Embedded Files
▼ Research Overview
Data-Efficient AI
It aims to leverage existing datasets effectively, especially under the constraints of limited or sparse data, as commonly found in real industrial scenarios.
[Top left] "Discovery of Design Insights": Extract physical patterns and design rules from given limited dataset to enable more informed and efficient design processes.
[Top middle] "Multi-Fidelity Data Fusion": Integrate different fidelity data to maximize predictive accuracy while minimizing computational cost.
[Top right] "Generative AI for SciML": Generate physically plausible designs or data samples to augment small datasets and expand the design space.
Physically-Robust AI
It aims to build trustworthy AI models by incorporating temporal dynamics, physics, and uncertainty awareness for reliable performance in real-world physics.
[Bottom left] "Spatio-Temporal Prediction": Predict how physical fields evolve over time and space to accurately capture dynamic system behaviors.
[Bottom middle] "Physics-Embedded AI": Embed governing physical laws directly into the AI model to ensure physically consistent and generalizable predictions.
[Bottom right] "Uncertainty-Aware Prediction": Quantify predictive uncertainty to enhance trust, support decision-making, and detect non-physical outcomes.
Page updated
Google Sites
Report abuse