, the one-back task RT was significantly slower at the Between-Chunk boundaries compared to the Within-Chunk positions. These findings provide potential behavioral evidence that motor chunks are stored and executed as consolidated, insulated units, with only the chunk boundaries requiring significant central executive resources for retrieval and planning.

Humans must determine when to reuse an abstract rule and when to reconstruct it. This study investigates how these opposing demands—generalization and adaptation—emerge in behavior and large-scale neural dynamics (Friston et al., 2014).

Forty-two adults completed an fMRI concept-learning task in which polyomino stimuli varied orthogonally in texture, color, and orientation. Two sequential blocks manipulated latent rule validity through three regimes: Preserve (rule maintained), Expand (rule extended), and Destroy (rule violated). Behavioral accuracy and trial-based learning curves quantified rule acquisition, and a separate neuropsychological battery assessed individual differences in cognitive flexibility.

Behaviorally, learning was rapid and near-ceiling in Preserve and Expand, whereas Destroy elicited slower convergence and a clear reconstruction cost. Neurally, Preserve and Expand preferentially engaged medial temporal and medial prefrontal regions associated with abstract structural representation and schema reuse (Constantinescu et al., 2016; Baram et al., 2021; Mack et al., 2020). In contrast, Destroy recruited lateral prefrontal, cingulo-insular, parietal control networks, and ventral-striatal prediction-error signals, consistent with belief updating and exploratory control states (Schwartenbeck et al., 2016). Direct contrasts (Destroy > Preserve/Expand) highlighted a fronto-cingulo-insular cluster commonly linked to conflict processing and rule revision.

These findings support a dual-route account of cognitive flexibility: efficient generalization relies on an abstract-model system, whereas adaptation engages broader control and error-processing circuitry to rebuild task structure. Ongoing analyses will relate these neural patterns to individual differences captured in the neuropsychological measures.

Creative brain is not only shaped by the experience but also influence its perception. Previous studies already reveal the brain activities related to creative thinking, but the difference between high and low trainings on the brain connectivity still remains largely undefined. Therefore, 40 students with and without art training were collected to compare their brain difference at resting state with EEG coherence method. Current results showed that students with art training exhibit more connections in the delta, theta and alpha bands, especially between frontal and central areas. These differences assume to the processing of creative thinking and could be in related to the executive plan for motor preparation. These findings reveal brain networks that can distinguish the creative behavior at rest state, providing an important basis to understand the effect of art training on the creative thinking.

Confrontation naming is a core language function that declines during aging and in Alzheimer’s disease (AD). Although neural mechanisms of naming impairment have been extensively investigated in post-stroke or post-surgical populations, much less is known about how anomia emerges in normative aging. Given that different languages impose distinct neural demands, it remains uncertain whether naming relies on a universal neural architecture. This study employed a multimodal neuroimaging approach to characterize neural correlates of aging-related anomia in Mandarin-speaking older adults, a population underrepresented in current research.

We examined 602 older adults from the prospective community-based Taiwan Precision Medicine Initiative on Cognitive Impairment and Dementia (TPMIC) cohort, including 121 with mild cognitive impairment (MCI) and 481 cognitively normal individuals. Seventy-six participants completed longitudinal follow-up (mean = 1.87 years). Naming ability was assessed using the Boston Naming Test. A comprehensive neuropsychological battery, T1-weighted imaging, diffusion tensor imaging (DTI), and resting-state fMRI were analyzed. Neural correlates were identified using multiple linear regression with correction for multiple comparisons. Logistic regression evaluated whether these markers predicted naming decline. 

Naming correlated strongly with attention, executive function, and memory. Anomia was associated with marital status, alcohol use, diabetes, and hypertension. After adjusting for confounders, poorer naming performance was linked to atrophy in the bilateral hippocampus, entorhinal cortex, insula, and left amygdala, as well as reduced white matter integrity in the right uncinate fasciculus and forceps minor. Functional connectivity across occipital, medial temporal, and frontal regions appeared to support lexical retrieval. These neural markers predicted future naming decline, with significant interactions observed with MCI status.

In conclusion, this study identifies structural and functional connectivity markers of anomia preceding dementia onset. Right-lateralized DTI patterns may signal Mandarin-specific neurodegeneration. A distributed visual–perisylvian–frontal network emerges as essential for naming resilience, providing potential biomarkers for AD and a foundation for next-generation AI-based language assessment tools.

