Recent Publications

Embeds physics in graph networks for stable long-horizon multiphysics simulation.

Prunes symbolic regressors with physics residuals and sensitivity, yielding cleaner equations.

Recasts dynamic multiobjective optimization as a prompt-guided time-series prediction problem and shows that large language models can generate competitive solutions.

Infers interpretable intrinsic dynamics from video with bilevel optimization.

Generates 4D scenes that stay physically consistent across time and materials.

Uses T-NNGP to produce interpretable multi-physics PDE solutions.

Makes neural operators respect interface physics in multiphase flow simulation.

Exposes a universal black-box weakness in 3DGS watermarking systems.

Produces interpretable PDE solutions rather than opaque black-box approximations.

Uses lower-confidence bounds to pick transferable solutions in multitask optimization.

Turns fairness diagnostics into actionable explanations for graph neural networks.

Learns from multiple graph views to reduce label dependence in subgraph learning.

Allocates computation dynamically to make evolutionary bilevel optimization practical.

Turns image-to-text attacks into a decision-based black-box method with no gradients.