Muge Zhang

Muge Zhang

he/him

Working on multilingual NLP, retrieval, and multimodal medical reasoning.

Curriculum Vitae

Education

Aug 2025 – May 2030 (expected)
Ph.D. in Computer Science and Engineering The Ohio State University, Columbus, OH Advisor: Sachin Kumar · GPA 4.00/4.00 Research: multilingual NLP, retrieval, and multimodal medical reasoning Coursework: Machine Learning, Speech & Language Processing, Data Mining, Data Visualization, Artificial Intelligence, Cybersecurity, Software Startups
Dec 2024
M.S. in Applied Computer Science Fairleigh Dickinson University GPA 3.92/4.00
June 2021
B.A. King's College London Upper Second-Class Honours

Awards

2025 – 2026
Graduate School Fellowship The Ohio State University · awarded January 2025, supporting the first year of the PhD

Research experience

2025 – present
Graduate Researcher The Ohio State University · Advisor: Sachin Kumar
Romanization for multilingual LM pretrainingEMNLP 2026
  • Ran a controlled study of input representations (orthographic text, IPA, Uroman romanization) for multilingual pretraining from scratch across 8 languages in 4 typological pairs. Romanization won across model scales and evaluation settings, and the benefit requires pretraining from scratch.
  • Built the full pretraining stack: modded-nanoGPT with custom BPE tokenizers and SLURM-based multi-node distributed training on H100 clusters at the Ohio Supercomputer Center.
Cross-lingual reasoning-intensive retrievalIn preparation
  • Showed that retrievers stable on multilingual benchmarks such as MMTEB collapse on cross-lingual reasoning-intensive queries, degrading 28 to 35% on the hardest task and language combinations.
  • Isolated the cause with same-backbone comparisons and alignment probes: the contrastive training recipe, not the backbone or data language, determines whether cross-lingual alignment survives.
Temporal reasoning over chest X-raysOngoing
  • Building a causal temporal-reasoning benchmark from MIMIC-CXR and MIMIC-IV with clinical collaborators: given a prior and recent X-ray pair plus long EHR context, identify the medical event explaining the radiographic change.
  • Training multimodal reasoners with SFT warmup followed by GRPO-based RL with evidence-grounded rewards.
2023 – 2025
Research Assistant Fairleigh Dickinson University · Advisors: Jeeho Ryoo and Wenyun Dai
  • Extended Med-DDPM conditional diffusion to synthesize Alzheimer's-specific 3D structural MRIs; hybrid real and synthetic training beat real-only segmentation baselines (Dice 0.7244). Published at IEEE MIPR 2025.
  • Profiled microarchitectural bottlenecks in ML and GNN workloads; optimized MapReduce hierarchical clustering (9 to 82% faster at over 90% accuracy). Four IEEE conference papers.

Publications

  1. M. Zhang, A. Jencks, K. Badikela, Y. Tsvetkov, S. Kumar. One Form to Transfer Them All: Pretraining Multilingual Language Models Beyond Native Orthography. EMNLP 2026, 2026.
  2. D. Lee, M. Zhang, Y. Yu, A. Manne, S. Koesters, F. Wen, B. Buchanan, L. Villagomez, O. Moninuola, J. Lim, K. Tobin, A. Srisuwananukorn, P. Zhang, S. Kumar. When Cases Get Rare: A Retrieval Benchmark for Off-Guideline Clinical Question Answering. arXiv (Preprint), 2026.
  3. M. Zhang, M. A. Khaliq, B. K. Lee, J. Ryoo. Synthetic Magnetic Resonance Imaging Generation for the Diagnosis of Alzheimer's Disease using Machine Learning. MIPR 2025, 2025.
  4. C. Davis, P. Stockton, M. Zhang, J. Ryoo, E. John. Microarchitectural Characterization of LightGCN and ExpressGNN and Architectural Implications. HiPEAC 2025 (Poster), 2025.
  5. W. Dai, M. Zhang. Hierarchical Agglomerative Clustering Optimization for Massive Data. ScalCom 2024, 2024.
  6. W. Dai, M. Zhang. A High Performance AI-Powered Cache Mechanism for IoT Devices. CyberC 2024, 2024.
  7. M. Zhang, D. Y. Lee, V. Janarthanan, J. Ryoo. Microarchitectural Analysis of Pre-Processing Stage in Machine Learning Workloads. ACAI 2024, 2024.

Full list on Google Scholar.

Work experience

Aug 2026 – present
Graduate Teaching Associate Department of Computer Science and Engineering, The Ohio State University Autumn 2026: CSE 3521 Introduction to Artificial Intelligence; CSE 5052 AI for Non-Majors
Sep 2024 – Dec 2024
Backend Engineer Intern Binance US · Remote
  • Developed and maintained internal backend tooling in Python in a high-security environment.
May 2021 – Aug 2023
Software Engineer Wefind AI · Beijing, China
  • Built a machine-translation workflow (Python, OpenAI API, spaCy and NLTK), improving accuracy by 30% and cutting manual post-editing by 50%.
  • Architected microservices (Flask, Docker, Kubernetes on AWS) serving over 10,000 daily requests at 99.9% uptime; added Redis caching to cut latency by 40%.

Skills and service

ML & systems
Python, C/C++, CUDA; PyTorch, HuggingFace Transformers, FlexAttention; distributed training (SLURM, multi-node); Docker, Kubernetes, AWS, Redis, Linux
Service
Reviewer, EMNLP 2026 System Demonstrations