The Complex Systems Response (CSR) function represents a second-order nonlinear model that characterizes emergent properties through interaction terms among constituent elements. This approach has demonstrated remarkable success in clinical applications, particularly in predicting individualized dosage responses across cancer therapy, organ transplantation, and HIV treatment, where it achieved significant therapeutic improvements with minimal calibration data. Motivated by these promising results in biological complex systems, this study evaluated whether CSR could similarly capture cognitive phenomena by examining individualized nonlinear models of Chinese character naming compared to traditional linear approaches. 

We fitted CSR models (45 parameters) and GLMs (9 parameters) to naming reaction times from 140 participants (~161 trials each) across eight psycholinguistic variables (LogCF, NS, CON, PC, SC, SAR, IMG, AoA). At the group level, while CSR yielded higher unadjusted R² (0.35 vs. GLM: 0.13), adjusted R² converged (CSR: 0.09; GLM: 0.08), indicating no explanatory advantage when accounting for parameter count. At the individual level, fitting CSR models with 45 parameters to exactly 45 data points through selective sampling inherently produces overfitting, evidenced by resampling analysis (10,000 iterations) showing severe coefficient instability, likely stemming from insufficient data-to-parameter ratios. Although interaction terms in CSR models collectively contributed substantially, this likely reflects their numerical abundance  rather than genuine interaction effects, with linear terms remaining more critical for model performance. 

These results suggest that selective sampling strategies successful in physical and biological systems may not generalize to noisier behavioral data, where traditional parsimonious models provide more stable and reliable explanations. Cognitive and linguistic data inherently involve substantially higher variability due to individual differences, attentional fluctuations, and neural noise that dominates variance in behavioral measurements. Under these conditions, the increased parameterization of CSR does not translate to improved predictive power but rather amplifies overfitting risks.

在各個領域中，如何保持持續的注意力集中以維持安全警覺、行為表現等是極爲重要的議題。過去研究中，已有相當數量的注意力訓練方法，尤以神經回饋訓練於近年來，在神經科學領域中愈發熱門。神經回饋訓練是一種透過操作制約方法：讓參與者接收到基於實時監測腦部的活動狀態，再藉由重複執行來改變大腦自我控制的訓練，進而達到對認知的改變。因此相較於其他常見的注意力訓練（如行爲訓練、正念及冥想等），神經回饋具有直接調控大腦活動、實時回饋促進個體調節學習、可量化效果與動態監測等優勢；除此之外，功能性磁振造影之高空間解析度、全腦覆蓋能力及強定量性，在進行神經回饋訓練時，亦能提供更精準之監測與回饋。

本研究之實驗設計以 deBettencourt 等人於 Nature Neuroscience（2015）發表之關鍵研究〈Closed-loop training of attention with real-time brain imaging〉為基礎，將32名參與者隨機平均分派之實驗組及控制組，進行三個階段的實驗，包含前測（兩個回合，共16分鐘之視覺go-no go作業)、神經回饋訓練（以兩小時為限）及後側（與前測一致）。實驗組在神經回饋訓練階段，係根據該參與者之注意力網絡腦區，經由多體素模態分析（正則化邏輯回歸），以該分析結果實時改變參與者所接受之視覺刺激；控制組則接受到虛假訊號（源自與其配對的實驗組之訊號,sham-paired)。

然而，考量實驗設備之硬體條件之侷限及前導研究的結果，對原有的神經回饋訓練技術、流程進行重新調整與設計，包含：技術上，為避免抵觸硬體瓶頸，將原始訓練中，使用全腦體素（n≊58000)作爲回饋訊號來源，替換為其原始研究結果提及、在該作業中與全腦訊號最接近之感興趣區域（注意力網絡，n≊927），以降低實時分析、訓練時之延遲；架構上，則參考前導研究參與者(n=8)的反饋，在前測階段新增調整刺激及練習回合，以決定個人化的最佳刺激對比參數，並協助參與者熟悉流程。

就現階段分析結果（n=10)：在神經回饋層面上，實驗組的回饋訊號之正確率為.599，高於隨機分類(.5)，且第一個訓練回合(.557)低於訓練之最後一個回合(.605）；在行為上，以敏感度（A’)作為指標，實驗組的後測相較於前測，有略微增加(.867,.88)但並無顯著差異(p=.298)。

When learning the name of objects, people need to form association between object images with corresponding words. Previous research on alphabetic readers has shown that English monolinguals represented onsets of English words most strongly, while Hebrew bilinguals only showed a relatively similar pattern on English but not on Hebrew words. These findings are interpreted as reflecting the effects of reading experience on how individuals represent novel orthographic words starting in the earliest stage of learning. To examine whether the same effect can be observed in a logographic writing system, we employed Chinese pseudo-characters in an orthographic word-referent mapping task. Specifically, real semantic and phonetic radicals in their legitimate spatial positions within characters were paired to form meaningless pseudo-characters that represent names of novel objects. The results showed that Chinese readers were able to learn the association between the novel objects and the pseudo-characters significantly above chance. When they did make mistakes, they tended to mistaken competing foils that shared the same semantic radical as the targets most in both the learning and testing phases of the word-referent mapping task. The differential reliance on semantic and phonetic radicals to represent object names in novel orthographic words reflected participants’ sensitivity to the distinct functions of these two types of radicals. The present findings from the logographic Chinese writing system provide converging evidence with that from alphabetic languages to highlight the influence from native orthography on learning and memory for novel orthographic words.

Substance use disorder (SUD) is associated with altered reward processing, heightened impulsivity, and learning inflexibility, all of which reinforce addiction persistence. Aberrant value computation, impulse control, and learning behaviors are prevalent across neuropsychiatric disorders, yet their underlying mechanisms remain unclear. Given that traditional clinical assessments are limited by subjectivity, potential self-deception, and difficulties in quantification and tracking, this study employs interactive computerized behavioral tasks for a relatively objective and data-driven evaluation.

Three tasks estimated distinct decision-making and learning metrics in economic contexts: a gambling task for risk attitudes, an intertemporal choice task for delay discounting, and a reversal-learning task for adaptive learning. Seventeen healthy controls and three individuals with alcohol use disorder (AUD) were recruited. Behavioral responses were analyzed using cognitive modeling to extract parameters reflecting sensitivity to uncertainty, impulsivity, and adaptive flexibility. Group comparisons and correlations between different task-derived measures were examined.

Preliminary gambling task results revealed that reference-point manipulation affected risk attitudes in healthy controls, suggesting sensitivity to stochastic reference points. Both groups shifted toward risk aversion as the expected value increased, but AUD patients showed stronger aversion. In the intertemporal choice task, AUD patients exhibited a steeper decline in subjective reward value (higher discounting rate), indicating elevated impulsivity. In the reversal learning task, healthy controls initially reduced optimal choices after reward probability reversal but adapted over time. AUD patients showed slower and less accurate adaptation, implying deficits in integrating prior experiences and external feedback to update predictions. These findings support using interactive behavioral tasks and cognitive modeling to assess neurocognitive dysfunctions in SUD, better complementing clinical interviews for diagnosis. Results highlight AUD patients’ altered risk attitudes, impulsivity, and impaired adaptive learning, underscoring the need for personalized assessments and interventions. A large-scale study is needed to validate these methods for transdiagnostic applications and early detection of addictive tendencies.

Successful behavior often requires acting at the right moment, yet “timing” is an invisible variable, making it difficult to determine how animals learn to use elapsed time as a reward-predictive cue. Although previous studies have described timing signals in motor and striatal circuits, the behavioral dynamics through which animals acquire temporal knowledge remain poorly understood. We developed an operant task for Long-Evans rats that isolates time as an abstract cue for reinforcement learning. After entering a fixation port, rats encountered one of three trial types with equal probability: (1) a sound instructing reward at the left port following a fast response without timing constraints (reaction-time trial), (2) a second sound instructing reward at the right port only when licking occurred within a specific temporal interval (timing trial), and (3) a “catch” trial in which no stimulus or reward was presented.

When the rewarded interval in timing trials was lengthened or shortened, more than half of the rats (14 of 23) flexibly shifted their response latencies, while behavior in reaction-time trials remained fast and unchanged. This learning was accompanied by two additional behavioral signatures: (1) the emergence of licking in catch trials toward the right reward port—with timing closely matching that in timing trials—indicating an internally generated expectation based on elapsed time; and (2) progressive withholding of the initial response in the fixation port to delay movement onset, reflecting emerging control of elapsed time.

