However, developing models that can process weight-space objects is challenging due to their high dimensional nature.

In this work, drawing insights from world models, we propose to instead learn a general representation that can adapt to be invariant or equivariant to different transformations by paying attention tocontext-- a memory module that tracks task-specificstates,actions,andfuturestates.

We integrate this model in a planning loop, where conditioning and classifier guidance let us softly break the symmetry for specific tasks as needed.

It is common that these data themselves have inherently complicated 3D geometric structures.

It is common that these data themselves have inherently complicated 3D geometric structures.