AI Paper Atlas

Research map / Probabilistic and scientific modeling / Variational and generative sampling

DiScoFormer: Plug-In Density and Score Estimation with Transformers

Vasily Ilin, Peter Sushko, Ranjay Krishna

ICML 2026 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Score-based generative modeling analysis · 0 citations (OpenAlex)

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
add whitening layer (ablation)DiScoFormer · d=1 GMM, OOD scalesscore MSE: 1.136 -> 0.020Improved
add whitening layer (ablation)DiScoFormer · d=1 GMM, OOD scaleslog-density MSE: 1.593 -> 0.123Improved
add whitening layer (ablation)DiScoFormer · d=1 GMM, in-distributionscore MSE: 0.118 -> 0.107Improved
add whitening layer (ablation)DiScoFormer · d=1 GMM, in-distributionlog-density MSE: 0.066 -> 0.058Improved
add test-time training with consistency loss (8 steps)DiScoFormer · 2D Student-t, n=128score MSE: 0.1980 -> 0.1450Improved
add test-time training (6 steps)DiScoFormer · 2D Student-t, n=512score MSE: 0.0574 -> 0.0485Improved
add more TTT steps (6 to 8)DiScoFormer · 2D Student-t, n=512score MSE: 0.0485 -> 0.0514Hurt
replace Scott KDE with DiScoFormerscore estimator · 2D Laplace, n=512score MSE: 0.3810 -> 0.3598Improved
replace Scott KDE with DiScoFormer (no TTT)score estimator · 2D Student-t, n=128score MSE: 0.1515 -> 0.1980Hurt
replace Scott KDE with DiScoFormerscore estimator · GMM d=10, n=2^8relative score MSE (%): 65.6 -> 7.49Improved
replace best KDE with DiScoFormerscore/log-density estimator · GMM d=100score MSE: 1.090 -> 0.167Improved
scale context size beyond training range (2^14 to 2^17)DiScoFormer · GMM d=10 extrapolationrelative score MSE (%): 2.83 -> 2.74Improved

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