AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

Universal Inverse Distillation for Matching Models with Real-Data Supervision (No GANs)

Nikita Maksimovich Kornilov, David Li, Tikhon Mavrin, Aleksei Leonov, Nikita Gushchin, Evgeny Burnaev, Iaroslav Sergeevich Koshelev, Alexander Korotin

ICLR 2026 · Image generation › Fast diffusion sampling · approach: Flow matching distillation

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
incorporate real data into distillation via RealUID coefficients with beta/alpha near 0.98 or 1.02 (ablation)UID (FGM) flow distillation · CIFAR-10 unconditionalFID: 2.58 -> 2.23Improved
incorporate real data via RealUID with beta/alpha near 0.98 (ablation)UID (FGM) flow distillation · CIFAR-10 conditionalFID: 2.21 -> 2.02Improved
use real data with beta/alpha = 1 or values outside [0.98, 1.02] (ablation)RealUID · CIFAR-10FID: unconditional baseline 2.58; beta/alpha=0.96 gives 2.66-2.97Hurt
add GAN adversarial loss with discriminator for real data (ablation)UID (FGM) · CIFAR-10 unconditionalFID: 2.58 -> 2.29Improved
increase GAN loss weights to 5/15 (ablation)UID + GAN · CIFAR-10 unconditionalFID: 2.29 -> 2.54Hurt
use RealUID instead of GAN for real data (ablation)UID (FGM) · CIFAR-10 conditionalFID: 2.12 (GAN) -> 2.02 (RealUID)Improved
add real-data coefficients to speed convergenceRealUID · CIFAR-10iterations to saturation: ~300k -> ~100k iterationsImproved
add fine-tuning stage with new alpha_FT, beta_FT from best checkpointRealUID · CIFAR-10 unconditionalFID: 2.23 -> 1.98Improved
add fine-tuning stageRealUID · CIFAR-10 conditionalFID: 2.02 -> 1.87Improved
use RealUID with fine-tuning instead of GAN with fine-tuning (ablation)UID · CIFAR-10 unconditionalFID: 2.10 -> 1.98Improved
use RealUID with fine-tuning instead of GAN with fine-tuning (ablation)UID · CIFAR-10 conditionalFID: 1.88 -> 1.87No clear effect
add real data, student vs teacherRealUID · CIFAR-10FID: Teacher 3.57 -> 1.98 (uncond); 5.56 -> 1.87 (cond)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy