口頭發表
Oral Presentation 

 The dynamic interaction between the brain and gastrointestinal system, collectively termed the gut–brain axis, has become an emerging focus in understanding human physiology and mental health. This bidirectional communication network links the central nervous system and enteric nervous system, coordinating neural and visceral processes. In this study, we examined cross-system interactions by quantifying phase–amplitude coupling (PAC) between infra-slow gastric rhythms and cortical alpha oscillations (8–13 Hz) during resting state. Simultaneous EEG and EGG recordings were obtained from 23 healthy participants. To capture the influences of analytic flexibility, we conducted a multiverse analysis encompassing 216 defensible preprocessing pipelines. Consistent PAC patterns were observed, particularly within the left fronto-central region, indicating region-specific gut–brain coupling. Specification curve analysis further highlighted that certain preprocessing parameters, such as FIR filtering and manual ICA, enhanced the stability and magnitude of coupling effects. These findings, validated through permutation-based inference tests and topographical inspection of alpha-band activity, emphasized the methodological sensitivity of gut–brain electrophysiology. Rather than asserting fixed physiological mechanisms, this study illustrated how multiverse frameworks could advance transparency and reproducibility, paving the way for more reliable inference in future gut–brain studies.

Motor imagery (MI), the internal simulation of movement without overt execution, is proposed to engage greater cognitive resources than motor execution (ME). This study investigated how MI interacts with spatial memory processes and how its neural correlates differ from those of ME. Participants performed a spatial orientation memory task involving physical head rotation (ME), head rotation imagery (MI), or passive scene motion (visual perception; VP) during encoding, followed by a pointing retrieval task. Spatial memory performance was quantified as the absolute angular error between the retrieved and target orientations across six rotation magnitudes (±9°, ±10°, ±11°, ±32°, ±33°, ±34°).

Behavioral results showed that MI led to larger retrieval errors than both ME and VP, suggesting that the cognitive resources required for MI interfere with spatial memory construction. This interference was amplified with greater angular deviations, indicating that increasing movement difficulty intensifies the cognitive load. EEG analyses revealed that MI induced stronger parietal alpha event-related desynchronization (ERD) than ME, reflecting heightened spatial attentional demands. Parietal alpha suppression was positively correlated with retrieval error, linking attentional resource allocation to behavioral accuracy. Additionally, both MI and ME elicited enhanced frontal theta power relative to VP, suggesting engagement of executive control and error-monitoring mechanisms.

These findings support a motor–cognitive model in which MI recruits extensive neural resources for attentional and executive functions. The results highlight shared yet distinct neurocognitive processes between imagined and executed actions, providing insights into how internal movement simulation involves spatial memory and executive control.

How do motivational forces evolve as individuals weigh the temptation of effortless engagement against the demand for effortful, goal-directed action? Capturing this dynamic requires a model that treats motivation as a latent, time-varying process rather than a fixed trait. This study introduces a unified state-space framework that reconstructs trial-by-trial motivational states by integrating economic valuation, Bayesian inference, and neural dynamics within a single generative architecture.

Participants performed an 18-trial temporal decision-making task that required allocating limited time between digital-media use (DMU) and inventory completion (INV). Each trial presented a structured switch–continue dilemma—whether to switch from DMU to INV or continue with DMU—eliciting recurrent motivational conflicts across different levels of procrastination tendency. Before each trial, participants rated each option’s desirability and effortfulness on 0–100 % interactive logit-normal scales, while EEG was recorded continuously throughout evaluation, choice, and execution phases.

Methodologically, the model formalizes subjective value and effort as logit-normal random variables adjusted by entropic risk parameters (λ), yielding risk-weighted evaluations. These feed into a ΔV/ΔE softmax utility governed by latent motivational states (αₜ: drift, βₜ: evaluation sensitivity, κ: value–effort tradeoff). A Kalman filter–Polya–Gamma hybrid performs sequential Bayesian inference, recovering identifiable trajectories of αₜ and βₜ with uncertainty quantification. EEG-derived affective states (ψₜ) are modeled as autoregressive processes linked to temporal-difference learning, grounding neural adaptation in predictive coding of motivational change.

