Hello, I'm
Ph.D. Candidate in Electrical Engineering
Case Western Reserve University
I am a Ph.D. candidate in Electrical Engineering at Case Western Reserve University, expected to graduate in August 2027. My research aims to build an end-to-end generalizable AI framework for industrial automation and robotic systems in three connected stages:
Scalable and interpretable representations from large-scale data.
Self-supervised learning, interpretable AI
ML models across edge–cloud architectures.
Edge AI, IIoT
Continuously, as unseen data and conditions emerge after deployment.
Continual learning, federated learning
I worked on site at General Motors for six months, validating my pretext-training framework for inline laser-welding monitoring and delivering it to internal customers in manufacturing. The framework is now being adapted to more manufacturing processes. I am also working to commercialize EdgeSense, an edge–cloud AI system for industrial monitoring developed from my research.
Visiting Scientist of AI, Manufacturing R&D · On-site
Entrepreneurial Lead · Remote
Visiting Student, Research Collaboration · Remote
Undergraduate Research · On-site
Ongoing Ph.D. research
Developing an interpretable training framework that helps the model disentangle concepts from the data, so it can identify and protect true knowledge while adapting to out-of-distribution data. Expected to generalize to both time-series and image data.
NeurIPS 2026 Main Track
Investigated the internal mechanisms of six VLA models (80M–7B parameters) with interpretability methods across 394,000+ rollout episodes on four benchmarks. Showed that visual representations strongly influence action generation and that learned motor programs are tied to scene coordinates rather than task semantics. Released Action Atlas, an interactive explorer of the learned representations.
IEEE Internet of Things Journal (in revision); U.S. provisional patent
Designed an edge-computing node for real-time condition monitoring of rotating machinery. Implemented continual self-supervised learning entirely on the microcontroller within 3 MB of memory, without cloud-based model updates.
J. Manufacturing Systems 2025; J. Manufacturing Processes 2026
Developed multi-level contrastive learning frameworks (MoCo- and DINO-style) to learn from unlabeled factory data for both welding images and multichannel photodiode signals. Combining labeled lab data with unlabeled plant data improved transfer to defect classification and weld-quality prediction.
Manufacturing Letters 2025
Co-developed a transfer-learning-enhanced Transformer that estimates in-process weld quantities from available signals, serving as a digital twin that reduces the need for direct measurement.
* equal contribution
"How Should the Edge Adapt: Designing Reliable On-Device Continual Self-Supervised Learning for Condition Monitoring"
IEEE Internet of Things Journal, in revision.
"Toward Foundational Manufacturing Surrogates: Physics-Informed Machine Learning for Welding and Additive Manufacturing Across Geometries, Materials, and Processes"
Manuscript for Journal of Manufacturing Systems, in revision.
"Harnessing Unlabeled Plant Data and Labeled Lab Data for Enhanced Quality Prediction in Laser Welding"
Journal of Manufacturing Processes, 157, pp. 1015–1034.
"Unleashing the Power of Unlabeled Plant Data: A Hierarchical Contrastive Learning Framework for Dynamic Manufacturing Process Monitoring"
Journal of Manufacturing Systems, 83, pp. 483–493.
"Transfer Learning-Enhanced Transformer for Virtual Process Sensing in Resistance Spot Welding"
Manufacturing Letters, 45, pp. 13–16.
"Applicable and Generalizable Machine Learning for Intelligent Welding in Automotive Manufacturing"
Welding in the World, pp. 1–36.
"Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models"
Advances in Neural Information Processing Systems (NeurIPS), Main Track. arXiv:2603.19233. Earlier version: ICLR 2026 Workshop on Multimodal Intelligence.
"Efficient and Generalizable Machine Learning for Inline Defect Detection in Battery Laser Welding"
International Conference on Precision Engineering (ICPE). Sendai, Japan.
"A Deployable Edge Computing Solution for Machine Condition Monitoring"
IEEE International Instrumentation and Measurement Technology Conference (I2MTC). Glasgow, United Kingdom, pp. 1–6.
"Laser Welding Signal Analysis System and Method"
U.S. patent application, General Motors, filed May 2026.
"Edge Computing Models for Apparatus Monitoring"
U.S. Provisional Patent Application No. 63/955,650, Case Western Reserve University, filed Jan 2026.
"Transformer-Based Multi-Level Contrastive Learning for Multi-Scale Welding Signal Representation"
General Motors internal patent disclosure, in process.
Case Western Reserve University
Aug 2023 – Aug 2027 (expected)
University of Kentucky
2021 – 2023
Beijing University of Technology
2019 – 2023