BWhat changed
Applied different noise levels across token subsets and matched mixed-input features to a cleaner internal target, combining representation prediction with standard flow loss without an externally pretrained representation teacher.
Scientific scope
Applied different noise levels across token subsets and matched mixed-input features to a cleaner internal target, combining representation prediction with standard flow loss without an externally pretrained representation teacher. Flow matching, denoising and self-supervised alignment have predecessors, including independently noised sequence tokens. Scoped delta is dual-noise internal feature prediction combined with flow training, not invention of arbitrary heterogeneous noise. No external teacher encoder does not mean no training data, text supervision, autoencoder or extra forward cost. Report/index/arXiv dates differ; no exact first-disclosure-day priority inferred.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Flow matching, denoising and self-supervised alignment have predecessors, including independently noised sequence tokens. Scoped delta is dual-noise internal feature prediction combined with flow training, not invention of arbitrary heterogeneous noise. No external teacher encoder does not mean no training data, text supervision, autoencoder or extra forward cost. Report/index/arXiv dates differ; no exact first-disclosure-day priority inferred.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- Self-Supervised Flow Matching for Scalable Multi-Modal Synthesis
- Self-Flow — original Black Forest Labs project
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction