Label-free minibatch OT coupling beats conditional OT, hits 1.91 FID on ImageNet-256
AlexanderTong7 · x · 2026-10-13
- Alexander Tong's team released a new preprint on Global Transport (GT) couplings for classifier-free guided flows.
- Surprising finding: simple minibatch optimal transport works remarkably well in the conditional setting.
- By dropping labels, GT stabilizes high-scale CFG, unexpectedly beating both class-conditional OT and independent couplings at high guidance scales.
- Achieves FID 1.91 on ImageNet-256, showing simple label-free coupling wins when pushing CFG scale up.
More from Research
- EMNLP paper: higher-accuracy LLM agents fail to flag uncertainty under knowledge conflict — DanielKhashabi · 2026-10-13
- Center for AI Safety releases CheatBench to measure reward gaming in agents — dhadfieldmenell · 2026-10-13
- 300M lab tests show menopause is a physiological cliff, not a slow slope — MaxUnfried · 2026-10-13
- Hackathon challenge: a self-healing agent that hot-patches tool schemas at runtime — parathalovespaneer · 2026-10-13
- Anthropic red team sees early signs of self-replicating AI spreading across cheap edge devices — CurieuxExplorer · 2026-10-13
- Top-5 ranked MIT econ PhD applicants publish 3x more than next 20, paper finds — paulnovosad · 2026-10-13