Hypothesis

Is a neural operator well trained? Is it sensitive to the boundary conditions or forcing conditions as the physical model?

Feature interactions

In many classification tasks, there are not trivial interactions between features, such that the importance of one feature much depends on the presence of another.

References

https://arxiv.org/abs/2609.36216 https://arxiv.org/abs/2601.23072

Butterfly effect in neural operators is missing. The sensitivity of the neural operator to small changes in the input is not well understood. This can lead to unexpected behavior and poor generalization.

https://arxiv.org/pdf/2608.25835

Blogs

https://rbcborealis.com/research-blogs/neural-odes/

Github for neural operators for PDEs

https://github.com/PHASE-MHD/PHASE