I am a Ph.D. candidate in Computer Science at Case Western Reserve University, advised by Prof. Yinghui Wu.

My research focuses on graph machine learning, explainable AI (XAI), and temporal graph neural networks. I am particularly interested in developing counterfactual and interpretable explanation methods for dynamic graph models.

My recent work includes TemGX, a training-free counterfactual explanation framework for temporal graph models, accepted at ICLR 2026. I have also worked on multimodal learning systems that integrate image segmentation, knowledge graphs, and graph neural networks for scientific applications.

In addition, I have developed large-scale benchmarking systems for materials image segmentation, achieving strong performance across SEM, AFM, and XCT datasets, and building pipelines for model evaluation and interpretability.

News

I am happy to share that our paper CATGX: Causal-Aware Temporal Graph Explanation via Scalable Motif Sampling and Adjustment has been accepted to KDD 2026 Research Track.

I am happy to share that our paper Training-free Counterfactual Explanation for Temporal Graph Model Inference has been accepted to ICLR 2026.
Our work on materials segmentation and multimodal learning has been accepted to ICLR 2026 AI4Mat Workshop.
Our paper Graph Compression for Interpretable Graph Neural Network Inference at Scale has been accepted to PVLDB 2025.

Publications

Mingjian Lu, Haolai Che, Yangxin Fan, Qu Liu, Fei Shao, Tingjian Ge, Xusheng Xiao, Yinghui Wu.

Training-free Counterfactual Explanation for Temporal Graph Model Inference.

ICLR 2026

Mingjian Lu, Pawan Kumar Tripathi, Mark Shteyn, Debargha Ganguly, Roger H. French, Vipin Chaudhary, Yinghui Wu.

Context Determines Optimal Architecture in Materials Segmentation.

ICLR 2026 AI4Mat Workshop

Yangxin Fan, Haolai Che, Mingjian Lu, Yinghui Wu.

Graph Compression for Interpretable Graph Neural Network Inference at Scale.

PVLDB 2025

Mingjian Lu, Nalin Venkat, Augustino J., et al.

Image Processing Pipeline for Fluoroelastomer Crystallite Detection in Atomic Force Microscopy Images.

Integrating Materials and Manufacturing Innovation, 2023

Hernandez, K. J., Ciardi, T. G., Yamamoto, R., Mingjian Lu, et al.

L-PBF High-Throughput Data Pipeline Approach for Multi-Modal Integration.

Integrating Materials and Manufacturing Innovation, 2024

Nalin Venkat, S., Ciardi, T.G., Mingjian Lu, et al.

A General Materials Data Science Framework for Quantitative 2D Analysis of Particle Growth from Image Sequences.

Integrating Materials and Manufacturing Innovation, 2024

Mingjian Lu, Chen, Q.Q., Chen, Y.J., Sun, W.J.

Micromanagement in StarCraft Game AI: A Case Study.

Procedia Computer Science (IIKI), 2019

Research & Projects

Temporal Graph Explanation

Developed TemGX, a training-free counterfactual explanation framework for temporal graph neural networks (TGNNs), accepted at ICLR 2026.
Designed influence and resistance-distance based metrics to quantify temporal impact and identify key substructures driving model predictions.
Achieved strong performance in fidelity and sparsity on real-world datasets, including METR-LA, PEMS-BAY, and Bitcoin transaction networks.
Enabled interpretable analysis of dynamic systems such as traffic flow, financial transactions, and cyber-attack propagation.

Multimodal Scientific AI & Image Segmentation

Built a benchmarking framework for materials image segmentation across multi-modal datasets (SEM, AFM, XCT, Optical).
Designed and evaluated high-performance encoder–decoder architectures (e.g., UNet variants, DeepLabv3+, Transformer-based models), achieving IoU > 0.90 in high-contrast scenarios.
Integrated quality control pipelines with OOD detection and interpretability tools (e.g., counterfactual heatmaps) to improve model reliability.
Released standardized datasets and evaluation tools to support reproducible research in materials science.

Knowledge Graphs & Scientific Discovery

Constructed knowledge graphs and scene graphs from image segmentation outputs to represent defect structures and spatial relationships.
Integrated graph learning with domain-specific features to support fatigue analysis and materials degradation modeling.
Developed pipelines that connect visual data, graph representations, and machine learning models for scientific interpretation.

Visual-Graph Multimodal Learning (Ongoing)

Developing a multimodal framework combining SEM image features and graph neural networks for fatigue life prediction.
Designed dual GraphSAGE-based encoders to model different physical zones (e.g., fatigue vs overload regions).
Exploring counterfactual explanations at both visual and graph levels to identify decision-critical microstructural features.
Leveraging LLMs to translate model outputs into interpretable domain knowledge.

Industry Experience

Research Assistant, Microsoft, Beijing, China

Jul 2018 – Sep 2018

Assisted in implementing data-driven strategies for enterprise search engine optimization within the Belt and Road initiative.
Analyzed conversion rates and click-through rates (CTR) for Bing Ads campaigns to improve targeting and performance.
Conducted data analysis to optimize campaign effectiveness and support decision-making.

Teaching

Teaching Assistant, Case Western Reserve University

CSDS 433: Database Systems Fall 2024, Spring 2025

CSDS 221: Full Stack Web Development Fall 2025

Skillsets

Technical Expertise

Research Areas:

Graph Machine Learning, Explainable AI (XAI), Temporal Graph Neural Networks, Knowledge Graphs, Multimodal Learning
Machine Learning & Frameworks:

PyTorch, TensorFlow, Scikit-learn
Programming:

Python, Java, SQL, C++
Systems & Data:

Large-scale machine learning pipelines, scientific image analysis, graph data processing

Page updated

Report abuse