These findings position this paradigm as a powerful tool to study how animals learn the concept of timing and whether elapsed time becomes a learned predictor of  reward. Ongoing experiments measure reward-prediction error signals in basal forebrain bursting neurons and midbrain dopamine neurons to test whether, in animals that successfully learn to time, the internal interval becomes a measurable neural reward-prediction signal.

Studies of the basal forebrain (BF) have traditionally focused on its cholinergic neurons, yet recent work shows that BF GABAergic neurons also play an important role in promoting reward-seeking behavior. However, BF GABAergic neurons are highly heterogeneous, and it has remained unclear which specific subtypes contribute to reward processing and through what mechanisms. Here, we identify GABAergic neurons expressing natriuretic peptide receptor 3 (Npr3) as a critical BF subpopulation whose transient activation robustly promotes reward-seeking behavior by generating an endogenous reward prediction error (RPE) signal. Using head-fixed mice performing an auditory detection task in which sound cues of varying intensities predicted water reward, we found that a brief (10 ms) optogenetic activation of BF Npr3⁺ neurons at sound onset was sufficient to significantly increase detection probability across intensities and accelerate reaction times. To determine how Npr3⁺ activation improves behavior, we recorded BF neuronal activity during stimulation. Npr3⁺ activation selectively recruited a specialized subset of BF neurons—BF bursting neurons, known to encode RPE signals—and enhanced their stimulus-evoked phasic responses. Despite stimulation being delivered only at sound onset, Npr3⁺ activation also reduced BF bursting neuron responses to reward delivery and increased neuronal inhibition during reward omission, both hallmark features of an enhanced RPE signal. Furthermore, transient optogenetic activation of BF Npr3⁺ neurons in the absence of sound cues also selectively activated BF bursting neurons, with response amplitudes proportional to their responses to reward-predicting sounds, effectively mimicking an endogenous RPE signal. Together, these findings identify BF Npr3⁺ neurons as a key GABAergic subtype that promotes reward-seeking behavior by selectively driving BF bursting neurons and increasing reward expectation. They also reveal a novel mechanism for the generation of endogenous RPE signals in the BF, offering new insight into how reward expectation is computed at the circuit level.

The current fMRI hyperscanning experiment explored the risk preference of individuals when they made individual or group-relevant choices. Compared to isolated decision-making, participants became more risk-seeking when exposed to information about others’ choices and outcomes. Secondly, their choices were more synchronized when group members were in symmetric roles. In addition, they collectively earned fewer rewards under either dictatorship or majority voting compared to under individual decision-making.

Short bowel syndrome (SBS), resulting from extensive intestinal resection, disrupts gut microbiota composition and may influence gut–brain axis (GBA) signaling. This study examined microbiota alterations and GBA-related biomarkers in a rat model of SBS. SBS was induced by small-intestine resection, and controls received sham surgery. Fecal, serum, and brain samples were collected 21 days after surgery. Gut microbiota composition was analyzed using 16S rRNA sequencing, with alpha and beta diversity assessed, and differentially abundant taxa identified by edgeR and LEfSe. SBS produced marked microbial shifts, with increases in Enterococcus, Clostridium, Alistipes, and Odoribacter, and decreases in Romboutsia, Schaedlerella, and Turicibacter. Alpha diversity showed no significant differences, whereas beta diversity revealed distinct group separation (PERMANOVA, P < 0.01). Differential abundance analysis identified Enterococcus, Drancourtella, Citrobacter, and Neglectibacter as enriched in SBS rats (FDR < 0.05). Serum cortisol was higher in SB group (69.95 ± 39.04 vs. 42.84 ± 21.79 ng/mL, P = 0.17), while serum serotonin remained unchanged. Brain serotonin was significantly lower in SB group (1.22 ± 0.24 vs. 1.81 ± 0.38 ng/g, P < 0.01), whereas brain GABA showed no significant difference. These results indicate that SBS induces microbiota dysbiosis and alters GBA-related biomarkers, suggesting potential disruption of gut–brain communication.

The hippocampus supports memories by transforming information about tasks and environments, resulting in state representations. Current understanding often relies on qualitative comparison of representations of neuronal outputs and external variables. We employed high-density electrophysiological probe Neuropixels 2.0 to simultaneously record cortico-hippocampal regions from upstream sensory to downstream memory-constructing tri-synaptic circuits, a pipeline designed for mice performing a spatial, two-alternative-choice virtual-reality task involving memory-guided pattern separation.

To identify information shared by these sensory and serially connected hippocampal regions, we first applied classic Canonical Correlation Analysis (CCA). By linearly aligning neural manifolds across regions, we visualized the communication subspace and the structure of shared information across the circuit. Spatial topology emerged as the dominant shared component between the sensory and the hippocampus, whereas communication between hippocampal subregions was mainly organized around discrete task zones.

Shared information, identified from maximally correlated dimensions, does not specify how information transforms through neural circuits as input-output processing because the analysis is agnostic to directionality as well as millisecond time-scale synaptic computation—the neural structure found could arise from mere co-alignment to external variables. To address this, we investigated the same problem after inferring putative synaptic connectivity using GLMCC (GLM Cross-Correlation), a method that isolates fast neuronal coupling by absorbing slower network fluctuations. Within the hippocampus, we found that task-related information potentially transferred through neuronal pairs with significant functional connections, while other aspects of the task (e.g., movement or time) might not. As a contrast, sensory cortex and hippocampus shared very clear joint representation of spatial topology and task information (like trial-type), yet few connected pairs were found. Therefore, some similarities between representations are more likely a consequence of synaptic-timescale transformation from hippocampal state machines, but others more indirect, non-local and multi-circuit computations.

This novel framework provides empirical constraints for a massive space of cortico-hippocampal computational models.

The internal models of the world are believed to be constructed in the hippocampus. CA3 maintains memory-based spatial representations, while CA1 generates a predictive map by integrating inputs about space and context. The entorhinal cortex (EC) provides relatively strong excitatory drive to distal CA1 dendrites, conveying path-integration information, whereas the retrosplenial cortex (RSC), which might detect mismatches between perception and memory, targets the same compartments. How EC and RSC together shape CA1 predictions, particularly during unreliable sensory inputs, remains unknown.

We first recorded CA1 pyramidal cells in mice navigating a linear virtual-reality (VR) track. Place cells lost most firing and subthreshold synaptic input when visual cues were unexpectedly removed, suggesting that CA1 prediction relies on the convergence of visual and other information.

We next investigated how RSC and EC inputs are anatomically organized within CA1 using one-step transsynaptic viral tracing. It showed that each pathway targeted largely non-overlapping interneuron and pyramidal-cell populations, forming partially segregated excitatory-inhibitory modules. RSC-recipient interneurons were enriched at the border between the proximal and distal apical dendrites, which were potentially perforant path- (PP-) or Schaffer collateral- (SC-) associated interneurons, whereas approximately half of EC-recipient interneurons were neurogliaform cells (NGFCs) that offer spatiotemporally broader inhibition to distal tuft dendrites. These anatomical organizations suggest that distinct cortical pathways could differentially contribute to CA1 processing.

Guided by these functional-circuital priors of RSC and EC, we performed electrophysiology-validated inhibition of RSC, EC, or CA1 in mice running in VR with ultrapotent chemogenetics (PSAM). Dorsal CA1 inactivation broadly disrupted anticipatory behaviors shown toward the upcoming reward location. RSC inactivation increased variability of reward-anticipating licks selectively during unexpected loss of visual cues, while EC inactivation increased variability across all cue conditions.

The functional and microcircuit constraints identified in this study will help revamp current ideas of the computational processes underlying place cells.

Our brain makes decisions for moving around and acting in space; on the other hand, our body moves through space and helps the brain build a mental map for the environment, an important process for memory and cognition. However, how self-motions are specifically used is hard to study due to the lack of tools selectively impairing their functions. We set up an experiment in which mice swam in a water maze deprived of external cues, applying path integration to reach a hidden escape platform, with only one-trial experience through the guidance of an L-shaped corridor. The underlying algorithm was inferred by quantifying seconds-level behavioral dynamics.

We used knockout of acid-sensing ion channels (ASICs), mechanosensing cation channels expressed in peripheral proprioceptors, as a tool to compromise proprioception and directly test its role in navigation. Our study showed that control mice tended to navigate in a more indirect way and try to mimic the pre-trained “route” (the experience ~15 minutes ago), whereas ASIC knockouts navigated in a straighter path. Interestingly, with the dorsal hippocampus inactivated, relatively direct path also emerged, possibly using a direction-reckoning strategy. These observations suggest a functionally converging result in terms of hippocampal and proprioceptive functions, yet it was surprising as hippocampus is conventionally believed to support learning flexible maps. We speculate that this proprioception-dependent motor “repetition”, through hippocampus-dependent habitual learning of sequences, may provide a more reliable solution in highly unreliable environments. Accordingly, in other tasks with reliable environments (reference memory and spatial alternation for Y-maze and figure-8 maze), proprioception appeared totally dispensable. In conclusion, in contrast to the norm of viewing hippocampus as a robust machine for slow map-learning, our one-shot path-integration experiment highlights its pursuit of reliability in a tradeoff of motor-policy and map-computing complexities. We will test this machine-learning-inspired idea computationally based on quantitative data.

Developmental prosopagnosia (DP) is a neurodevelopmental condition characterized by impaired face recognition. The 20-Item Prosopagnosia Index (PI20) is a commonly used self-reporting tool for face recognition difficulties. To provide a linguistically appropriate tool for Mandarin speakers, this study validated the Traditional Chinese 20-Item Prosopagnosia Index (PI20-TW) and developed the Taiwanese Famous Face Memory Test (FFMT-TW). Seventy adults (21 males, Mage = 25.9) participated in the study. They completed PI20 Traditional Chinese and English Versions (PI20-TW, PI20-EN) and four computerized measures: the Cambridge Face Memory Test (CFMT), the Cambridge Car Memory Test (CCMT), the Cambridge Face Perception Test (CFPT), and the newly developed Taiwanese Famous Face Memory Test (FFMT-TW). Results showed that the mean scores for PI20-TW and PI20-EN were similar (M = 57.3 for both), while objective test scores were M = 62.7 for CFMT, 61.8 for CCMT, and 77.3 for FFMT-TW. The results strongly supported the cross-language consistency hypothesis: the PI20-TW total score positively correlated with the PI20-EN score (r = 0.929, p <.001). Validity analyses revealed a weak but significant negative correlation between the self-report measure (PI20-TW) and the objective unfamiliar face memory test (CFMT, r = −0.207, p = 0.047) but not with the familiar face test (FFMT-TW, r = −0.077). Discriminant validity was established, as the PI20-TW-CCMT (Car Memory) correlation (r = −0.059) was not significant. In sum, our preliminary results supported the cross-language consistency hypothesis. To establish the test's definitive psychometric properties, we are conducting additional data collection and refining the FFMT-TW to eliminate age bias.

Temporal lobe epilepsy reflects distributed network disruption, yet metabolic connectivity remains difficult to characterize from short clinical PET acquisitions. We explored a hybrid feasibility approach combining Gaussian-kernel similarity embedding with sparse inverse covariance estimation to derive subject-level metabolic network structure from five 3-minute FDG-PET frames. Dynamic ^18F-FDG PET data were obtained from patients with left (LTLE, n = 16) and right (RTLE, n = 14) temporal lobe epilepsy recruited from the Taipei Veterans General Hospital Epilepsy Center. All datasets underwent motion correction, MRI coregistration, FreeSurfer parcellation (109 ROIs), kernel construction, QUIC precision modeling, and graph-theoretic profiling. Group differences were assessed using 10,000-iteration permutation testing, FDR control, and Cohen’s d. Global topology was highly similar across groups (|d| < 0.25; q > 0.78), with a nonsignificant trend toward higher clustering in RTLE (0.659 vs. 0.640; d = 0.85). In contrast, regional effects suggested complementary reorganization patterns: LTLE showed increased dominant semantic–mnemonic circuitry (hippocampus d = –0.71; parahippocampal d = –0.88) and contralateral limbic–DMN involvement (right amygdala d = –0.67), whereas RTLE strengthened ipsilateral limbic/subcortical pathways (pallidum d = +0.97) alongside dominant-hemisphere language/attention engagement (supramarginal gyrus d = +0.73). These patterns align with hemispheric specialization in cognitive epilepsy: LTLE reorganization may scaffold verbal semantic memory via anterior temporal and contralateral limbic–DMN reinforcement, while RTLE compensatory recruitment of attention-parietal and basal ganglia systems may support visuospatial and attentional processing despite nondominant temporal disruption. These preliminary findings highlight the potential of kernel-assisted precision modeling for capturing cognition-relevant lateralization mechanisms from short PET data. Future work will directly test these compensatory hypotheses using neuropsychological assessment to determine whether network signatures predict cognitive outcome and inform treatment planning.

