论文 | 罗骁齐

2027

AAAI-27
NeuralIntent: Bridging Noninvasive EEG and Embodied Foundation Models with Continuous Neural Intent Tokens

Xiaoqi Luo, Weixiang Gao, Zhengnan Zhang, and 4 more authors

In The Forty-First AAAI Conference on Artificial Intelligence (Under Review). First author · AAAI-27 submission · Under Review · Submission 37074. Framework: NeuralIntent; benchmark: EIG-Bench. , 2027

Abs Bib HTML PDF

Individuals with severe motor or communication impairments may retain clear object goals but lack a reliable way to convey them to assistive robots. We present NeuralIntent, a framework that maps noninvasive electroencephalography (EEG) into continuous Neural Intent Tokens (NITs) and calibrates them to condition an embodied foundation model. NeuralIntent maps the EEG-derived representation into the policy condition space through an embodiment-specific adapter. The EEG-derived representation specifies the target object, while the embodied policy generates the actions needed to interact with it. We introduce the Embodied Intent Grounding Benchmark (EIG-Bench). On THINGS-EEG2, Qwen3-VL Layer 6 supervision improves retrieval among 200 candidates from 25.27% to 40.58%. Under strict leave-one-subject-out evaluation, calibrated NITs achieve 48.13% target selection accuracy. In a separate modular evaluation, the system achieves 65.63% task success, with all correctly selected objects subsequently grasped and lifted.
@inproceedings{luo2027neuralintent, title = {NeuralIntent: Bridging Noninvasive EEG and Embodied Foundation Models with Continuous Neural Intent Tokens}, author = {Luo, Xiaoqi and Gao, Weixiang and Zhang, Zhengnan and Xu, Taojie and Zhou, Mo and Shi, Xun and Gao, Yunyuan}, booktitle = {The Forty-First AAAI Conference on Artificial Intelligence (Under Review)}, year = {2027}, keywords = {EEG, brain-computer interface, neural intent token, embodied AI, VLA, assistive robotics}, }

AAAI-27

MSPhys-Diff: Diffusion with Multiscale Physiological Priors for Parkinson’s Disease EEG Generation

Weixiang Gao, Yunyuan Gao, Xiaoqi Luo, and 3 more authors

In The Forty-First AAAI Conference on Artificial Intelligence (Under Review). Third author · AAAI-27 submission · Under Review · Submission 46558. Framework: MSPhys-Diff. , 2027

Bib HTML

@inproceedings{gao2027msphysdiff,
  title = {MSPhys-Diff: Diffusion with Multiscale Physiological Priors for Parkinson's Disease EEG Generation},
  author = {Gao, Weixiang and Gao, Yunyuan and Luo, Xiaoqi and Lu, Jiangwen and Zhang, Zhengnan and Zhou, Mo},
  booktitle = {The Forty-First AAAI Conference on Artificial Intelligence (Under Review)},
  year = {2027},
}

2026

ASOC
Progressive Cross-Modal Attention Network for Brain–Computer Interface Decoding Using EEG and Functional Near-Infrared Spectroscopy (fNIRS)

Yunyuan Gao, Xiaoqi Luo, Zhengnan Zhang, and 2 more authors

Applied Soft Computing. IF≈7.8 · JCR Q1 · 中科院计算机 2区 TOP. Second author (advisor Yunyuan Gao 高云园 first author). Framework: PCMAN. , Jun 2026

Abs DOI Bib HTML PDF

Electroencephalography (EEG) offers millisecond-level temporal resolution but limited spatial localization, whereas functional near-infrared spectroscopy (fNIRS) provides better spatial resolution but shows a hemodynamic delay of several seconds. A hybrid EEG-fNIRS brain-computer interface fuses the complementary strengths of both modalities but faces challenges from intrinsic spatiotemporal heterogeneity between electrophysiological and hemodynamic signals. To address this, this study proposes a Progressive Cross-Modal Alignment Network (PCMAN), which achieves deep EEG-fNIRS fusion and alignment through a three-layer progressive mechanism: physical alignment (a Spatiotemporal correlation Dynamic Fusion framework, SDF), semantic alignment (a Multi-Dimensional Feature Enhancement Attention network, MDFEA), and task alignment (a Cross-modal Adaptive Decoding Network, CADN). On three cognitive tasks–mental arithmetic (MA), word generation (WG), and motor imagery (MI)–PCMAN achieved accuracies of 88.18%, 81.93%, and 72.94%, outperforming state-of-the-art hybrid BCI methods.
@article{gao2026pcman, title = {Progressive Cross-Modal Attention Network for Brain--Computer Interface Decoding Using EEG and Functional Near-Infrared Spectroscopy (fNIRS)}, author = {Gao, Yunyuan and Luo, Xiaoqi and Zhang, Zhengnan and Shao, Xu and Qin, Yanhua}, journal = {Applied Soft Computing}, volume = {199C}, pages = {115318}, year = {2026}, month = jun, publisher = {Elsevier}, doi = {10.1016/j.asoc.2026.115318}, url = {https://doi.org/10.1016/j.asoc.2026.115318}, keywords = {Hybrid brain-computer interface, EEG-fNIRS fusion, Cross-modal attention, Progressive alignment, Multimodal deep learning}, }
Under Review
Neuro-SSCL: A Neuro-Aware Semi-Supervised Contrastive Learning for Cross-Subject EEG Emotion Recognition

