<a href=\"https://cdn-uploads.huggingface.co/production/uploads/69b1e787fc5716730171aba3/4uKgQDptbMGu0DDhVKvRQ.jpeg\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/69b1e787fc5716730171aba3/4uKgQDptbMGu0DDhVKvRQ.jpeg\" alt=\"Consistency_Overview\"></a><br>Consistency-based, test-time self-supervised error estimation with a bidirectional diffusion model: a directional<br>flag c_d = ±1 selects forward or backward rollout; reversing it rolls back to an estimate of the starting point and the amount by which we miss can be used to predict roll out error.</p>\n","updatedAt":"2026-08-08T04:01:02.859Z","author":{"_id":"69b1e787fc5716730171aba3","avatarUrl":"/avatars/0389e0a510d65cdc6ac11e94629671fb.svg","fullname":"Alexander Scheinker","name":"AScheinker","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.28729426860809326},"editors":["AScheinker"],"editorAvatarUrls":["/avatars/0389e0a510d65cdc6ac11e94629671fb.svg"],"reactions":[],"isReport":false}},{"id":"6a794d1cdf4e886d3d3e8eb4","author":{"_id":"69b1e787fc5716730171aba3","avatarUrl":"/avatars/0389e0a510d65cdc6ac11e94629671fb.svg","fullname":"Alexander Scheinker","name":"AScheinker","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-08-10T04:01:32.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"A diffusion model that predicts its own rollout errors — code and physics benchmarks linked on the paper page.\n","html":"<p>A diffusion model that predicts its own rollout errors — code and physics benchmarks linked on the paper page.<br><a href=\"https://cdn-uploads.huggingface.co/production/uploads/69b1e787fc5716730171aba3/rEGsDSDexR3VqGfE5eLqh.jpeg\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/69b1e787fc5716730171aba3/rEGsDSDexR3VqGfE5eLqh.jpeg\" alt=\"Consistency_Overview\"></a></p>\n","updatedAt":"2026-08-10T04:01:32.620Z","author":{"_id":"69b1e787fc5716730171aba3","avatarUrl":"/avatars/0389e0a510d65cdc6ac11e94629671fb.svg","fullname":"Alexander Scheinker","name":"AScheinker","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.44489672780036926},"editors":["AScheinker"],"editorAvatarUrls":["/avatars/0389e0a510d65cdc6ac11e94629671fb.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.00675","authors":[{"_id":"6a73240021d743496486c3e3","user":{"_id":"69b1e787fc5716730171aba3","avatarUrl":"/avatars/0389e0a510d65cdc6ac11e94629671fb.svg","isPro":false,"fullname":"Alexander Scheinker","user":"AScheinker","type":"user","name":"AScheinker"},"name":"Alexander Scheinker","status":"claimed_verified","statusLastChangedAt":"2026-08-08T00:45:04.539Z","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/69b1e787fc5716730171aba3/2lV2pETi7xEBqlligVbjR.jpeg"],"publishedAt":"2026-08-01T13:49:46.000Z","submittedOnDailyAt":"2026-08-10T00:00:00.000Z","title":"Round-Trip Consistency: Bidirectional Diffusion Models Can Predict Their Own Rollout Errors","submittedOnDailyBy":{"_id":"69b1e787fc5716730171aba3","avatarUrl":"/avatars/0389e0a510d65cdc6ac11e94629671fb.svg","isPro":false,"fullname":"Alexander Scheinker","user":"AScheinker","type":"user","name":"AScheinker"},"summary":"Autoregressive models accumulate error over long rollouts, yet at deployment there is no ground truth to measure it against. We train a single conditional latent diffusion model that steps a dynamical system forward or backward in time via a direction flag, and show that this bidirectionality supplies a measurement-free test-time error signal: rolling forward i steps and then backward i steps must return the model to its start, so the round-trip discrepancy C_i is a self-supervised proxy for the unobservable rollout error: no ensembles, no held-out data, no governing equations, for one extra rollout. We validate on compressible magnetohydrodynamics (MHD), an astrophysical turbulent radiative mixing layer, and natural face videos (CelebV-HQ). On held-out MHD trajectories, C_i ranks rollout error (Spearman 0.91-0.98 at fixed depth; 0.69 pm 0.16 within trajectories), and a simple calibrator fit on training rollouts predicts