Ongoing analyses aim to reveal how latent motivational states correspond to neural signatures of resisting or yielding to low-effort engagement. This hierarchical, multimodal framework offers a principled account of how the balance between task-directed control and effortless diversion unfolds over time.

Natural language unfolds over multiple nested timescales: Words form sentences, sentences form paragraphs, and paragraphs build into full narratives. Correspondingly, the brain exhibits a hierarchy of processing timescales, spanning from lower- to higher-order regions. During narrative comprehension, neural activation patterns have been shown to propagate along this cortical hierarchy with increasing temporal delays (lags). To investigate the mechanisms underlying this lag gradient, we systematically manipulate the structure of a recurrent reservoir network. In the biologically inspired “Limited-Canal” configuration, word embeddings are received by a limited set of sensory neurons and transmitted through a series of local connections to the distal end of the network. This configuration endows the network with an intrinsic lag gradient, inducing a cascade of activity as information propagates along the network. We found that, similar to the human brain, this intrinsic lag gradient is enhanced by naturalistic narratives. The interaction between naturalistic input and network structure becomes evident when manipulating local connectivity through the “canal width” parameter, which determines how closely the Limited-Canal model mirrors the human brain’s sensitivity to narrative structure. In addition, we found that processing cost, as a computational proxy for the BOLD signal, increases more slowly in later neurons, which can account for the emergence of the lag gradient. Our results demonstrate that narrative-driven neural dynamics can emerge from macroscale anatomical topology alone without task-specific training. These fundamental topological properties of the human cortex may have evolved to effectively process the hierarchical structures ubiquitous in the natural environment.

Multiple subcortical neuromodulatory systems are simultaneously engaged during behavior, yet how these systems interact remains poorly understood. Recent evidence indicates that subsets of basal forebrain (BF) neurons encode reward prediction signals similar to those conveyed by dopamine (DA) neurons in the ventral tegmental area (VTA), raising the question of whether these two systems transmit independent signals or coordinate their activity. Here we show that a subgroup of noncholinergic BF neurons, termed BF bursting neurons, exhibits robust temporal and causal coupling with midbrain DA neurons, consistently following DA neuron activity with a ~10-ms delay. Simultaneous recordings from freely moving rats performing reward-based associative learning tasks revealed strong trial-by-trial correlations in response amplitude and spike timing between putative DA neurons and BF bursting neurons. DA neuron firing consistently preceded BF neuron activity, whether during well-trained tasks, novel learning contexts, or baseline periods without explicit behavioral events. Optogenetic activation of VTA DA neurons in DAT-Cre mice reliably evoked responses in BF bursting neurons at latencies matching those observed during behavior, and similar effects were obtained by directly activating DA axon terminals within the BF. Conversely, optogenetic inhibition of DA neurons at either site abolished BF bursting responses to reward-predicting cues and markedly suppressed their baseline firing. These findings establish that VTA DA neuron activity is both necessary and sufficient for reward prediction signaling by BF bursting neurons, revealing a previously unrecognized direct causal coupling between two major neuromodulatory systems, and expanding the traditional view of DA function beyond its established role in modulating striatal circuits.