Jiangwen Lu, Xiaoqi Luo, Weixiang Gao, and 2 more authors

IEEE Journal of Biomedical and Health Informatics (Under Review). IF≈6.8 · JCR Q1 · 中科院医学 2区 TOP · Under Review (Initial Submission). Second author (corresponding author: advisor Yunyuan Gao). Framework: Neuro-SSCL. , 2026

Abs Bib

Electroencephalography (EEG) has been widely used in emotion recognition, yet the scarcity of labeled samples remains a major obstacle to its broader application. Existing semi-supervised EEG-based emotion recognition methods mainly rely on data-driven distribution alignment strategies and have not fully explored the neurophysiological priors underlying affective neural activity. To address this issue, we propose a neuro-aware semi-supervised contrastive learning framework, termed Neuro-SSCL. It incorporates neuroscience-derived inductive biases into model design and optimization. Neurophysiological topology-aware soft contrastive learning (NT-SCL) explicitly integrates local spatial correlations induced by volume conduction and global functional connectivity topology into contrastive learning, while spectrum-adaptive guided multi-scale graph contrastive learning (SA-MGCL) enhances emotion-relevant spectral components and imposes multi-scale graph contrastive constraints. A semi-supervised multi-domain adaptation strategy further mitigates distribution shifts across labeled/unlabeled source and target domains via adversarial alignment. On SEED, SEED-IV and SEED-V under a semi-supervised cross-subject leave-one-subject-out protocol, Neuro-SSCL consistently outperforms state-of-the-art methods under different incomplete-label conditions.
@article{lu2026neurosscl, title = {Neuro-SSCL: A Neuro-Aware Semi-Supervised Contrastive Learning for Cross-Subject EEG Emotion Recognition}, author = {Lu, Jiangwen and Luo, Xiaoqi and Gao, Weixiang and Huang, Ruyue and Gao, Yunyuan}, journal = {IEEE Journal of Biomedical and Health Informatics (Under Review)}, year = {2026}, keywords = {EEG, emotion recognition, contrastive learning, semi-supervised learning, neuroscience}, }
Under Review
Attention-Guided Low-Rank Tensor Decomposition for Feature Analysis of MI-EEG Signals

Xiao Wu, Weixiang Gao, Xiaoqi Luo, and 2 more authors

IEEE Transactions on Emerging Topics in Computational Intelligence (Under Review). IF≈6.5 · JCR Q1 · 中科院计算机 2区 · Under Review (Initial Submission). Third author. Framework: AG-LRTD. , 2026

Abs Bib

Motor imagery electroencephalography (MI-EEG) decoding remains challenging because EEG signals are weak, nonstationary, and strongly coupled across spatial, temporal, and spectral domains. Tensor decomposition is effective for high-dimensional feature mining, but conventional EEG tensor decomposition methods often overlook physiological characteristics and impose task-agnostic low-rank constraints. This paper proposes an attention-guided low-rank tensor decomposition (AG-LRTD) framework for MI-EEG feature analysis and classification. MI-EEG signals are organized as four-order tensors (channel, temporal, frequency-band, and trial modes) to preserve multi-dimensional spatial-temporal-spectral structure. A convolutional attention module is embedded into a tensor-product-based low-rank decomposition to enhance task-relevant spatial patterns; tensor-product factorization with a tubal-rank constraint performs structured dimensionality reduction and noise suppression, while residual-dense-block-based implicit regularization mitigates the limitation of a single low-rank prior for nonstationary EEG. On three public MI-EEG datasets under a subject-dependent binary protocol, AG-LRTD achieves 95.83%, 94.72%, and 90.07% on Dataset 1, Dataset 4a, and the binary left/right subset of Dataset 2a, respectively, while retaining an interpretable tensor-structured modeling process.
@article{wu2026aglrtd, title = {Attention-Guided Low-Rank Tensor Decomposition for Feature Analysis of MI-EEG Signals}, author = {Wu, Xiao and Gao, Weixiang and Luo, Xiaoqi and Zhang, Zhengnan and Gao, Yunyuan}, journal = {IEEE Transactions on Emerging Topics in Computational Intelligence (Under Review)}, year = {2026}, keywords = {Motor imagery EEG, brain-computer interface, attention mechanism, low-rank tensor decomposition, EEG classification}, }
Under Review
PSSF: A Phase-Space and State-Space Fusion Method for EEG Source Localization and Feature Decoding

