# Annealed Co-Generation: Disentangling Variables via Progressive Pairwise Modeling

Source: [Annealed Co-Generation: Disentangling Variables via Progressive Pairwise Modeling](<https://www.sota2.com/research/paper/69afa57c2d603e6de3e439b3>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-11T05:00:04.181Z

Authors: Hantao Zhang, Jieke Wu, Mingda Xu, Xiao Hu, Yingxuan You, Pascal Fua

Published: 2026-02-20T07:42:11.000Z

## Abstract

For multivariate co-generation in scientific applications, we advocate pairwise block rather than joint modeling of all variables. This design mitigates the computational burden and data imbalance. To this end, we propose an Annealed Co-Generation (ACG) framework that replaces high-dimensional diffusion modeling with a low-dimensional diffusion model, which enables multivariate co-generation by composing pairwise variable generations. We first train an unconditional diffusion model over causal variables that are disentangled into pairs. At inference time, we recover the joint distribution by coupling these pairwise models through shared common variables, enabling coherent multivariate generation without any additional training. By employing a three-stage annealing process-Consensus, Heating, and Cooling-our method enforces consistency across shared common variables and progressively constrains each pairwise data distribution to lie on a learnable manifold, while maintaining high likelihood within each pair. We demonstrate the framework's flexibility and efficacy on two distinct scientific tasks: flow-field completion and antibody generation. All datasets and code will be made publicly available upon publication.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Multi-specific antibody co-design on Split-B](<https://www.sota2.com/research/sota/multi-specific-antibody-co-design-on-split-b.md>): Multi-specific antibody co-design; Split-B; RMSD (Ag. A): 2.96.

- [Flow-field inpainting on Flow-field New Case](<https://www.sota2.com/research/sota/flow-field-inpainting-on-flow-field-new-case.md>): Flow-field inpainting; Flow-field New Case; MSE: 0.2443.

- [Flow-velocity field inpainting on Flow-velocity field dataset](<https://www.sota2.com/research/sota/flow-velocity-field-inpainting-on-flow-velocity-field-dataset.md>): Flow-velocity field inpainting; Flow-velocity field dataset; MSE: 0.1027.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2603.06615>)

[PDF](<https://arxiv.org/pdf/2603.06615.pdf>)