Best Paper Competition

Best Paper Competition

Best Paper Candidate
01 Dual Attention Enhanced Multi-Scale Inverted Residual Neural Network for Image Inpainting
Kuan-Hsien Liu and Chun-Chieh Chang (National Taichung University of Science and Technology, Taiwan); Tsung-Jung Liu (National Chung Hsing University, Taiwan)

We propose DA-MSIRNet, an efficient image inpainting framework that enhances the U-Net by integrating Sparse Self-Attention, Context Anchor Attention, Multi-Scale Inverted Residual (MSIR) Blocks, and Structural Similarity Index Measure (SSIM) Loss. The Sparse Self-Attention, inspired by Spa-former, effectively balances global and local feature modeling. The MSIR optimizes U-Net's skip connections, while Context Anchor Attention reduces computational overhead and enhances reconstruction details. Additionally, the SSIM Loss improves structural integrity and texture consistency in the inpainted regions. Our framework addresses key limitations of existing approaches, including GANs, U-Net, and Transformers, achieving superior inpainting quality with improved accuracy and efficiency. Extensive experiments on the Places2 and CelebA-HQ datasets show that our method outperforms state-of-the-art techniques. The code for our DA-MSIRNet is publicly available on GitHub: https://github.com/nutcliu2507/DA-MIRNet.
02 AI-Powered VR Simulations for Knee Post-Surgical Physiotherapy and Recovery
Wai Yie Leong (INTI International University, Malaysia)

Knee post-surgical physiotherapy is a crucial phase of patient recovery, requiring consistent rehabilitation exercises to restore mobility, strength, and function. However, traditional physiotherapy methods face challenges such as high costs, limited accessibility, and low patient adherence. AI-powered Virtual Reality (VR) simulations offer a promising alternative by providing personalized, engaging, and data-driven rehabilitation experiences. By integrating artificial intelligence (AI) with immersive VR environments, these simulations can dynamically adjust rehabilitation exercises, track patient progress in real-time, and deliver instant feedback to enhance recovery outcomes. This paper explores the methodologies, case studies, challenges, and future directions of AI-powered VR in knee post-surgical rehabilitation. A comparative analysis between traditional physiotherapy and AI-VR-assisted therapy highlights significant improvements in patient adherence, pain reduction, and recovery time. Furthermore, a case study involving 100 post-operative knee surgery patients demonstrates the effectiveness of AI-driven VR therapy in enhancing rehabilitation outcomes. The study identifies ethical considerations, technological limitations, and implementation challenges, emphasizing the need for further research and development in this field. The findings suggest that AI-powered VR has the potential to optimize patient recovery, reduce healthcare burdens, and make physiotherapy more accessible, paving the way for next-generation rehabilitation solutions.
03 Block-Variable Combination Strategy for Distributed Blind Estimation in Fading Channel
Chi Sulin (Otemon Gakuin University, Japan); Hsuan Fu Wang (National Formosa University, Taiwan); Tetsuya Shimamura (Saitama University, Japan)

In long-distance communication systems, distributed blind estimation has gained significant attention as it enables the estimation of transmitted data signal through a wireless sensor network (WSN) by using a group of cooperative sensors, without the knowledge of the training signal in advance at the receiver. However, the performance of distributed blind estimation is highly sensitive to transmission channels. There have been very few studies investigating the impact of channel conditions on distributed blind estimation, particularly under dynamically changing channels. In this paper, we consider a fading channel for distributed blind estimation. To achieve optimal signal estimation and mitigate intersymbol interference for distributed blind estimation, a block-variable combination strategy is proposed. The distributed diffusion generalized Sato algorithm is employed to design the blind equalizer for signal estimation. In computer simulations, the average mean square error (MSE) is used to evaluate the performance of distributed blind equalization with the proposed method compared to conventional method.
04 A novel LLM-enhanced anomaly prediction framework for IoT time-series data
Peng Qi, Dan Tao and Ruipeng Gao (Beijing Jiaotong University, China)

The rapid development of IoT has enhanced production flexibility, but concomitantly increased the risk of system failures. Existing anomaly prediction approaches necessitate the adaptation of feature extraction or model retraining for different scenarios, which complicates practical engineering. To address it, this paper proposes an LLM-enhanced anomaly prediction framework for IoT time-series data. First, a general LLM is fine-tuned to equip it with domain prediction capabilities. Next, a lightweight TCN-LSTM model is designed for resource-constrained edges. An adversarial distillation strategy is then proposed to align the hidden vectors of TCN-LSTM with those of the LLM, facilitating knowledge transfer from the fine-tuned LLM to TCN-LSTM. Experimental results demonstrate the efficacy of the proposed design.
05 A Reinforcement Learning-Based Mobile Charging Agent for Wireless Rechargeable Sensor Networks
Cuijuan Shang and Qiaoyun Zhang (Chuzhou University, China); Wan-Chi Yang (National Taipei University of Nursing and Health Sciences, Taiwan); Chih-Yung Chang (Tamkang University, Taiwan)

Wireless Rechargeable Sensor Networks (WRSNs) are crucial for many applications, including environmental monitoring, healthcare, and smart cities. However, optimizing the charging schedule of a mobile charger to maximize network coverage while minimizing charging latency and energy consumption remains a challenge. In this paper, we design a reinforcement learning-based mobile charging agent (RLMCA), which integrates dynamic window search (DWS), Q-learning, and SARSA to improve charging efficiency. The proposed RLMCA combines dynamic threshold-based charging requests and an improved reward function that accounts for sensor coverage contribution. Extensive simulations demonstrate that RLMCA outperforms conventional methods in terms of charging latency, energy usage efficiency, and network coverage.
Best Student Paper Candidate
01 Clustering-based optimization of data collection points using minimum bounding circles for UAV-assisted edge computing systems
Nao Maeda (Kansai University, Japan); Tomotaka Kimura (Doshisha University, Japan); Kouji Hirata (Kansai University, Japan)

Unmanned aerial vehicles (UAVs) have emerged as essential components in cognitive radio (CR) networks, enabling enhanced spectrum awareness and dynamic spectrum access. A key function in CR systems is automatic modulation classification (AMC), which identifies modulation schemes to optimize signal processing and spectrum management. While AMC techniques perform well in controlled terrestrial environments, their deployment on UAV platforms faces new challenges due to UAV-induced vibrations and Doppler effects. This paper quantifies these impacts using a simulation framework based on the RadioML2016.01A dataset and a deep learning-based AMC model, explicitly incorporating UAV motion dynamics. Results reveal that phase distortions and frequency drifts significantly degrade AMC performance. These findings guide the development of resilient AMC techniques for UAV-enabled CR systems in dynamic environments.
02 Homemade Potentiostat with Cyclic Voltammetry Measurement Capability
Long Huang (National Sun Yat-sen University, Taiwan); Yin-Da Feng (National Kaohsiung University of Science and Technology, Taiwan); Mei-Ying Chang and Wen-Yao Huang (National Sun Yat-sen University, Taiwan); Tung-Li Hsieh (National Kaohsiung University of Science and Technology, Taiwan)

While commercially available potentiostats are expensive, traditional electrodes are not ideal for independent research, and are bulky, building a simple potentiostat with simple electronic components can help us detect unknown chemicals when needed. We use cyclic voltammetry (CV) for circuit analysis and transmit the data to the mobile app via Bluetooth to view the measurement results in real time.
03 Impact of UAV-Induced Vibrations and Doppler Effects on Deep Learning-Based Automatic Modulation Classification
Ying-Chen Chen (Nation Central University, Taiwan); Hai-Yan Huang and Yu-Jia Chen (National Central University, Taiwan)

Unmanned aerial vehicles (UAVs) have emerged as essential components in cognitive radio (CR) networks, enabling enhanced spectrum awareness and dynamic spectrum access. A key function in CR systems is automatic modulation classification (AMC), which identifies modulation schemes to optimize signal processing and spectrum management. While AMC techniques perform well in controlled terrestrial environments, their deployment on UAV platforms faces new challenges due to UAV-induced vibrations and Doppler effects. This paper quantifies these impacts using a simulation framework based on the RadioML2016.01A dataset and a deep learning-based AMC model, explicitly incorporating UAV motion dynamics. Results reveal that phase distortions and frequency drifts significantly degrade AMC performance. These findings guide the development of resilient AMC techniques for UAV-enabled CR systems in dynamic environments.
04 A Dual-Patch Bidirectional Mamba Model for IoT Network Traffic Classification
Yuan-Cheng Yu, Yu-Ling Lin, Yen-Chieh Ouyang and Chun-An Lin (National Chung Hsing University, Taiwan)

In recent years, machine learning (ML) has been widely studied for analyzing Internet of Things (IoT) traffic. We propose a novel IoT network traffic classification algorithm, the Dual-Patch Bidirectional Mamba (DPBM) model. Leveraging bidirectional inputs and the incorporation of patch operations, the model is enabled to better capture local traffic features. We evaluate the proposed approach on two publicly available traffic datasets, and the results demonstrate its effectiveness in identifying traffic patterns in edge computing environments.
05 High Speed and Low Power Simple Serial Peripheral Interface Using TSPC-based DFF
Yu-An Chen, Chia-Chen Chao and Yue-Fang Kuo (Yuan Ze University, Taiwan); Hung-Che Wei (National Kaohsiung University of Science and Technology, Taiwan)

This paper proposes a simple structure of n-bit SPI based on SIPO and PISO shift registers. Both shift registers uniformly use the proposed TSPC-based DFF circuit to eliminate the unnecessary switching activities of the internal nodes and reduce extra power consumption. The proposed SPI is designed in the TSMC 180nm CMOS process and 1.8V voltage supply. Compared with the simulated results of SPI with the traditional TSPC DFF, the power consumption of the proposed 8-bit and 12-bit SPI at 2GHz and 1GHz is reduced by 22% and 28%, respectively.
06 Optimal Assignment Method for a Relay AMR to AMR ad hoc Networks for High Throughput Communications
Kazuki Maeda, Tutomu Murase, Risa Takeuchi and Jingcheng Huang (Nagoya University, Japan)

In this study, we propose a relay autonomous mobile robot (AMR) assignment control method so that as many AMRs as possible can achieve a fixed value or higher throughput. In this system, multiple AMRs that communicate with an edge server form ad-hoc networks in which the AMRs relay communications. A heuristic method is used to assign the optimal relay AMR by creating a simple pseudo-path for each AMR and using an indicator calculated along that path. By predicting the throughput of the AMR when creating the pseudo-path, it is possible to make good allocations. After qualitative and quantitative evaluation, it was found that this method was highly effective, with the throughput duration improving by 19.5% compared to the results of the conventional method.