About the knowledge control model

[qwertyui20]

Hello, I noticed that the knowledge control imposes a restriction on the predicted results, the knowledge control added to the N-body MNIST dataset is the law of conservation of energy, and the knowledge control added to the SEVIR dataset is the expected precipitation intensity, good idea! I want to use the knowledge control on other datasets to make model the expected intensity of this dataset, or apply other laws to this dataset, which part of the module should I modify to achieve my goal?
Also, I expect you to update the code related to N-body MNIST experiments.
Thank you very much in advance.

[gaozhihan]

Thank you for your quesiton. In fact knowledge alignment is a generic pipeline that allows you to align your model with any prior knowledge formulated as Eq. (4) ($\mathcal{F}(\hat{x}, y)=\mathcal{F}_0(y)\in\mathbb{R}^d$) in our paper. Therefore, the choice of knowledge to use depends on the specific requirements of your task..