As brain-computer interfaces become increasingly precise in reading human emotions, a practical question arises: Will a model trained today still be able to recognize the emotions of the same person tomorrow, in a few weeks, or even months later? Recently, the team led by Shuang Liu from the Medical School of Tianjin University/Haihe Laboratory of Brain-Computer Interaction and Human-Machine Integration published a research paper titled "Toward Sustainable Cross-Time Affective Brain–Computer Interfaces: Explicit Modeling and Structural Calibration of Temporal Variability" in the National Science Review (NSR). This study offers a cross-time "EEG ruler" for affective brain-computer interfaces (aBCIs). To address temporal variability in cross-time emotion recognition, the researchers propose a Multi-level Dynamic Integrated Perception Network (MDIN), which explicitly models and structurally calibrates temporal variability on three levels: emotional neural representation, dynamic functional connectivity, and cross-time discriminative structures, offering a new technical pathway for long-term stable affective brain-computer interfaces. This work may further accelerate the transition of aBCIs from short-term experimental validation to long-term, adaptive use in real-world mental health monitoring and human-machine interaction.

Researchers have developed a high‑precision state of health prediction model that combines physical information constraints and multi‑dimensional features for lithium‑ion batteries under random variable load conditions. A neural network‑transformer framework based on physical information was constructed, converting the consistency laws of irreversible battery aging into regularization loss terms in the training process to ensure physical rationality. Multidimensional coupled features including voltage, current, and cycle count were extracted, and Bayesian optimization was applied to adaptively adjust key hyperparameters. Experimental results show that the model outperforms comparison models in terms of MAE, RMSE, and R² metrics, effectively suppressing non‑physical fluctuations and balancing high prediction accuracy with physical rationality.