Wenchang CHAI (柴文昌)

PolyU, Hong Kong

Project: Decentralized Clean Energy Verification Framework via Edge LLMs

Advisor: Chair Prof. Jiannong Cao & Prof. Jiang Shan, PolyU, HK

Abstract: Proposed the DeCEN architecture (integrating edge computing, blockchain, and AI-driven verification) to advance the reliability, trust, and efficiency of decentralized clean energy networks, which offers a scalable pathway toward global renewable energy targets.

My work: Framework development and evaluation, data analysis (Analyzed the performance of DeepSeek-R1 models for edge deployment and identified 14B outperforms 70B model), paper writing. (1 paper published)

PolyU, Hong Kong

Project 1: Optimization of Direct-to-Satellite (DtS) Connectivity in Low Earth Orbit (LEO) Satellite IoT

Advisor: Prof. Xia Xianjin, PolyU, HK

Abstract: Conducted in-depth investigations with 27 tiny ground stations across 8 locations on 4 continents to evaluate the network availability, reliability, cost, and energy performance of Satellite IoT, to optimize the system's performance.

My work: Designed and assembled the Tiny Ground Station devices; Developed TLE-based scheduling algorithms with real-time Doppler compensation and LoRa duplex protocols; Data collection and analysis; Paper writing. (1 paper accepted)

Project 2: Spatial Reuse Improvement and Spectrum Efficiency Optimization in LPWANs

Advisor: Prof. Xia Xianjin, PolyU, HK

Abstract: Developed a novel framework HydraNet, which leverages the capture effect of LPWAN radios to enable robust concurrent transmissions, and achieved a higher spectrum utilization over the state-of-the-art.

My work: Hardware setup, System synchronization, Data collection. (1 paper accepted)

Advisor: Prof. Wenxiong Kang, SCUT, China Mainland

Abstract: Proposed a unified LLM framework for cross-modal physiological signal analysis (ECG/EOG/EEG/EMG).

My work: Co-designed cross-modal attention mechanisms enabling LLMs to interpret bioelectric signals. Dataset selection and data preprocessing. (1 paper submitted)

Advisor: Prof. Hejun Wu, SYSU, China Mainland

Abstract: Proposed a novel prompting paradigm Proactive Prompting, which allow LLMs to generate prompts to steer its reasoning steps. Developed a reasoning framework Think Before You Speak with proactive test-time scaling ability.

My work: Deploy Qwen2.5-7B-Instruction model; Analyzed TBYS and SC accuracy on MATH-500 and AIME; Coding and paper writing. (1 paper accepted)

Advisor: Prof. Hejun Wu, SYSU, China Mainland

Abstract: A litchi detection framework with 31.06% AP improvement and 92.3% recognition accuracy in orchards.

My work: Created AIA-guided queries and EFPN with four modules for accurate MFBB detection in complex branches; Constructed LitchiMFBB dataset through comprehensive annotation of litchi clusters and MFBBs. (1 paper submitted)

Advisor: Prof. Wu Jibin, PolyU, HK

Abstract: Teaching Assistant-GPT, an RAG system providing accurate information for students on academic programs, course syllabus, and services.

My work: Designed Generator module using Langchain with query reformulation; Achieved 4.4/5.0 user satisfaction in study with 50 students; Outperformed Qwen/Gemini models in accuracy. (URIS Research Proposal)

Advisor: Prof. Chan Chun Bun Henry, PolyU, HK

Abstract: Vision Pro-based accessibility system enabling disabled learners to input formulas via multimodal interfaces.

My work: Designed 3 assistive modes for students with mobility impairments, upper-limb dysfunction, or hearing limitations; Implemented Swift frontend featuring multimodal keyboard and teacher dashboard for assignment management. (Award)