MA CHENGJIE

MA CHENGJIE

Ph.D. student at Yonsei University, based in Sinchon, Seoul

About Me

I am MA CHENGJIE, a Ph.D. student in Electrical and Electronic Engineering at Yonsei University, based in Sinchon, Seoul. My research focuses on Efficient Machine Learning, Federated Learning, LLM Fine-tuning, and AI-RAN. If you enjoy badminton, cycling, hiking, or academic exchange, feel free to reach out. You can contact me at cjma@ramo.yonsei.ac.kr.

Download CV (PDF)

News

  • 2026: My first ICML paper, “Breaking the Capacity Bottleneck in Model-Heterogeneous Federated Learning via Gradual Model Restoration,” appears in PMLR, volume 306, pages 84807–84845.
  • Planned, November 2026: Technical exchanges during AI-RAN Summit activities (November 9–10, Seoul) and an initial CPU–GPU PHY acceleration demo at the AI-RAN Alliance All Member Meeting (November 11–12, Yonsei University).

Educations

  • 2024.09 - Present: Ph.D. student, Department of Electrical and Electronic Engineering, Yonsei University
    • GPA: 97.3/100
    • Advisor: Professor Seong-Lyun Kim
    • Research focus: Efficient Machine Learning and AI-RAN
  • 2021.09 - 2024.06: M.S. in Computer Science and Technology, Beijing University of Posts and Telecommunications
    • GPA: 92.69/100
    • Advisor: Professor Junping Du (IEEE Fellow)
    • Thesis: Communication Optimization in Federated Learning Based on Asynchrony and Pruning
  • 2017.09 - 2021.06: B.S. in Computer Science and Technology, Chongqing University of Posts and Telecommunications
    • GPA: 87.11/100, Top 2%
    • Honorary Bachelor’s Degree
    • Minor: Business Administration

Publications

Breaking the Capacity Bottleneck in Model-Heterogeneous Federated Learning via Gradual Model Restoration
Chengjie Ma, S. Oh, J. Park, and S.-L. Kim Venue: Proceedings of the 43rd International Conference on Machine Learning (2026) Volume: 306 Pages: 84807–84845 Publisher: PMLR
Efficient Federated Learning Based on Parameter Group Training
Chengjie Ma et al. Venue: Proceedings of the Korean Institute of Communications and Information Sciences (KICS) Conference (2025) Pages: 227–228
Federated Topic Model and Model Pruning Based on Variational Autoencoder
Chengjie Ma, Y. Li, M. Liang, and A. Li Venue: Chinese Intelligent Automation Conference (CIAC) (2023)
Topic Model Based on Co-occurrence Word Networks for Unbalanced Short Text Datasets
Chengjie Ma, J. Du, M. Liang, and Z. Guan Venue: IEEE International Conference on Data-driven Optimization of Complex Systems (DOCS) (2023)

View Google Scholar profile

Patent

  • Training method, topic prediction method and device for federated variational autoencoder topic model. Patent number: 202310826329; published January 31, 2024. Student first author.

Research Experience

  • 2026.07 – Present: AI-RAN: CPU–GPU Acceleration and PHY Scheduling, Yonsei University
    • Build a CPU–GPU heterogeneous computing platform using NVIDIA DGX Spark (GB10) and OCUDU 5G gNB, with PDSCH/PUSCH PHY processing and performance evaluation across CPU/GPU execution configurations.
    • Analyze latency, throughput capacity, and resource overhead across worker configurations, LDPC code block sizes, and PHY workloads; investigate Unified Memory data sharing, Static Routing, and Hybrid CPU–GPU PHY Scheduling.
    • Plan technical exchanges during AI-RAN Summit activities (November 9–10, Seoul) and an initial demo at the AI-RAN Alliance All Member Meeting (November 11–12, Yonsei University).
  • 2025.09 – Present: TriBLoRA: Parameter-Efficient Fine-Tuning (Submitted), Yonsei University
    • Propose a tri-factor blockwise low-rank adaptation method that separates public and private information through shared outer factors and interaction-specific middle factors, enhancing LoRA’s expressive capacity.
    • Achieve intra-layer rank diversity through non-uniform parameter allocation within layers under a fixed parameter budget, supported by theoretical analysis and experiments.
  • 2025.09 – 2026.07: AI-RAN, Yonsei University
    • Explore GPU-accelerated base station architectures and offloading parts of gNB computation from CPUs to GPUs.
    • Investigated and implemented replacement of the gNB physical layer (L1) on the OCUDU platform using NVIDIA Aerial.
  • 2024.09 – 2025.09: Federated Learning with Heterogeneous Clients, Yonsei University
    • Studied heterogeneous environments with diverse client devices, asynchronous training, and varying model capacities.
    • Analyzed how system and model heterogeneity introduce skewed updates and degrade training performance.
    • Found that adaptive model capacity improves the effectiveness of contributions from weak clients.
  • 2022.09 – 2024.06: Communication Optimization in Federated Learning, Beijing University of Posts and Telecommunications
    • Worked on a sub-project supported by the National Natural Science Foundation of China (62192784, U22B2038, 62172056).
    • Studied communication-efficient federated learning with a focus on asynchronous training and model pruning.
    • Explored federated topic modeling and its applications in decentralized learning scenarios.
  • 2022.03 – 2022.09: Shared Mobility Evaluation and Sentiment Analysis, BUPT–Didi Global
    • Conducted topic analysis on customer reviews for shared mobility platforms.
    • Analyzed taxi transfer feedback and user sentiment trends.

Skills

  • Programming and tools: PyTorch, federated learning systems, model pruning, LLM fine-tuning, Docker.
  • Research areas: Efficient Machine Learning, Federated Learning, Parameter-Efficient Fine-tuning, AI-RAN.
  • Languages: CET-4: 454; CET-6: 434; IELTS: 6.0.

Honors & Awards

  • 2024: Outstanding Graduate of Beijing, Provincial Level.
  • 2021–2023: First Class Graduate Scholarship (3 times), School Level.
  • 2022–2023: Outstanding Graduate Cadre (2 times), School Level.
  • 2021: CQUPT Honours Bachelor’s Degree, School Level.
  • 2021: Chongqing Advanced Individual in Enhancing Innovation Capability, Provincial Level.
  • 2020: Guo Changbo Scholarship, School Level.
  • 2020: 1st Prize, May Day Mathematical Modeling Competition, Provincial Level.
  • 2020: 2nd Prize, MathorCup Mathematical Modeling Challenge, Provincial Level.
  • 2018–2019: National Inspirational Scholarship (2 times), Provincial Level.

Services

  • 2026: ICML 2026 Silver Reviewer.
  • 2021.09 – 2024.06: Class Monitor.
  • 2023.09.27 – 2023.10.08: Volunteer, China Open Tennis Championship.

Contact