Yucheng Wang

Education

Sun Yat-sen University Guangzhou, China

Microelectronics Science and Engineering Sep. 2024 -- Aug. 2028

Academics: GPA 3.2 / 4.0 ; Sun Yat-sen University Outstanding Student Scholarship (Top 30%, 2024)
Relevant Courses: Digital Logic Design, Computer Architecture, Machine Learning, Signals and Systems, Hardware Description Languages (HDL) and FPGA.

Research on RISC-V Based Transformer 1.58-bit Matrix Acceleration Nov. 2025 -- Present

Research Assistant | Mentor: Xiang Chen (Sun Yat-sen University)

Operator Reconstruction: Addressing the "Memory Wall," I reconstructed FP32 GEMM into multiplier-free ternary (-1, 0, 1) operators based on BitNet b1.58. This significantly reduced FLOPs and resource overhead by eliminating heavy MAC operations, paving the way for efficient LLM compression on edge devices.
TDPU Design: Designed the Tensor Data Processing Unit (TDPU) with fine-tuned memory access patterns for Transformers. Established an efficient low-bit inference pipeline on FPGA, drastically reducing power consumption while maintaining consistency.

Multimodal Large Model Training System Optimization (PyTorch) Apr. 2025 -- May. 2025

Research Intern (Sun Yat-sen University)

Memory & I/O Optimization: Mitigated OOM for multimodal models (ResNet+BERT+GRU) via PyTorch AMP and dynamic FP32/FP16 scheduling. Resolved CPU-GPU bottlenecks using torch.multiprocessing and DistributedSampler to maximize hardware utilization and training throughput.
Compute Graph Customization: Reconstructed loss graphs for multi-state robot policy tasks. Implemented multi-task joint loss with Cosine Annealing to ensure smooth gradient descent and operator convergence, significantly improving model robustness and accelerating convergence speeds.

AI-Copilot Assisted Complex System Modeling (CUMCM) Sep. 2025

Research Team Leader (China Undergraduate Mathematical Contest in Modeling)

AI-Driven Engineering: Validated "Human-AI Co-programming" to translate math models (ODE/NLP) into vectorized solvers. Demonstrated the potential of "Agentic AI" for rapid, low-latency code generation in complex system modeling.
Algorithm Optimization: Solved high-dimensional constraints via AI-assisted heuristic algorithms (GA/SA). Conducted automated sensitivity analysis to verify model stability under parameter perturbations, winning the National Second Prize.

Intelligent Digital Health Intervention System Design Jan. 2026

Research Project Lead (Tallinn University Winter School Project)

System Architecture: Developed a COM-B intervention system with AI analysis for behavior modeling. Accurately identified distraction triggers through deep user pattern modeling and dynamic interventions.
Product Interaction: Applied Copilot concepts to design "AI-guided interactions." Implemented gamified, data-driven feedback loops to effectively close the user intervention cycle and enhance user engagement.

National Second Prize, China Undergraduate Mathematical Contest in Modeling (CUMCM) Sep. 2025
First Prize in Guangdong Province, "Shenzhen Cup" Mathematical Modeling Challenge May. 2025
Third Prize in Guangdong Province, National Undergraduate Electronic Design Contest (NUEDC) Aug. 2025
First Prize, Sun Yat-sen University Selection for CUMCM May. 2025
Third Prize, 1st Sun Yat-sen University Artificial Intelligence Challenge Apr. 2025
Third-Class Scholarship, Sun Yat-sen University 2024

Languages: Python, C/C++, MATLAB, SystemVerilog, Shell Scripting, LaTeX
AI/ML Frameworks: PyTorch (AMP, Distributed), TensorFlow, GNNs, LLMs (BitNet, Transformer), Copilot-Assisted Dev
Developer Tools: Linux/Unix, Git, Docker, FPGA (Vivado), RISC-V Toolchain, Slurm, Weights & Biases
Domain Knowledge: MLSys Optimization, Hardware-Software Co-design, 5G Protocols, Signal Processing