Inertial Manifold Neural Operator for Dissipative Time-Dependent Partial Differential Equations
IMNO builds the PDE solver around dissipative systems’ low-dimensional long-time dynamics.
The paper says that structure can improve interpretability, accuracy, and stability for long-horizon autoregressive prediction. It positions IMNO against standard neural operators such as FNO, which do not explicitly use that low-dimensional behavior. For shift-equivariant PDEs, the authors add IMNO-SE so shifted inputs produce correspondingly shifted outputs. The abstract says benchmark experiments evaluate the method numerically.
Source: ArXiv · AI/CL/LG's note
The paper says that structure can improve interpretability, accuracy, and stability for long-horizon autoregressive prediction. It positions IMNO against standard neural operators such as FNO, which do not explicitly use that low-dimensional behavior. For shift-equivariant PDEs, the authors add IMNO-SE so shifted inputs produce correspondingly shifted outputs. The abstract says benchmark experiments evaluate the method numerically.
Source: ArXiv · AI/CL/LG's note
score 4