New preprint: geometry-aware time reparameterization improves flow-map distillation
FrnkNlsn · x · 2026-10-08
An OIST team (Félix Dedek, Makoto Yamada) posted arXiv:2610.02427 on Geometry-Aware Time Reparameterization for Flow-Map Distillation. Idea: flow-map distillation learns finite-time transitions of a pretrained generative ODE for one/few-step generation; segments with large normal acceleration are harder to distill, so the paper allocates more student time to highly curved regions while preserving the teacher's geometric paths and terminal distribution. Key property: the shared clock is estimated once before distillation — no teacher retraining, no extra student parameters or inference cost. Results: on synthetic data, CIFAR-10 and CelebA-64, sample quality beats identity-time distillation at matched budgets, including one- and two-step generation; gains in one-step generation (where no intermediate times exist) highlight the value of time reparameterization during distillation.
More from Research
- Sebastian Raschka traces text classification from bag-of-words to the viral Jev decision model — AxSaucedo · 2026-10-08
- Scott Aaronson: AI labs are using internal models to attack core cryptographic protocols — skdh · 2026-10-08
- Hierarchical RL with mixed discount rates may unlock long-horizon agent tasks — jessi_cata · 2026-10-08
- Big Math Drop of Oct 6 analyzed: major progress toward, not yet, Millennium Problems — burny_tech · 2026-10-08
- Yukon's multiplayer autoresearch platform lets humans and AI agents beat benchmarks like Google Quantum AI's by 67% — RexDouglass · 2026-10-08
- OpenAI math repo formalizes ~42% of top-line results, withdraws 3 papers over sign error — danintheory · 2026-10-08