Weisheng Tang is a Postdoctoral Research Associate in the Min H. Kao Department of Electrical Engineering and Computer Science. His connection to AICIP runs long: he first joined the lab as a visiting PhD student (2017–2019, supported by a Chinese Academy of Sciences CSC scholarship) before returning as a postdoc.
His current research spans the lab’s Koopman-theoretic training-dynamics line — he is a co-author of the 2026 cross-regime study of structured predictability in neural training — and its remote-sensing efforts, including Koopman-based transition detection in satellite imagery and self-supervised fine-tuning strategies for multi-scale overhead imagery (IGARSS 2024). He also works on machine learning for multimodal biomedical data.
Projects
Koopman Dynamics
Treating neural network training as a dynamical system — predicting weights epochs ahead to accelerate optimization, and stabilizing policy learning in RL.
IARPA SMART
Automating the analysis of Landsat, Sentinel, and WorldView imagery to detect and characterize large-scale change anywhere on Earth.
Publications
Measuring Structured Predictability in Neural Training Dynamics: A Cross-Regime Study
arXiv 2026
Koopman-Based Transition Detection in Satellite Imagery: Unveiling Construction Phase Dynamics Through Material Histogram Analysis
IEEE International Geoscience and Remote Sensing Symposium (IGARSS) 2024
Advancing Multi-Scale Remote Sensing Analysis Through Self-Supervised Learning Fine-Tuning Strategies
IEEE International Geoscience and Remote Sensing Symposium (IGARSS) 2024