Curiosity motivates learning and memory, yet how curiosity-based exploration changes with age and relates to locus coeruleus (LC) function remains unclear. We developed the Photographic Art Storytelling Task to characterize age differences in how intrinsic motivation and reward prediction errors guide information-seeking. Participants viewed photographs, rated their curiosity, and later read stories associated with selected images. The stories were deliberately constructed to be either interesting or boring, functioning as rewards that elicited prediction errors. After reading each story, participants reappraised their curiosity, allowing us to quantify how the story outcome updated their curiosity. Pupillary responses indexed anticipatory arousal, and LC structural integrity was measured in a subset of older adults (3T) and young adults (7T) using high-resolution MRI. A total of 133 adults (65 older, 68 young) completed the task. Young adults showed stronger updating based on story content, consistent with reward prediction error mechanisms shaping exploration. Older adults relied more on initial curiosity, suggesting heightened sensitivity to novelty rather than outcome-based learning. Pupillometry revealed distinct arousal dynamics. In young adults, higher curiosity elicited larger anticipatory pupil dilation. In older adults, lower curiosity items produced greater dilation, potentially reflecting disengagement or aversive responding. Individuals with higher LC integrity showed reduced reliance on story outcomes, indicating a role for LC structure in stabilizing novelty-driven curiosity. Curiosity enhanced episodic memory in both age groups, demonstrating preserved motivational benefits for learning across adulthood. These findings identify curiosity as a robust enhancer of memory in aging and highlight LC integrity as a neural substrate supporting curiosity regulation. They suggest that targeting motivational mechanisms linked to LC function may help sustain exploratory behavior and cognitive health in later life.

Recent advances in Generative Adversarial Networks (GANs) have enabled hyperrealistic facial synthesis, producing AI-generated faces that often appear indistinguishable from real ones and are perceived as highly trustworthy. This study investigates how adults from diverse racial backgrounds—Taiwanese, South Asian, and Black—distinguish real and synthetic faces and explores the perceptual cues for detecting facial artificiality and realism. We recruited 32 adults in each of the three racial groups. Each participant completed two computerized tasks and a post-experimental survey. In the Face Authenticity Judgment Task, participants viewed individual faces and judged each as real or synthetic. Across all three groups, performance was near chance (Taiwanese: M=.515, South Asian: M=.513, and Black: M=.530), suggesting limited spontaneous sensitivity to facial artificiality. In the Simultaneous Presentation Task, participants viewed a pair of real and synthetic faces (matched in race, gender, and overall age) and chose the real face. Without any explicit feedback, the group mean accuracy significantly improved for all three groups—Taiwanese (M=.577), South Asian (M=.576), and Black (M=.622). Individual analysis indicated that most participants fell into the response quadrants of improved accuracy through paired presentation. South Asians exhibited a strong own-race advantage with selectively higher accuracy for the South Asian faces. Taiwanese participants showed comparable accuracy for East Asian and South Asian faces, suggesting a near-race effect. Black participants demonstrated an own-race advantage with higher accuracy for Black faces. Post-experiment surveys revealed perceptual cues across groups: overly smooth or flawless skin and excessive symmetry were seen as artificial, while visible skin texture, natural shadows, and clear eye details contributed to realism. Together, these findings demonstrate that simultaneous presentation improves overall accuracy and that racial familiarity shapes perceptual sensitivity to AI-generated faces. Future work will extend this research to explore the neural mechanisms underlying facial authenticity judgement.

Facial expression recognition bias may reflect a vulnerability to affective disorders such as depression. However, the effects of exposure duration and the underlying neurophysiological mechanisms in nonclinical populations remained unclear. This study aimed to develop an integrative neural framework using EEG to identify markers of emotion perception bias in individuals with depressive tendencies. Forty-seven participants were recruited from National Central University, including 22 with depressive tendencies, and 25 neurotypical individuals, based on completed questionnaires of the State-Trait Anxiety Inventory and the Beck Depression Inventory. Participants performed two facial emotion recognition tasks with shorter and longer exposure durations on separate days while EEG was recorded. In each task, faces were morphed from happy to angry or sad across 11 intensity levels. A within-subject design ensured that all participants completed both conditions in a randomized sequence. Psychometric analysis yielded the point of subjective equality (PSE) and discrimination sensitivity (DS). Hilbert-Huang transform analysis was applied to task EEG. Preliminary results showed that individuals with depressive tendencies exhibited a negative bias under the longer exposure duration, evident in both angry–happy and sad–happy conditions, particularly for faces with less than 50% negative expression. Additionally, EEG results showed that alpha and beta oscillations enhanced in depressive tendency individuals compared to neurotypical individuals. Furthermore, posterior theta oscillations increased in individuals with depressive tendencies, particularly for morphed faces containing less than 50% negative expression. These findings may indicate that longer exposure durations allow depressive individuals to engage in more extensive processing of ambiguous faces. The enhanced posterior theta activity may reflect a memory-driven mechanism through which ambiguous expressions were more likely interpreted as negative. Future studies using connectivity analyses and brain stimulation are needed to clarify EEG markers of emotion bias.

