Research Highlights

Our group develops machine learning methods for high-dimensional, spatio-temporal, and sequential data from engineered and societal systems. Applications include manufacturing, energy and hydrogen infrastructure, transportation, aviation safety, disaster response, and medical imaging. Below are four themes of our recent work, each with representative papers. The Google Scholar profile and my CV have the full list.

Reinforcement learning for prognostics and decision making

We develop reinforcement learning methods, including distributional and flow-based value learning, for prognostics and for optimizing operating conditions from real-time degradation signals.

“Path-Coupled Bellman Flows for Distributional Reinforcement Learning”, Xu, Zou, Yang, Yan. ICML 2026. arXiv:2605.08253
“Retimed Bellman Flows: Escaping the Impossible Triangle of Velocity Bootstrapping”, Xu, Chen, Yan. arXiv preprint, 2026. arXiv:2609.32828
“General Value Functions for Remaining Useful Life and Failure-Mode Prediction”, Yan, Sarabi, Zou, Xu. arXiv preprint, 2026. arXiv:2607.22268
“Optimizing Multiple Condition Policy based on Real-time Degradation Signals via Model-based Reinforcement Learning”, Kang, Zhao, Feng, Yan. Journal of Quality Technology, 2026.
“Partially observable Markov decision process framework for operating condition optimization using real-time degradation signals”, Xu, Kang, Zhao, Yan, Ju. Journal of Quality Technology, 2025.

Generative models for anomaly detection and industrial data

We use flow-matching and diffusion models to detect and localize anomalies without labeled anomalies and to build surrogate models of complex systems, and we develop shape priors for image segmentation.

“Flow Mismatching: Unsupervised Anomaly Detection via Velocity Discrepancies in Flow Matching Models”, Chen, Moradi, Paynabar, Yan. NeurIPS 2026 (accepted). arXiv:2605.23070
“A Single Image Is All You Need: Zero-Shot Anomaly Localization Without Training Data”, Moradi, Chen, Yan, Paynabar. arXiv preprint, 2025. arXiv:2509.18354
“D-Convexity: A Unified Differentiable Convex Shape Prior via Quasi-Concavity for Data-driven Image Segmentation”, Chen, Yan. CVPR 2026 (Highlight, top 2%). Project page.
“Diffusion-Based Surrogate Modeling and Multi-Fidelity Calibration”, Shi, Yan, Guo, Al Kontar. IEEE Transactions on Automation Science and Engineering, 2025.
“Multi-modal Generative Modeling of Event Sequences and Time Series for Solar PV Systems”, Huang, Xu, Liu, Yan. IEEE CASE 2025 (Best Conference Paper Award).

Probabilistic, causal, and physics-informed learning

We build models that quantify uncertainty, learn causal structure, and incorporate domain knowledge and physics into neural networks.

“Structurally Separated DAG Learning with Multi-Scale Normalized Closure”, Feng, Yan. NeurIPS 2026 (accepted).
“Bayesian Entropy Neural Networks for physics-aware prediction”, Rathnakumar, Lu, Huang, Yan, Liu. Reliability Engineering & System Safety, 2026.
“Aleatoric-Aware Posterior Regularized Bayesian Neural Networks for Incorporating Domain Knowledge: Addressing Epistemic and Aleatoric Uncertainties”, Huang, Li, Setzer, Lure, Ju, Yan. INFORMS Journal on Data Science (accepted).
“Probabilistic Kolmogorov-Arnold Networks via sparsified deep Gaussian processes with additive kernels”, Zou, Yan. IEEE CASE 2025.
“Personalized Federated Hierarchical Gaussian Processes for Privacy-Preserving Modeling of Heterogeneous Distributed Systems”, Xie, Yan. arXiv preprint, 2026. arXiv:2609.19337

Tensor and spatio-temporal analytics

We develop tensor, sparse, and spatio-temporal methods for monitoring and modeling high-dimensional images, profiles, and data streams. Our work in this area has received the IEEE CASE Best Paper Award (2020), the ASQ Brumbaugh Award (2019), the IISE Transactions Best Paper Award (2019), and INFORMS Data Mining Section best paper awards (2017 and 2018). Earlier work on smooth-sparse decomposition for image and data-stream monitoring received best paper awards from INFORMS (Data Mining Section, 2014; Quality, Statistics and Reliability Section, 2015) and from the IISE Quality Control and Reliability Engineering Division (2015).

“Personalized Tucker decomposition: Modeling commonality and peculiarity on tensor data”, Hu, Shi, Al Kontar, Yan. Technometrics, 2025.
“Low-rank robust subspace tensor clustering for metro passenger flow modeling”, Sergin, Hu, Li, Zhang, Tsung, Yan. INFORMS Journal on Data Science, 4(1), 33–50, 2025.

Application areas

Energy and hydrogen infrastructure: Bayesian optimization for reactor design; predictive maintenance of photovoltaic plants; risk assessment for hydrogen pipelines.
Aviation safety: event extraction from accident reports; prognostics for air transportation systems.
Disaster response: adaptive sampling for rapid damage assessment after disasters.
Medical imaging and public health: 3D dental CBCT segmentation and lesion detection; detection of HIV transmission clusters.