Computer Science Ph.D. student at Florida State University, advised by Prof. Guang Wang. Previously B.S. in Big Data Management and Application, Peking University.
🌐 Homepage · 📄 CV · 🎓 Google Scholar · DBLP · ORCID · LinkedIn · ✉️ dahai.yu@fsu.edu
I build trustworthy machine learning systems for the physical world — uncertainty quantification for spatiotemporal prediction, generative models for mobility and energy data, and decision pipelines that stay reliable when the downstream stakes are high.
- Uncertainty-aware spatiotemporal prediction — graph neural networks and selective state space models that report calibrated uncertainty alongside their point predictions.
- Generative models for urban and energy data — diffusion models that synthesize or repair mobility traces, human activity, and utility readings.
- Uncertainty quantification for LLM reasoning — estimating when a fluent reasoning trace should be trusted, via answer re-elicitation and symbolic verification.
- From prediction to decisions — predict-then-optimize pipelines where the uncertainty estimate actually changes the allocation.
Published at AAAI, IJCAI, ACM SIGKDD, ACM SIGSPATIAL, and ACM IMWUT (UbiComp).
| Venue | Paper | Code |
|---|---|---|
| AAAI 2026 | TrustEnergy: A Unified Framework for Accurate and Reliable User-level Energy Usage Prediction | TrustEnergy |
| IJCAI 2026 | HealthMamba: An Uncertainty-aware Spatiotemporal Graph State Space Model for Healthcare Facility Visit Prediction | HealthMamba |
| KDD 2026 | EnergyMamba: An Uncertainty-Aware Graph-Enhanced Selective State Space Model for Energy Consumption Prediction | EnergyMamba |
| IMWUT 2026 | SynHAT: A Two-stage Coarse-to-Fine Diffusion Framework for Synthesizing Human Activity Traces | — |
| SIGSPATIAL 2025 | UQGNN: Uncertainty Quantification of Graph Neural Networks for Multivariate Spatiotemporal Prediction | UQGNN |
| IJCAI 2025 | Uncertainty-aware Predict-Then-Optimize Framework for Equitable Post-Disaster Power Restoration | — |
| arXiv | TrAC: Trace-Conditioned Answer Consistency for Efficient UQ in LLMs | TrAC |
| arXiv | SymboUQ: Symbolic Uncertainty Quantification for Spatial Reasoning in LLMs | SymboUQ |
Full list → ufodestiny.github.io/publications
Also here: POPST, a unified benchmarking framework for spatiotemporal forecasting with conformal quantile regression, and OD-ZeroCal for zero-aware calibrated origin–destination demand prediction.
Outside research I write things for games I play — EU5-Patcher (achievements outside ironman for Europa Universalis V), UFO-Bannerlord, and CK3 Smaller Map.
Always happy to talk about spatiotemporal foundation models, calibration, or urban data.