Jia Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jia Wang has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Daliang Wang
Datatang (Beijing) Technology Co., Ltd., Beijing 101100, China
1 shared publication - Dan Jiang
School of Information Engineering, Beijing Institute of Graphic Communication, Beijing, China
1 shared publication - Guanyu Zou
School of Electornic and Computer Engineering, Beijing Institute of Graphic Communication, Beijing 102600, China
1 shared publication - Han Zhang
Beijing Institute of Graphic Communication
1 shared publication - Liang Lan
School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China
1 shared publication - Wenyong Wang
School of Computer Science and Engineering, Macau University of Science and Technology, Macau 999078, China
1 shared publication
Latest Publications
- 14 July 2026
- Published by Springer Nature
- p. 76-87
Controlled LoRA Adapter Placement for Wan2.1 Text-To-Video Generation
- 14 July 2026
- Published by Springer Nature
- p. 76-87
Controlled LoRA Adapter Placement for Wan2.1 Text-To-Video Generation
- Open Access
- 16 November 2025
- Published by MDPI AG in Electronics
- Vol. 14 (22) , 4466
SPaRLoRA: Spectral-Phase Residual Initialization for LoRA in Low-Resource ASR
Parameter-efficient fine-tuning (PEFT) methods like Low-Rank Adaptation (LoRA) are widely used to adapt large pre-trained models under limited resources, yet they often underperform full fine-tuning in low-resource automatic speech recognition (ASR). This gap stems partly from initialization strategies that ignore speech signals’ inherent spectral-phase structure. Unlike SVD/QR-based approaches (PiSSA, OLoRA) that construct mathematically optimal but signal-agnostic subspaces, we propose SPaRLoRA (Spectral-Phase Residual LoRA), which leverages Discrete Fourier Transform (DFT) bases to create speech-aware low-rank adapters. SPaRLoRA explicitly incorporates both magnitude and phase information by concatenating real and imaginary parts of DFT basis vectors, and applies residual correction to focus learning exclusively on components unexplained by the spectral subspace. Evaluated on a 200-h Sichuan dialect ASR benchmark, SPaRLoRA achieves a 2.1% relative character error rate reduction over standard LoRA, outperforming variants including DoRA, PiSSA, and OLoRA. Ablation studies confirm the individual and complementary benefits of spectral basis, phase awareness, and residual correction. Our work demonstrates that signal-structure-aware initialization significantly enhances parameter-efficient fine-tuning for low-resource ASR without architectural changes or added inference cost.
- 16 November 2025
- Published by MDPI AG in Electronics
- Vol. 14 (22) , 4466
Open AccessSPaRLoRA: Spectral-Phase Residual Initialization for LoRA in Low-Resource ASR
Parameter-efficient fine-tuning (PEFT) methods like Low-Rank Adaptation (LoRA) are widely used to adapt large pre-trained models under limited resources, yet they often underperform full fine-tuning in low-resource automatic speech recognition (ASR). This gap stems partly from initialization strategies that ignore speech signals’ inherent spectral-phase structure. Unlike SVD/QR-based approaches (PiSSA, OLoRA) that construct mathematically optimal but signal-agnostic subspaces, we propose SPaRLoRA (Spectral-Phase Residual LoRA), which leverages Discrete Fourier Transform (DFT) bases to create speech-aware low-rank adapters. SPaRLoRA explicitly incorporates both magnitude and phase information by concatenating real and imaginary parts of DFT basis vectors, and applies residual correction to focus learning exclusively on components unexplained by the spectral subspace. Evaluated on a 200-h Sichuan dialect ASR benchmark, SPaRLoRA achieves a 2.1% relative character error rate reduction over standard LoRA, outperforming variants including DoRA, PiSSA, and OLoRA. Ablation studies confirm the individual and complementary benefits of spectral basis, phase awareness, and residual correction. Our work demonstrates that signal-structure-aware initialization significantly enhances parameter-efficient fine-tuning for low-resource ASR without architectural changes or added inference cost.
- Open Access
- 5 May 2025
- Published by Springer Nature in International Journal of Computational Intelligence Systems
- Vol. 18 (1) , 1-20
Defects Detection in Screen-Printed Circuits Based on an Enhanced YOLOv8n Algorithm
Defect detection is a crucial task in screen-printed circuit (SPC) production, where image processing method based on deep learning is often used. This field frequently encounters challenges, such as minute surface defects, a large number of model parameters, and high computational complexity. To address these challenges, a self-made SPC defect data set and an enhanced CAAB-YOLOv8n detection algorithm were developed. A CAD module was integrated into the backbone network to improve the model’s ability to detect bar-shaped features. In addition, the ASF feature fusion and RMT modules were combined to construct the ASF-CR neck structure, which enhances the model’s capability to detect small, localized defects. To expedite inference speed, the DBB-Head reparameterization module was incorporated. Experimental results show that the enhanced algorithm achieves 88.4 $$\%$$ accuracy, a mAP@50 of 90.2 $$\%$$ , and a parameter count of just 33.27 million, with a detection speed of 35.2 frames per second. The real-time requirements for SPC defect detection are met by these findings. This work lays a solid theoretical foundation for subsequent defect traceability and the optimization of printing process parameters.- 5 May 2025
- Published by Springer Nature in International Journal of Computational Intelligence Systems
- Vol. 18 (1) , 1-20
Open AccessDefects Detection in Screen-Printed Circuits Based on an Enhanced YOLOv8n Algorithm
Defect detection is a crucial task in screen-printed circuit (SPC) production, where image processing method based on deep learning is often used. This field frequently encounters challenges, such as minute surface defects, a large number of model parameters, and high computational complexity. To address these challenges, a self-made SPC defect data set and an enhanced CAAB-YOLOv8n detection algorithm were developed. A CAD module was integrated into the backbone network to improve the model’s ability to detect bar-shaped features. In addition, the ASF feature fusion and RMT modules were combined to construct the ASF-CR neck structure, which enhances the model’s capability to detect small, localized defects. To expedite inference speed, the DBB-Head reparameterization module was incorporated. Experimental results show that the enhanced algorithm achieves 88.4 $$\%$$ accuracy, a mAP@50 of 90.2 $$\%$$ , and a parameter count of just 33.27 million, with a detection speed of 35.2 frames per second. The real-time requirements for SPC defect detection are met by these findings. This work lays a solid theoretical foundation for subsequent defect traceability and the optimization of printing process parameters.