|
Jiazhen Hong | ζ΄ͺεζ‘’ Department of Electrical and Computer Engineering at Rutgers University in Piscataway, NJ. Jiazhen Hong is a Ph.D. candidate in Electrical and Computer Engineering at Rutgers University, advised by Professor Laleh Najafizadeh, and a Research Scientist at EMOTIV Research. He serves on the Technical Program Committee of the 2026 IEEE International Symposium on Biomedical Imaging (ISBI). His research interests lie in non-invasive brain-computer interfaces, electroencephalography (EEG) decoding, real-time neural communication systems, and foundation models for neural signals. He has led the development of closed-loop, real-time P300 speller BCI systems at the Integrated Systems & NeuroImaging Laboratory, and has contributed to the design of EEG foundation models at EMOTIV, with multiple publications and patent filings. |
|
π₯ Newsπ Awards & Honors
π’ Journal & Conference Papers
π Career Milestone
|
Research BackgroundIn Fall 2024, I worked as a Research Intern at Emotiv, focusing on Foundation Models and Generative AI for Brain-Computer Interface (BCI) applications. At Rutgers in 2024, I designed a P300 Speller BCI incorporating Generative AI. In 2023, I developed an LLM-based BCI mind-controlled speller system aimed at assisting individuals with disabilities in communication. In 2022, I created a channel selection method to enhance the speed and efficiency of BCIs in real-world applications. In 2020, I proposed a geometric approach to optimize the k-means algorithm, addressing issues related to local minima. Before joining Rutgers University in 2019, I conducted biostatistics research at Harvard Medical School, focusing on pancreatic cancer diagnosis. This project utilized Gene Set Enrichment Analysis (GSEA) to identify significant biological differences using protein data, including subsets of around 1300 proteins and small gene panels with as few as 5-10 genes. |
|
Enhancing Typing Speed in LLM-based P300 Speller BCIs Using A New Data-Driven Dynamic Stopping Strategy
Jiazhen Hong, Laleh Najafizadeh, NER, 2025 IEEE EMBS 12th Annual International Conference on Neural Engineering Link We introduce a novel data-driven dynamic stopping strategy for LLM-based P300 speller BCIs, which adaptively determines when to finalize character selection based on real-time EEG decoding. By analyzing probability transitions from a reference group, we derive heuristic and salient thresholds that guide early stopping without requiring fixed repetition or arbitrary thresholds. This approach significantly reduces the number of flashes and enhances semantic communication speed, achieving over 260% improvement in information transfer rate compared to traditional methods. |
|
ChatBCI4ALS: A High-Performance, LLM-Driven Intent-Based BCI Communication System for Individuals with ALS
Jiazhen Hong, Pradyumna Rao, Weinan Wang, Shan Chen, Laleh Najafizadeh, EMBC, 2025 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society Link We present ChatBCI-4-ALS, the first intent-based P300 BCI communication system tailored for individuals with ALS. By integrating large language models (LLMs) with a dynamic flash algorithm, the system enables efficient semantic prediction and rapid character selection based on EEG. A novel GUI and ALS-relevant prompting strategy support intuitive message composition. Online results show record-breaking performance, achieving up to 42.16 characters/min and 128.85 bits/min information transfer rate, significantly advancing real-time BCI communication. |
|
TopoEEG: a TimeSformer-Based Topographic Image Representation Method for Single-Trial Early Detection of P300
Jiazhen Hong, Laleh Najafizadeh, ISBI, 2025 IEEE International Symposium on Biomedical Imaging Link We introduce TopoEEG, an innovative image-based framework that generates sequences of topographic maps from EEG data, preserving both temporal and spatial information for decoding neural dynamics. These topographic maps are processed using TimeSformer, a state-of-the-art video classification model with joint and divided space-time attention mechanisms. |
|
ChatBCI: A P300 Speller BCI Leveraging Large Language Models for Improved Sentence Composition in Realistic Scenarios
Jiazhen Hong, Weinan Wang, Laleh Najafizadeh, IEEE Brain, 2024 IEEE Brain Discovery and Neurotechnology Workshop. arXiv We present ChatBCI, an innovative P300 speller BCI that leverages the zero-shot learning capability of large language models (LLMs) to improve the speed of sentence writing for the user. The system retrieves word suggestions to either complete partially spelled words or predict the next word in a sentence, improving efficiency when in sentence composition. |
