Wei Zhang

Dr. Wei Zhang


Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
22 July 2025 in Acta Biomaterialia
Verified
Journal Article
Closed access
JHJixin Hou
Jixin Hou
KJKun Jiang
Kun Jiang
ARArunachalam Ramanathan
Arunachalam Ramanathan
AKAbhishek Saji Kumar
Abhishek Saji Kumar
WZWei Zhang
Wei Zhang
XWXianqiao Wang
Xianqiao Wang

2 August 2024 in arXiv
Verified
Preprint
Open access
PDF file
JWJiaqi Wang
Jiaqi Wang
HJHanqi Jiang
Hanqi Jiang
YLYiheng Liu
Yiheng Liu
CMChong Ma
Chong Ma
XZXu Zhang
Xu Zhang
SZShu Zhang
Shu Zhang
In an era defined by the explosive growth of data and rapid technological advancements, Multimodal Large Language Models (MLLMs) stand at the forefront of artificial intelligence (AI) systems. Designed to seamlessly integrate diverse data types-including text, images, videos, audio, and physiological sequences-MLLMs address the complexities of real-world applications far beyond the capabilities of single-modality systems. In this paper, we systematically sort out the applications of MLLM in multimodal tasks such as natural language, vision, and audio. We also provide a comparative analysis of the focus of different MLLMs in the tasks, and provide insights into the shortcomings of current MLLMs, and suggest potential directions for future research. Through these discussions, this paper hopes to provide valuable insights for the further development and application of MLLM.

30 January 2025 in Electronics
Verified
Journal Article
Open access
PDF file
AZArif Hassan Zidan
Arif Hassan Zidan
AJAfrar Jahin
Afrar Jahin
YBYu Bao
Yu Bao
WZWei Zhang
Wei Zhang
Deep learning has become indispensable for identifying hierarchical spatial features (SFs), which are crucial for linking neurological disorders to brain functionality, from functional Magnetic Resonance Imaging (fMRI). Unfortunately, existing methods are constrained by architectures that are either linear or nonlinear, limiting a comprehensive categorization of spatial features. To overcome this limitation, we introduce the Semi-Nonlinear Deep Efficient Reconstruction (SENDER) framework, a novel hybrid approach designed to simultaneously capture both linear and nonlinear spatial features, providing a holistic understanding of brain functionality. In our approach, linear SFs are formed by directly integrating multiple spatial features at shallow layers, whereas nonlinear SFs emerge from combining partial regions of these features, yielding complex patterns at deeper layers. We validated SENDER through extensive qualitative and quantitative evaluations with four state-of-the-art methods. Results demonstrate its superior performance, identifying five reproducible linear SFs and eight reproducible nonlinear SFs.

2 April 2026 in Contemporary Mathematics
Verified
Journal Article
Open access
AZArif Hassan Zidan
Arif Hassan Zidan
AJAfrar Jahin
Afrar Jahin
YBYu Bao
Yu Bao
TLTianming Liu
Tianming Liu
WZWei Zhang
Wei Zhang
While gradient-based optimizers that incorporate randomization often demonstrate superior performance on complex optimizations, the theoretical foundations of this advantage remain underexplored. A central question arises: What role does randomization play in dimension-free, nonsmooth, nonconvex optimization? To address this gap, we examine both the theoretical and empirical impact of permutation randomization within gradient-based optimization frameworks, using it as a representative case to investigate broader implications. From a theoretical perspective, our analysis reveals that permutation randomization disrupts the shrinkage behavior characteristic of gradient-based optimizers, allowing for continued progress toward the global optimum with sufficient iterations. Moreover, we prove that permutation randomization preserves the convergence rate of the underlying optimizer. Empirically, we conduct extensive numerical experiments comparing permutation-randomized optimizers with three baseline methods. These experiments span tasks such as training deep neural networks with stacked architectures and optimizing noisy objective functions. The results not only support our theoretical findings but also demonstrate the practical benefits of permutation randomization. In summary, this work provides both rigorous theoretical justification and compelling empirical evidence for the effectiveness of permutation randomization, establishing a foundation for extending such analyses to broader implications of randomized strategies.

4 December 2024 in IBRO Neuroscience Reports
Verified
Journal Article
Open access
AZArif Hassan Zidan
Arif Hassan Zidan
AJAfrar Jahin
Afrar Jahin
YBYu Bao
Yu Bao
WZWei Zhang
Wei Zhang
Mapping consistent longitudinal intrinsic connectivity networks (ICNs) is crucial for understanding brain functional development over various life stages. However, achieving consistent longitudinal ICNs has been challenging due to the lack of methodologies that maintain temporal consistency. To address this gap, we introduce an innovative approach named Longitudinal Sparse Dictionary Learning (LSDL). This method utilizes an additional Frobenius norm to bridge gaps between consecutive ICNs, facilitating the continuous transfer of the learned feature matrix to subsequent stages. Moreover, Matrix Backpropagation (MBP) is employed to effectively mitigate potential accumulative errors. Our validation results demonstrate that LSDL can successfully extract 21 consistent longitudinal ICNs in macaque brains. In comparative empirical evaluations with established methodologies, Fast Independent Component Analysis (FICA) and Sparse Dictionary Learning (SDL), LSDL showcases superior efficacy in modeling longitudinal functional Magnetic Resonance Imaging (fMRI) data. This approach opens new avenues for research into developmental brain dynamics and neurodegenerative disorders, providing a robust framework for tracking the evolution of brain connectivity over time.

28 October 2024
Verified
Conference Paper
Open access
WZWei Zhang
Wei Zhang
XJXi Jiang
Xi Jiang
SZShijie Zhao
Shijie Zhao
NQNing Qiang
Ning Qiang
YBYu Bao
Yu Bao
JLJinglei Lv
Jinglei Lv
Deep Neural Networks (DNNs) have successfully uncovered hierarchical functional connectivity networks (FCNs) in the human brain. In particular, the FCNs identified in the deep layers of DNNs usually represent the combinations of traditional FCNs, which are closely relevant to neurological disorders. We define these deep-layer FCNs as large-scale FCNs (LS-FCNs). A few existing studies reported LS-FCNs, but reproducibility remains a challenge. To address this, we propose an innovative deep matrix factorization technique, named Deep Linear Matrix Approximate Reconstruction (DELMAR), designed to identify reproducible LS-FCNs. DELMAR introduces several methodological advancements in functional Magnetic Resonance Imaging (fMRI) analytics. Firstly, it incorporates an efficient, non-fully connected layer to reduce the computational complexity associated with traditional fully connected layers. Additionally, we have developed a hyperparameter tuning technique that enables automatic hyperparameter estimation, and we introduce matrix backpropagation to minimize potential accumulative errors. Validation experiments were conducted using Multi-band Multiecho (MBME) resting-state fMRI data alongside two peer methods to assess the reproducibility of LS-FCNs. The results demonstrate that DELMAR effectively extracts various reproducible, and even previously unreported LS-FCNs.