Prediction error refers to the discrepancy between expected and actual sensory input. Its hierarchical organization has been demonstrated through decomposed brain responses in the local-global paradigm, where temporal regularities are established locally at individual stimulus transitions and globally across multi-tone sequence structures. In macaques and marmosets, local and global prediction-error signals have both been found in high-frequency oscillations (60-150 Hz) recorded with ECoG. In humans, however, these signals have primarily been observed in a lower frequency range (30-100 Hz) using EEG. Here, we recorded human ECoG to achieve higher signal fidelity, enabling examination of neural oscillations above 100 Hz (high-frequency activity, HFA), which are believed to be closely linked to local spiking activity. Eight participants listened to auditory sequences that either followed their local and global regularities (local and global standards) or violated them (local and global deviants). Robust HFA responses were observed for the local deviants, but these responses were reduced when the deviants were expected based on the global regularity, indicating both levels of prediction-error processes contribute to the observed activity. Importantly, these HFA responses could be decomposed into two subcomponents: an early component reflecting local prediction-error signals localized to lateral auditory regions, and a late component reflecting global prediction-error signals prominent in both lateral auditory and frontal cortices. Together, these findings demonstrate that neural oscillations above 100 Hz encode hierarchical prediction errors not only in non-human primates but also in humans.

Mindfulness-based stress reduction (MBSR), the 8-week training program that uses mindfulness meditation practices, shows the capability to improve emotion regulation in the face of aversive events and has become a trend nowadays in both clinical and neuroscience fields. Neuroimaging investigations of MBSR revealed changes of brain functionality before and after the MBSR training, among which most of the literature investigated the network connectivity in the resting state. However, the brain function in the resting state does not reflect the MBSR effect on emotion regulation. Therefore, we recruited 32 participants (14 waitlist control and 18 MBSR group) to investigate the functional connectivity changes on emotion arousal before and after the MBSR training. The result showed that after 8-week training, the MBSR group increased trait mindfulness and presented the connectivity between amygdala and postcentral gyrus under emotion arousal, which implies the association between body-sensation awareness and emotion arousal.

Acoustic features such as word onset, speech envelope, and fundamental frequency in spoken discourse contributes to discourse comprehension, which displays in neural synchronization with the acoustic features (Brown & Bacon, 2010; Goswami, 2018; Goswami, 2022). Yet, second language (L2) learners of different language proficiency levels showed a different brain response pattern when listening to natural speech in their second language (Ihara et al., 2021). Holo-Hilbert Spectra Analysis (HHSA) was used to decode brain oscillations of continuous auditorial stimuli and its frequency modulations (Lee et al., 2022). Thus, this study aims to explore the brain responses of the interactions from spoken discourse features for L2 learners using HHSA. 

A group of native English speakers (n=33) from Alice dataset (Bhattasali et al., 2020), and another group of native Mandarin speakers who learn English as a second language learners (n=26) are included in the present study. With which passively listens to Alice’s Adventures in Wonderland Chapter One audiobook while conducting EEG recording. The preliminary results from Layer 1 Intrinsic Mode Functions (IMFs) reveal that native speakers exhibit significantly more stable and rhythmic carrier oscillations compared to L2 learners. L2 learners displayed higher signal fluctuations, particularly in the 1.2s window following word onset. Furthermore, Holo-Spectra is suggested to uncover whether that native listeners possess a more focused frequency-coupling pattern, whereas L2 learners exhibit scattered power distributions. These findings indicated that L2 speech comprehension is possibly characterized by less efficient neural carrier stability, reflecting the increased cognitive load required for non-native linguistic processing.