Technology News

Jun Ye Advances Neural Interface Technology Through Brain–Computer Systems and Generative AI

Neural interface technology continues to face challenges in signal acquisition accuracy and real-time decoding as brain–computer interfaces (BCIs) become increasingly vital for patients with motor impairments. Traditional non-invasive systems struggle with weak signals, susceptibility to interference, and complex decoding pipelines. Recent work in the International Journal of Big Data Intelligent Technology introduces a comprehensive framework that integrates advanced preprocessing with deep learning architectures to improve EEG interpretation and assistive communication. The analytical foundation addresses core limitations via a segmentation–recombination preprocessing method that reduces a 31-class character task to an optimized 5-class problem. By analyzing temporal complexity and reassembling similar segments using Spearman correlation, the framework expands dataset utility while enhancing feature extraction. Subject attention is monitored through TGAM-chip integration, collecting EEG only when attention exceeds 60%, thereby maintaining high-quality data throughout extended sessions. The recognition architecture leverages an MLSTM network—combining mapping modules with LSTM structures—for direct EEG-to-text conversion. To mitigate gradient explosion and vanishing common in traditional RNNs, the model employs forget, input, and output gates for long-term memory control. L2 regularization helps prevent overfitting and supports generalization across subjects. Experimental results show 96.87% accuracy at 128 hidden units, outperforming baseline implementations including LSTM (91.57%), RNN (90.08%), GRU (88.36%), CNN (84.89%), and SVM (78.58%). Practical implementations span assistive technology and education. The handwritten-text BCI enables individuals with stroke, spinal cord injury, or amyotrophic lateral sclerosis to communicate via imagined handwriting captured by 24-electrode caps, processed in real time under attention monitoring. Analysis across 50 participants indicates highest accuracy among those aged 20–25, with male participants achieving marginally higher averages across demographics, supporting applicability across diverse user populations. Contributing to this research is Jun Ye, whose background bridges research and industry. He holds an M.S. in Electrical and Computer Engineering from Carnegie Mellon University. From 2019 to 2025, his work at Meta focused on production-grade sEMG decoding for neural wristband control, achieving >90% population coverage. His technical contributions include distributed training and benchmarking pipelines for multi-terabyte datasets, continuous data-quality assurance computer vision based pipeline recovering ~5% otherwise unusable data, deep autoencoder spike sorting architectures improving decoding accuracy by ~20%, and sub-minute onboarding web platform replacing much lengthier per-session setup procedures. Building on his neural-interface experience, Jun Ye has extended his innovations into education with generative AI. His Sound Pages application earned second place at the OpenMobile AI Hackathon, demonstrating the translation of advanced ML into socially beneficial products. The mobile app converts children’s prompts into fully illustrated, narrated storybooks within minutes via a multi-agent pipeline (planner, writer, illustrator, audio director), with age-appropriate guardrails, automatic mobile pagination, and consistent character synthesis. Interactive features—tap-to-highlight read-along, vocabulary callouts, and optional comprehension prompts—encourage parent–child engagement. Collectively, this body of work represents significant contributions across assistive technology and educational innovation. By combining attention-supervision paradigms with advanced neural architectures for BCIs and deploying accessible generative-AI applications, these efforts establish new benchmarks for neural-interface reliability and demonstrate broader impact in healthcare rehabilitation and early childhood literacy within an increasingly interconnected technological ecosystem.