Zheng Hu, Xiaoqi Luo, Xu Shao, and 2 more authors

Journal of Neural Engineering (Under Review). IF≈4.0 · JCR Q2 · Under Review (Manuscript Received, JNE-111255). Second author (corresponding authors: advisor Yunyuan Gao 高云园, Yanhua Qin). Framework: PSSF. , 2026

Abs Bib PDF

Scalp EEG signals are affected by volume-conduction-induced mixing, whereas source-localized time series contain nonlinear dynamics together with redundant activity from task-irrelevant regions of interest (ROIs). This study proposes PSSF, a phase-space and state-space fusion method for EEG source localization and feature decoding. Scalp EEG is projected into cortical source space using sLORETA to obtain ROI-level source time series. Phase-space reconstruction via time-delay embedding, together with approximate, fuzzy, and permutation entropy, characterizes nonlinear source dynamics; in parallel, a state-space model performs sparse source estimation and BIC-based model selection to identify active ROIs. Features retained under active-ROI constraints are ranked by ReliefF and classified with a linear SVM. PSSF achieved mean accuracies of 77.81% on BCI Competition IV 2a (binary MI) and 98.35% on a self-collected adult ADHD dataset; ablation showed gains of 6.98 and 5.72 percentage points for fuzzy and hybrid entropy after jointly introducing phase-space reconstruction and state-space ROI screening.
@article{hu2026pssf, title = {PSSF: A Phase-Space and State-Space Fusion Method for EEG Source Localization and Feature Decoding}, author = {Hu, Zheng and Luo, Xiaoqi and Shao, Xu and Gao, Yunyuan and Qin, Yanhua}, journal = {Journal of Neural Engineering (Under Review)}, year = {2026}, keywords = {electroencephalography, phase-space reconstruction, state-space model, source localization, EEG decoding}, }
Accepted
Neurovascular Alignment and Reliability-Aware Fusion for Continuous EEG–fNIRS Emotion Regression

Yunyuan Gao, Xiaoqi Luo, Zhengnan Zhang, and 3 more authors

IEEE Transactions on Affective Computing. IF≈9.8 · JCR Q1 · 中科院计算机 1区 TOP · Accepted on 2026-08-19. Second author (advisor Yunyuan Gao 高云园 first author). Framework: PhysioSync. , 2026

Abs Bib PDF

Continuous affect tracking is central to closed-loop affective brain–computer interfaces and adaptive human–computer interaction. Continuous EEG–fNIRS regression is well suited to this setting because EEG captures rapid neural dynamics and fNIRS provides complementary hemodynamic information, but direct fusion often underperforms EEG-only baselines when neurovascular coupling (NVC) introduces subject- and region-dependent delays. PhysioSync addresses this by performing HRF-like temporal alignment and residual-aware reliability control before multimodal fusion. A learnable dual-gamma kernel maps EEG-derived drive to the fNIRS timeline through finite-window causal convolution, and the resulting residuals modulate the contribution of hemodynamic evidence during fusion. A linear-complexity Mamba encoder provides long-context EEG representation. On REFED, PhysioSync reduces Avg MAE by 5.8% relative to the best fusion baseline and by 7.2% relative to EEG-only when evaluated across three within-subject folds, while showing lower error in 1–3 s trajectory forecasting. With lightweight task-specific adaptations, ENTER and ZBra-music results indicate conditional applicability to discrete classification rather than uniform transferability, with performance depending on EEG coverage and HbO/HbR chromophore decomposition.
@article{gao2026physiosync, title = {Neurovascular Alignment and Reliability-Aware Fusion for Continuous EEG--fNIRS Emotion Regression}, author = {Gao, Yunyuan and Luo, Xiaoqi and Zhang, Zhengnan and Lu, Jiangwen and Zhang, Yingchun and Huang, Ruyue}, journal = {IEEE Transactions on Affective Computing}, year = {2026}, keywords = {physiological affect recognition, EEG--fNIRS fusion, continuous emotion regression, neurovascular coupling, multimodal affective BCI}, }