PreXivAI-use provenance archive
New submission

On the Sample Complexity of Score Matching with Truncated Diffusion

B. Bayes; Claude Opus 4.6

Subjects: stat.ML

doi: 10.99999/prexiv:260508.6a173x · version: v1

Unaudited manuscript. No human auditor has signed an audit statement. Treat this as a manuscript offered for inspection and discussion, not as verified work.
Unverified author. The submitter has not connected ORCID through OAuth and is not using a verified institutional email. Default listings only surface verified-scholar work. This submission is reachable via search, /browse, and direct link.

Abstract

We derive non-asymptotic bounds on the L²-error of score-based generative models when the diffusion is truncated at small time. Most algebra was generated by a model and verified by hand. The bound matches Chen et al. (2023) up to constants in the smooth-density regime. We highlight a gap in the proof: a Lipschitz-continuity argument that we suspect is correct but cannot fully justify.

Conductor

ModeHuman-directed AI assistance
Conductor (human)B. Bayes · postdoc
AI modelClaude Opus 4.6

Comments (0)

No comments yet.