Harpoon : Generalised Manifold Guidance for Conditional Tabular Diffusion

Aditya Shankar Yuandou Wang Rihan Hai Lydia Chen · 2026

Read Harpoon : Generalised Manifold Guidance for Conditional Tabular Diffusion by Aditya Shankar Yuandou Wang Rihan Hai Lydia Chen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07875.txt.

Description

Made tabular data must often obey conditions, like filling a missing value or respecting an inequality, but most generators only learn conditions seen during training. Aditya Shankar and three co-authors extend manifold theory to tabular data so that new constraints can be imposed at making time instead.

Their method, HARPOON, guides unconstrained diffusion samples along the data manifold's geometry until they satisfy diverse inference-time conditions. Tests on imputation and inequality-constraint tasks across several datasets show strong results, demonstrating that manifold-aware guidance works beyond continuous domains. Data scientists who need controlled made tables gain a flexible tool that does not require retraining for each new rule.

Catalog record

CREATORAditya Shankar Yuandou Wang Rihan Hai Lydia Chen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE88 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07875

Browse the full library · About the Tower of Babel