CAPTAIN.
Forecasts with a fuller
picture of uncertainty.
A framework that models uncertainty within sources, across sources, and over time, bringing conformal prediction to multi-source time-series forecasting.
Shuaicheng Zhang Researcher & AI engineer
Exploring an
open world.
I build AI systems that connect
structure, context, and evidence.
AI Engineer at LinkedIn.
Ph.D. in Computer Science, Virginia Tech.
Curiosity. Evidence. Impact.Research interests
Research interests
From better models.
To useful systems.
My work connects agentic reasoning, graph learning, and reliable AI. I’m interested in how models use structure and context to make sense of an open, changing world.
I completed my Ph.D. at Virginia Tech, advised by Dr. Dawei Zhou at the Sanghani Center for AI and Data Analytics. At LinkedIn, I bring that research depth to modeling and systems at scale.
Reasoning → action
Intelligence that takes the next step.
Agents that plan, use tools, and coordinate reasoning across complex evidence. Connecting model capabilities to the decisions a working system needs to make.
Selected publications
Ideas, put to the test.
AgentPDM.
Reasoning from signals.
Decisions with context.
Context-driven agentic reasoning for predictive maintenance, connecting sensor evidence with physics and domain knowledge to produce structured diagnoses and maintenance recommendations.
KDD 2025 · Best Paper · Datasets & Benchmarks
When Heterophily Meets Heterogeneity
All publications 2022—2026
2026
-
CAPTAIN: Conformal-Prediction-Based Multi-Source Time-Series Forecasting
Transactions on Machine Learning Research · 2026
Overview
Uncertainty quantification is critical for real-world forecasting applications such as predictive maintenance, patient health monitoring, and environmental sensing, where decisions must account for confidence levels. Multi-source time-series forecasting introduces additional complexity due to inter-source interactions and temporal dependencies, which existing methods struggle to capture within a unified probabilistic framework. CAPTAIN is a two-stage framework that uses Normal Inverse Gamma distributions and a meta-source to model source-specific and inter-source uncertainty, then applies temporal copulas to provide robust, theoretically valid uncertainty coverage. Across five diverse datasets, CAPTAIN achieves valid coverage of at least 90 percent on every benchmark while producing competitive interval widths.
-
AgentPDM: Context-Driven Agentic Reasoning for Predictive Maintenance
Proceedings of the 35th ACM International Conference on Information and Knowledge Management (CIKM) · 2026Accepted
Overview
AgentPDM is a context-driven agentic framework for predictive maintenance that combines data-driven time- and frequency-domain signal evidence with physics and domain knowledge. A context-aware decision agent reconciles this evidence to produce structured, actionable, and auditable diagnoses covering fault type, stage and severity, key evidence, uncertainty, and maintenance recommendations.
2025
-
HeroFilter: Adaptive Spectral Graph Filter for Varying Heterophilic Relations
Proceedings of the 39th Conference on Neural Information Processing Systems (NeurIPS) · 2025
-
When Heterophily Meets Heterogeneity: Challenges and a New Large-Scale Graph Benchmark
Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · 2025
-
MentorPDM: Learning Data-Driven Curriculum for Multi-Modal Predictive Maintenance
Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 1 · 2025
-
Scientific Hypothesis Generation and Validation: Methods, Datasets, and Future Directions
CoRR · 2025arXiv preprint
2024
-
UnifiedGT: Towards a Universal Framework of Transformers in Large‑Scale Graph Learning
IEEE International Conference on Big Data (BigData) · 2024
2023
-
Personalized Federated Learning under Mixture of Distributions
Proceedings of the 40th International Conference on Machine Learning (ICML) · 2023
-
TGEditor: Task‑Guided Graph Editing for Augmenting Temporal Financial Transaction Networks
Proceedings of the Fourth ACM International Conference on Artificial Intelligence and Finance (ICAIF) · 2023
2022
-
Extracting Temporal Event Relation with Syntax‑guided Graph Transformer
Findings of the Association for Computational Linguistics: NAACL · 2022
A new chapter
The latest.
And what comes next.
ResearchAgentPDM is
Explore the work AgentPDM is
heading to CIKM.
Our work on context-driven agentic reasoning for predictive maintenance has been accepted at CIKM 2026.
PublicationCAPTAIN.
Read the paper CAPTAIN.
Published in TMLR.
Bringing source interactions and temporal dependence together for uncertainty-aware forecasting.
MilestonePh.D. complete.
The journey so far Ph.D. complete.
Curiosity continues.
Passed my final defense at Virginia Tech. Grateful for the people, questions, and collaborations along the way.
Earlier moments
Presented an agentic co-design system for hypothesis generation in SNN design at Virginia Tech.
Gave an invited talk at Microsoft Research on open-world graph learning.
Started an ORNL internship on GraphRAG for scientific discovery.
Presented a NeurIPS 2025 poster in San Diego.
One paper accepted at NeurIPS 2025.
KDD Best Paper Award, Datasets & Benchmarks track.
Joined Microsoft Research to work on GNNs for database foundation models.
Paper accepted at KDD Datasets & Benchmarks. Recognized as an Excellent Reviewer, Top 25%.
Outstanding Reviewer, Top 10%, KDD 2025 August Track.
One paper accepted at IEEE BigData 2024.
One paper accepted at KDD 2025.
One paper accepted at ICAIF 2023 with an oral presentation.
Presented at ICML 2023 in Honolulu.
Joined MIT–IBM Watson AI Lab. One paper accepted at ICML.
Received an Amazon ML Day travel grant.
Selected as a CCI Cyber Innovation Scholar.
Attended CIKM and received an NSF travel award.
Served as Publicity Chair of the first TrustLOG workshop at CIKM.
One paper accepted at Findings of NAACL.
Experience
Different teams.
A wider lens.
Each chapter has shaped how I think about research, collaboration, and building things that matter.
View my CVAI Engineer
Modeling · Intelligence at scale
Oak Ridge National Laboratory
Research Intern
GraphRAG for scientific discovery
Microsoft Research
Research Intern
GNNs for database foundation models
MIT–IBM Watson AI Lab
Research Intern
Deep graph generative models
Deloitte AI
NLP Research Intern
Natural language processing
Hundsun Technologies
Software Engineer Intern
Trust systems engineeringA life outside the lab
My favorite collaborators.
Jojo and Joy. Experts in curiosity, comfort,
and perfectly timed keyboard interruptions.
A few more moments
Around the world
Ideas travel.
A world of curious minds.
Thanks for stopping by, wherever you are.
Our global community MapMyVisitors
The next conversation
Good ideas start
with a hello.
zshuai8@vt.edu Open the PDF directly if the preview is unavailable.