Yunfei Ma

Yunfei Ma

I enjoy building computational ML pipeline on large geospatial datasets

About Me

I’m a PhD student in Computational Science and Engineering at McMaster University, advised by Prof. Elkafi Hassini and Prof. Saiedeh Razavi.

My research interest is broadly in the convergence of geospatial data science, computational analytics, operations research, deep learning, and GenAI.

Currently, I’m affiliated with the Smart Freight Centre and working as a Research Scientist at Geotab under the Altitude team. In addition to my research, I am a sessional instructor for a statistics course at McMaster University.

Before that, I earned my B.S. in Mathematics (Applied) from the University of Illinois Urbana-Champaign.

Working Papers

  • Ma, Y., Foda, A., Hassini, E., & Mohamed, M. Classifying E-Commerce Vehicle Trajectories Using Vision-Based Representation Learning and LLM-Augmented Metadata Labeling.
EC Classification Method EC Classification Method
  • Ma, Y., Hassini, E., & Razavi, S. Impact of Bus Rapid Transit on Freight Movement: A Telematics Approach.
BRT Clusters BRT Emission BRT Results
  • Ma, Y., Hassini, E., & Razavi, S. Evaluation of Emission Trade-offs in Optimized Urban Truck Routes under Dynamic Traffic.
Eco-routing Trade-off

Publications

  • Ma, Y., Liu, C. A., Hassini, E., & Razavi, S. (2024). A Network-Based, Data-Driven Methodology for Identifying and Ranking Freight Bottlenecks. Data Science for Transportation, 6(3), 20. https://doi.org/10.1007/s42421-024-00107-z
FB Graph Labelling
  • Ma, Y., Amiri, A., Hassini, E., & Razavi, S. (2022). Transportation data visualization with a focus on freight: a literature review. In Transportation Planning and Technology (Vol. 45, Issue 4, pp. 358–401). Informa UK Limited. https://doi.org/10.1080/03081060.2022.2111430
Visual Method Visual Decision Support Tool

Patent

  • Ma, Y., & Liu, C. A. (2024). Systems and methods for identifying and ranking traffic bottlenecks. U.S. Patent Application No. US20240161606A1.
FB Patent 1 FB Patent 2 FB Patent 3 FB Patent 4

Projects

Project Description
MBC AI Dashboard LLM-driven pipeline for sermon analytics, featuring a data dashboard with NLP-based feature extraction and visualization toolkit.
WerewolfGPT Multi-agent LLM system simulating conversational deduction in Werewolf game, leveraging natural language reasoning.
TumorRxGPT LLM fine-tuned on EMRs for tumor phenotyping and treatment recommendation, integrating structured/unstructured data processing.
MosaicNFT Generator C++ optimized mosaic generator using kd-trees for spatial indexing and LUV color space for perceptual similarity matching.
TriCount Proof Protocol Interactive sum-check protocol implementation with prover/verifier for triangle counting in graphs, leveraging polynomial verification.

Essays