its magnitude to within 1.14times (68%) and 1.29times (95%) with near-nominal coverage - one nat beyond a depth-only predictor, transferring to all six decoded physical fields. The same signal flags the out-of-distribution Orszag-Tang vortex (AUROC 0.98; 1.0 by depth 10) exactly where sampling-dispersion baselines invert, and it cuts incurred error by 15% at 80% coverage - three times the depth-only baseline. Bidirectional training comes at negative cost, beating direction specialists in both directions, and the backward direction doubles as a fast inverse solver. On LE-PDE-UQ's turbulent Navier-Stokes benchmark, a single bidirectional model reaches accuracy within 1.3times of their ten-model ensemble at a tenth of the training cost, with the best training-free pixel-level calibration. Round-trip consistency turns reversibility into a practical trust signal for generative models.","upvotes":2,"discussionId":"6a73240021d743496486c3e4","projectPage":"https://alexscheinker.github.io/roundtrip.html","githubRepo":"https://github.com/alexscheinker/round-trip-consistency","githubRepoAddedBy":"user","githubStars":8},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"69b1e787fc5716730171aba3","avatarUrl":"/avatars/0389e0a510d65cdc6ac11e94629671fb.svg","isPro":false,"fullname":"Alexander Scheinker","user":"AScheinker","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.00675.md","query":{}}">
Round-Trip Consistency: Bidirectional Diffusion Models Can Predict Their Own Rollout Errors
Abstract
Autoregressive models accumulate error over long rollouts, yet at deployment there is no ground truth to measure it against. We train a single conditional latent diffusion model that steps a dynamical system forward or backward in time via a direction flag, and show that this bidirectionality supplies a measurement-free test-time error signal: rolling forward i steps and then backward i steps must return the model to its start, so the round-trip discrepancy C_i is a self-supervised proxy for the unobservable rollout error: no ensembles, no held-out data, no governing equations, for one extra rollout. We validate on compressible magnetohydrodynamics (MHD), an astrophysical turbulent radiative mixing layer, and natural face videos (CelebV-HQ). On held-out MHD trajectories, C_i ranks rollout error (Spearman 0.91-0.98 at fixed depth; 0.69 pm 0.16 within trajectories), and a simple calibrator fit on training rollouts predicts its magnitude to within 1.14times (68%) and 1.29times (95%) with near-nominal coverage - one nat beyond a depth-only predictor, transferring to all six decoded physical fields. The same signal flags the out-of-distribution Orszag-Tang vortex (AUROC 0.98; 1.0 by depth 10) exactly where sampling-dispersion baselines invert, and it cuts incurred error by 15% at 80% coverage - three times the depth-only baseline. Bidirectional training comes at negative cost, beating direction specialists in both directions, and the backward direction doubles as a fast inverse solver. On LE-PDE-UQ's turbulent Navier-Stokes benchmark, a single bidirectional model reaches accuracy within 1.3times of their ten-model ensemble at a tenth of the training cost, with the best training-free pixel-level calibration. Round-trip consistency turns reversibility into a practical trust signal for generative models.
Community
Consistency-based, test-time self-supervised error estimation with a bidirectional diffusion model: a directional
flag c_d = ±1 selects forward or backward rollout; reversing it rolls back to an estimate of the starting point and the amount by which we miss can be used to predict roll out error.
A diffusion model that predicts its own rollout errors — code and physics benchmarks linked on the paper page.
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.00675 in a model README.md to link it from this page.
Cite arxiv.org/abs/2608.00675 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.00675 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.