第 1 天 - 2025年06月13日

全天 到会

第 2 天 - 2025年06月14日

北京09:00
(巴黎3:00, 纽约21:00)		 开幕式
北京9:00
(巴黎3:00, 纽约21:00)
Semantic Processing and Decoding Using Intracranial Recordings in Humans
Natural conversation relies on our ability to produce and understand language, a complex process that involves both assembling and deciphering streams of words. Despite significant advances, little is known about how neuronal activity responds to individual word meanings, nor do we know whether it is feasible to decode these meanings from neural activity. To explore this, we utilized the rare opportunity of single-neuron and intracranial sEEG recordings in humans. We examined how single-neuron and local field potential (LFP) coding support the processing of word meanings during language comprehension and production. Our findings reveal that single neurons display selective responses to specific semantic domains, enabling us to decode word meanings with high accuracy from both single-neuron activity and LFP. Additionally, we identified a distributed left-lateralized brain network that supports semantic planning, which includes the precentral gyrus, pars triangularis, and middle temporal cortex. Together, we demonstrated the feasibility of extracting the meaning of words during speech and suggested the potential for decoding semantic content from unconstrained, natural speech.
蔡静
Emotion regulation and affective psychopathology: Capturing context and flexibility
In this talk, I will first briefly present evidence on how situational and cultural contexts shape the effectiveness of emotion regulation (ER). This work underscores the importance of considering ER flexibility instead of discrete strategies. To illustrate this, I then present evidence from ecological momentary assessment (EMA) and laboratory psychophysiology studies demonstrating the affective benefits of flexibility across different stages of emotion regulation—evaluation, implementation, and monitoring—in both healthy and clinical samples and across cultures. Together, these findings underscore the need for a more comprehensive examination of person-context interactions in psychopathology and highlight flexibility as a key target for developing personalized, context-sensitive interventions. I will end the talk with a call to prioritize the study of context in affect and psychopathology research.
陈树铨
Representation of visual uniformity in the lateral prefrontal cortex
Visual illusions tend to have early visual cortical correlates. However, this general trend may not apply to our subjective impression of a detailed and uniform visual world, which may be considered illusory given the paucity of peripheral processing. Using a psychophysically calibrated visual illusion, we assessed the patterns of hemodynamic activity in the human brain that distinguished between the illusory percept of uniformity in the periphery (i.e., Gabor patches having identical orientations) from the accurate perception of incoherence. We identified voxel patterns in the lateral prefrontal cortex that predicted perceived uniformity, which could also generalize to scene uniformity in naturalistic movies. Because similar representations of visual uniformity can also be found in the intermediate and late layers of a feedforward convolutional neural network, the perception of uniformity may involve high-level coding of abstract properties of the entire scene as a whole, that is distinct from the filling-in of specific details in early visual areas.
葛一君
青少年睡眠模式的可穿戴监测及神经表征分析
睡眠在维持生命活动、调节生理功能以及促进脑发育和认知能力方面发挥着关键作用。青春期是睡眠模式发生显著变化的阶段，常表现为入睡延迟、睡眠时间缩短以及昼夜节律的重新调整，这些变化通常与神经系统的持续成熟过程相伴。由于这些神经系统对认知发展和成年功能准备至关重要，深入理解睡眠与脑发育之间的关系，尤其是对认知能力的影响，具有重要意义。尽管成人群体中关于睡眠的神经机制已有较多研究，青春期，尤其是青春期早期个体的睡眠神经表征仍缺乏系统性探讨。本报告基于3000余名11至12岁青少年，通过腕戴设备获取的客观睡眠数据，结合机器学习方法，识别与不同睡眠模式相关的脑结构和功能连接特征，并进一步探讨这些模式与认知能力及学业表现之间的关系。
马庆
Unraveling Human Cognition During Sleep: From Active Stimulation to Behavioral Response
Sleep is more than just rest—it is a dynamic state essential for memory consolidation, emotional regulation, and cognitive adaptation. My research explores how active stimulation during sleep can be used not only to study these processes but also to modify them, offering new possibilities for improving mental health. In this talk, I present a series of studies demonstrating how techniques like targeted memory reactivation (TMR) and sleep learning can reshape aversive memories, reduce emotional distress, and even influence decision-making across different states of consciousness, including wakefulness, light sleep, and lucid dreaming. These findings highlight sleep’s untapped potential for transforming how we approach memory, emotion, and mental health interventions—from treating PTSD to enhancing therapeutic outcomes. This research illustrates how sleep science can open new doors for understanding and improving the human mind.
夏涛
Neural Repurposing from Proto-Word Areas in Macaques to the Human Visual Word Form Area
The Visual Word Form Area (VWFA) is believed to develop by repurposing a pre-existing area through literacy, but which specific area is repurposed and why it is chosen remain unclear. Given the presence of similar category-selective regions (e.g., face, body, scene) in both macaques and humans, could the human VWFA develop from a proto-word area that can also be identified in macaques?
Using fMRI, we identified word-selective regions spanning from the posterior to anterior inferotemporal (IT) cortex in word-naïve macaques. Widefield calcium imaging and high-density electrophysiological recordings confirmed a high concentration of word-selective neurons in these regions through measuring responses to thousands of words and non-word objects. Additionally, objects similar to words in the object space elicited stronger activity, suggesting that proto-word areas in primates may develop through exposure to such objects. This idea is further supported by simulations using deep neural networks.
To examine the homology between the macaque word patch and the human VWFA, we conducted human fMRI experiments, showing that the same object space model could explain human VWFA responses.
While both species' word-selective areas follow the object space model, notable differences persist. By measuring responses to nearby objects, faces, and words in human adults and macaques using fMRI, we observed that the macaque word area favors nearby objects over words, whereas the human VWFA shows the opposite preference. Furthermore, responses from preschool and primary school children revealed a shift from a preference for nearby objects to a preference for words as reading experience increased. This study highlights how the human brain repurposes a homologous word-selective area, identified in macaques, to specifically represent words through literacy.
