Researchers have developed DyPSI, a novel framework for dynamic physics sensing that jointly generates future spatiotemporal fields and sensor trajectories. Unlike previous methods that assume static sensing or optimize placements independently, DyPSI formulates the problem as a joint generation process conditioned on historical observations. The system utilizes continuous Sensor Position Fields and a shared Functional Tucker representation, trained with a joint diffusion model. Experiments on turbulent flow, sea surface temperature, and 3D car aerodynamics demonstrate DyPSI's superior performance over static strategies, particularly under limited sensor budgets. AI
IMPACT This research could lead to more efficient and accurate data collection in dynamic environments, improving scientific modeling and simulation.
RANK_REASON The item is a research paper published on arXiv detailing a new framework for physics sensing. [lever_c_demoted from research: ic=1 ai=1.0]
- 3D car aerodynamics
- alphaXiv
- A-optimality orthogonal forward regression algorithm using branch and bound
- arXiv
- CatalyzeX
- DagsHub
- Functional Tucker
- Gaussian-process temporal kernel
- GLORYS12
- Gotit.pub
- Hugging Face
- ScienceCast
- sea surface temperature
- Sensor Position Fields
- turbulent flow
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
