Efficient Diffusion Models for Symmetric Manifolds
This work addresses the computational bottleneck for researchers and practitioners using diffusion models on symmetric manifolds, offering a significant speed improvement over existing methods.
The authors tackled the problem of inefficient diffusion models on symmetric manifolds by introducing a framework that uses a spatially-varying covariance and a projection of Euclidean Brownian motion to bypass heat kernel computations, resulting in training steps with O(1) gradient evaluations and nearly-linear-in-d arithmetic operations (O(d^{1.19})), outperforming prior methods in speed and sample quality on synthetic datasets.
We introduce a framework for designing efficient diffusion models for $d$-dimensional symmetric-space Riemannian manifolds, including the torus, sphere, special orthogonal group and unitary group. Existing manifold diffusion models often depend on heat kernels, which lack closed-form expressions and require either $d$ gradient evaluations or exponential-in-$d$ arithmetic operations per training step. We introduce a new diffusion model for symmetric manifolds with a spatially-varying covariance, allowing us to leverage a projection of Euclidean Brownian motion to bypass heat kernel computations. Our training algorithm minimizes a novel efficient objective derived via Ito's Lemma, allowing each step to run in $O(1)$ gradient evaluations and nearly-linear-in-$d$ ($O(d^{1.19})$) arithmetic operations, reducing the gap between diffusions on symmetric manifolds and Euclidean space. Manifold symmetries ensure the diffusion satisfies an "average-case" Lipschitz condition, enabling accurate and efficient sample generation. Empirically, our model outperforms prior methods in training speed and improves sample quality on synthetic datasets on the torus, special orthogonal group, and unitary group.