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

Any explanation that faithfully explains this type of model needs to be in agreement with this invariance property.

A.1.1 V ariance Preserving SDE as a potential function We can re-express the widely used V ariance Preserving (VP-SDE) (DDPM): dY

Generative diffusion models have recently emerged as a leading approach for generating high-dimensional data.

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

D.2 Countries Hyperparameters are summarized in table 6. We ran all experiments on a single CPU (Apple M2). 15 optimizer AdamW learning rate 0.0003 learning rate schedule cosine training epochs 100 weight decay 0.00001 batch size 4 embedding dimensions 10 embedding initialization one-hot, fixed neural networks LeNet5 max search depth / Table 5: Hyperparameters for the MNIST -addition experiments.