Bodily posture is a fundamental expression of how the brain represents and organizes the body in space. Maintaining and reorganizing complex postures engages not only sensorimotor systems but also the cognitive machinery that supports body schema in space, and the action planning. Postural organization thus embodies the interface between perception, action, and cognition—linking how the body feels, moves, and is perceived by others.

To probe the cortical network for posture coding, we passively and simultaneously displaced the two forearms of two macaques during fMRI, thus defining four conditions: staying uncrossed (UU), staying crossed (CC), moving from uncrossed to crossed (UC), or vice versa (CU). Across all pairwise contrasts, cortical activations were determined by final posture rather than movement history. In other words, the static posture contrasts CC vs. UU (resp. UU vs. CC) captured the activations of most *C vs. *U (resp. *U vs. *C) contrasts. These two contrasts shared a core posture network encompassing S1, motor, premotor, insula and STP. The crossed condition (CC vs. UU) additionally activated occipito-parietal visual areas, the visual motion complex, the object-directed action network (LIP/VIP/AIP/area 7), as well as S2, Tpt, and cingulate areas 23/24. In contrast, the uncrossed condition (UU vs. CC) activated occipito-temporal areas (TEO/TE). When analyzing the dynamic contrasts in more detail, strongest activations were observed for uncrossing (CU vs. UU) or maintaining crossed configuration, while weakest activations were observed for dynamic crossing (UC vs. CC).

In sum, postural coding engages a sensory-integration network (parietal, temporal, insular cortices) that synthesizes multisensory body- and world-centered signals, and a motor-executive network (premotor, motor, cingulate cortices) that generates and scales compensatory actions. On top of these, crossed posture activates the dorsal stream, planning “how” to execute an action, while resting uncrossed posture triggers the ventral stream, possibly searching for object-directed actions.

Human feelings toward visual stimuli are shaped by how our perception represents what we see. Yet it remains unclear which aspects of perceptual representations contribute to sensory valuation. In this study, we tested the hypothesis that universality of visual representation, namely, the degree to which different perceptual systems converged on similar internal representations of the same image, can predict sensory valuation. We selected nearly 9,000 complex, naturalistic images drawn from two large public datasets that include human ratings of beauty and enjoyment for analysis. We used seven pretrained visual neural networks, each with different architectures, training objectives, and datasets, as computational analogues of diverse perceptual systems. For each image, we extracted its representation from every model and compared them to quantify how consistently it was encoded across models in terms of representational geometry and axis. This inter-model agreement served as an index of representational universality. Using matrix decomposition and inter-model representational similarity analyses, we found that images with higher representational universality tended to receive higher pleasure ratings from human observers. This effect persisted after controlling for visual complexity and was stronger than correlations with various image statistics. To test whether this effect was specific to purely visual processing, we repeated the analysis using text captions of the same images processed through seven language models with distinct constructions. After variance partitioning analysis, a similar relationship emerged, suggesting that universality in both visual and linguistic representations separately contributes to the sensory pleasure of images. Overall, these findings suggest that what humans find pleasurable to look at may be linked to how consistently different cognitive systems encode the same visual content, implying that sensory valuation reflects an optimization of internal representation. This study also provides a novel computational approach for using artificial neural networks to uncover the computational principles of human cognition.