杨佳
北京13:30
(巴黎7:30, 纽约1:30)		 Keynote
Unveiling the Dynamics of Visual Learning: New Approaches in Modeling and Analysis
Visual perceptual learning (VPL) can lead to dramatic performance improvements with training and has broad applications in enhancing visual and cognitive functions. However, VPL is not a unitary process—learning involves multiple components, including general improvement, session-to-session consolidation, rapid relearning, and within-session adaptation or fatigue. Traditional analyses often overlook these dynamics due to coarse temporal resolution.
This talk will present findings from a large-scale, multi-task VPL study revealing individual differences in general learning ability, session-level dynamics, and task interference. I will introduce novel data-analytic techniques, including hierarchical Bayesian models, that allow fine-grained, trial-by-trial analysis and prediction of learning trajectories. These approaches enhance our ability to model, forecast, and optimize perceptual learning.
I’ll conclude with a look ahead: integrating component processes into generative models, examining classic VPL effects through this lens, and building frameworks to enhance learning and generalization across domains.
Zhong-Lin Lv 吕忠林
北京14:30
(巴黎8:30, 纽约2:30)
Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex
Understanding functional representations within higher visual cortex is a fundamental question in computational neuroscience. While artificial neural networks pretrained on large-scale datasets exhibit striking representational alignment with human neural responses, learning image-computable models of visual cortex relies on individual-level, large-scale fMRI datasets. The necessity for expensive, time-intensive, and often impractical data acquisition limits the generalizability of encoders to new subjects and stimuli. BraInCoRL uses in-context learning to predict voxelwise neural responses from few-shot examples without any additional finetuning for novel subjects and stimuli. We leverage a transformer architecture that can flexibly condition on a variable number of in-context image stimuli, learning an inductive bias over multiple subjects. During training, we explicitly optimize the model for in-context learning. By jointly conditioning on image features and voxel activations, our model learns to directly generate better performing voxelwise models of higher visual cortex. We demonstrate that BraInCoRL consistently outperforms existing voxelwise encoder designs in a low-data regime when evaluated on entirely novel images, while also exhibiting strong test-time scaling behavior. The model also generalizes to an entirely new visual fMRI dataset, which uses different subjects and fMRI data acquisition parameters. Further, BraInCoRL facilitates better interpretability of neural signals in higher visual cortex by attending to semantically relevant stimuli. Finally, we show that our framework enables interpretable mappings from natural language queries to voxel selectivity.
Andrew Luo
How to Study “Thought”? Toward Understanding the Brain’s Algorithm for General Intelligence
How does the brain implement the computational principles that enable general intelligence? Despite remarkable advances in neuroscience and artificial intelligence, the mechanisms that govern thought, reasoning, and decision-making remain elusive. In this talk, I will outline my research to discover the brain’s algorithm for general intelligence by integrating principles from computational neuroscience, cognitive science, and artificial intelligence.
杨千里
How the Brain Forms New Word Concepts: A Neurocomputational Integration of Priors and Learning
Human word-concept learning transcends simple associations, leveraging prior knowledge to generalize from minimal examples. While Bayesian models explain such behavior, their neural underpinnings remain unclear. Our study introduces a neural Bayesian model (NBM) to elucidate how prior semantic representations guide new word learning. Using fMRI, we measured neural activity as participants learned novel words paired with familiar objects (rich priors) or novel shapes (weak priors). The NBM, which integrates neural priors derived from ventral occipitotemporal cortex (VOTC) activity, predicted learned concept neural representations and generalization behavior for familiar objects, outperforming control models lacking such neural priors. Conversely, hippocampal activity, not explained by NBM, underpinned learning with novel shapes, reflecting simple associative mechanisms. Comparisons with large language models (LLMs) revealed LLMs’ inferior alignment with human generalization, underscoring gaps in grounding word learning in nonverbal priors. These findings dissociate neural computational systems for concept learning: VOTC mediates prior-based Bayesian inference, while hippocampus supports novel associations. The results bridge computational theories of word learning with neural mechanisms, highlighting the dynamic interplay of semantic and episodic memory, and further promoting the incorporation of Bayesian-based learning mechanisms for large language model development.
张光耀
Exploration, Combination, Communication: A New Paradigm for Studying Conceptual Innovation
Many new things, if not all, come from combining existing elements in novel ways: mixing flour and water makes a dough, and feeding large dataset to neural networks can create powerful AI agents. Behind such creative synthesis lies a core set of cognitive mechanisms: forming hypotheses about how elements can be recombined, testing those hypotheses, and sharing discoveries with others. These abilities help us navigate an open-ended space of possibilities, even given limited cognitive resources. In this talk, I will present an interactive task we recently developed: a 2D discovery game where participants can freely move, pick up, drop, and combine items, harvesting points by consuming what they create. This flexible environment integrates multiple cognitive functions into a single framework—from reward maximization and rule induction to exploration-exploitation tradeoffs—and allows full control over the game parameters. As a starting point, we compared several existing learning strategies with online participants and found supporting evidence that people rely on structured mental representations to explore the game world.
赵博囡

第 3 天 - 2025年06月15日

全天 离会