Contrary to prior assumptions, we identify extraction of weights, not extraction of weight signs, as the critical bottleneck.

Using low-dimensional demonstrations (such as agent behavior) to guide the generation of high-dimensional outputs (policy parameters) is a challenging problem.

This approach allows predictions to be traced back to specific concept patterns that users can understand and potentially intervene on. However, existing CBMs' task predictors are not fully interpretable, preventing a thorough analysis and any form of

PaGoDA's pipeline can be applied directly in the latent space, adding compression alongside the pre-trained autoencoder in Latent Diffusion Models (e.g., Stable Diffusion).

We further design an effective mechanism to inject PH information into GP at both feature and topology levels, with a novel topology-preserving loss function.