|
P3T: A Transformer Model for Enhancing Character Recognition Rates in P300 Speller Systems
Jiazhen Hong, Laleh Najafizadeh, Asilomar (ACSSC), 2024 Asilomar Conference on Signals, Systems, and Computers Poster We introduce P300-Transformer (P3T), a new single-trial P300 detec- tor transfer model, designed to optimize the information transfer rate (ITR) in P300-BCI speller systems, while maintaining a high character recognition rate. |
|
A Geometric Approach to k-means
Jiazhen Hong, Wei Qian, Yudong Chen, Yuqian Zhang, TDKE , IEEE Transactions on Knowledge and Data Engineering arXiv, TKDE We propose a flexible framework for k-means problem by harnessing the geometric structure of local solutions. It provides a theoretical foundation for future work to design detection routines for varying cluster distributions. |
|
|
A deep learning framework based on dynamic channel selection for early classification of left and right hand motor imagery tasks
Jiazhen Hong, Foroogh Shamsi, Laleh Najafizadeh, EMBC, 2022 IEEE / Rutgers Research Day / Rutgers Brain Health Institute Symposium We introduce a deep learning framework that utilizes dynamic channel selection for early classification of left versus right hand motor imagery (MI) tasks. This approach reduces data dimensionality, thereby accelerating future related brain-computer interface (BCI) technologies. |
π Publications |
Accepted
- J. Hong, W. Qian, Y. Chen, and Y. Zhang, βA geometric approach to k-means,β is accpeted by IEEE Transactions on Knowledge and Data Engineering.
- J. Hong G. Mackellar, S. Ghane, "SpellerSSL: Self-Supervised Learning with P300 Aggregation for Speller BCIs, is accpeted by the 39 Annual Conference on Neural Information Processing Systems (NeurIPS 2025), Data on the Brain & Mind Workshop.
- J. Hong G. Mackellar, S. Ghane, "An Efficient Self-Supervised Framework for Long-Sequence EEG Modeling" is accpeted by IEEE International Conference on Data Mining (ICDM 2025), AI for Time Series Workshop.
- J. Hong and L. Najafizadeh, βEnhancing Typing Speed in LLM-based P300 Speller BCIs Using A New Data-Driven Dynamic Stopping Strategy,β IEEE EMBS 12th Annual International Conference on Neural Engineering (NER 2025).
- S. Mai, J. Hong, T. Shors and L. Najafizadeh, βLongest Sustained Cortical Activity Duration in Temporal Bisection Task Aligns with Geometric Mean for Log-Spaced Probes,β IEEE EMBS 12th Annual International Conference on Neural Engineering (NER 2025).
- J. Hong, P. Rao, W. Wang, S. Chen and L. Najafizadeh, βChatBCI4ALS: A High-Performance, LLM-Driven Intent-Based BCI Communication System for Individuals with ALS,β IEEE EMBS 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2025).
- J. Hong and L. Najafizadeh, βTopoEEG: a TimeSformer-Based Topographic Image Representation Method for Early Single-Trial Detection of P300,β 22nd IEEE International Symposium on Biomedical Imaging (ISBI 2025).
- J. Hong, W. Wang and L. Najafizadeh, βChatBCI: A Fast P300 Speller Brain-Computer Interface Incorporating Generative AI-Based Word Prediction, 2024 IEEE Brain Discovery and Neurotechnology Workshop. (Spotlight) β Machine Learning and Computer Paradigms for Brain Discovery Posters.
- J. Hong, L. Najafizadeh, βP3T: A Transformer Model for Enhancing Character Recognition Rates in P300 Speller Systems,β 58th Annual Asilomar Conference on Signals, Systems, and Computers (ACSSC 2025).
- J. Hong, F. Shamsi, and L. Najafizadeh, βA Deep Learning Framework Based on Dynamic Channel Selection for Early Classification of Left and Right Hand Motor Imagery Tasks,β Proc. of 44th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBCβ22), Glasgow, Scotland, July 2022, pp. 3550-3553.
Under Review
- J. Hong, W. Wang, and L. Najafizadeh, βChatBCI: A P300 Speller BCI Leveraging Large Language Models for Improved Sentence Composition in Realistic Scenarios,β Under Review.
- J. Hong, G. Machellar, and S. Ghane, SAMBA: An Efficient Mamba-2-Based Self-Supervised Framework for Long-Sequence EEG Modeling,β Under Review.
In Preparation
- J. Hong, W. Wang, S. Haghani, and L. Najafizadeh, βSubject-specific Channel Selection Based on Davies-Bouldin Index for EEG Motor Imagery Classification,β in preparation.
π§βπ Supervised Students
|