Reproduce “Finite-Time Online Sample Complexity for Conditional Diffusion Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mudit Gaur, Prashant Trivedi, Sasidhar Kunapuli, Amrit Singh Bedi, Vaneet Aggarwal
The authors analyze conditional diffusion after finite online stochastic-gradient training, separating score-class approximation error from optimization error and propagating both through reverse sampling. Under stated approximation, tail, intrinsic-geometry, and uniform optimization-geometry conditions, capacity M = Θ̃(ε⁻⁽²⁺ᵈˣ′⁄ᵅˣ⁺ᵈʸ′⁄ᵅʸ⁾) and N = Θ̃(ε⁻²) fresh denoising samples suffice for expected total-variation error of order ε against the positive-time smoothed conditional law, not necessarily the original data law.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.