Conditional Unbalanced Optimal Transport Maps: An Outlier-Robust Framework for Conditional Generative Modeling
Jiwoo Yoon, Kyumin Choi, Jaewoong Choi · 2026
Read Conditional Unbalanced Optimal Transport Maps: An Outlier-Robust Framework for Conditional Generative Modeling by Jiwoo Yoon, Kyumin Choi, Jaewoong Choi (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-2603-06972.txt.
Introduction
Yoon and colleagues introduce Conditional Unbalanced Optimal Transport maps, an outlier-robust framework for generative modeling given conditions. Transport seeks a cheapest map between source and target distributions under given conditions, now used for generation. Yet classical transport shares the outlier weakness of optimal transport from hard matching constraints, worse where each distribution under given conditions comes from a small data set. Their framework relaxes matching constraints through Csiszár divergence penalties and keeps the given marginals, and they derive dual forms.
Catalog record
| CREATOR | Jiwoo Yoon, Kyumin Choi, Jaewoong Choi |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 77896 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.06972
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Conditional Unbalanced Optimal Transport Maps: An Outlier-Robust Framework for Conditional Generative Modeling Jiwoo Yoon, Kyumin Choi, Jaewoong Choi Abstract Conditional Optimal Transport (COT) problem aims to find a transport map between conditional source and target distributions while minimizing the transport cost. Recently, these transport maps have been utilized in conditional generative modeling tasks to establish efficient mappings between the distributions. However, classical COT inherits a fundamental limitation of optimal transport, i.e., sensitivity to outliers, which arises from the hard distribution matching constraints. This limitation becomes more pronounced in a conditional setting, where each conditional distribution is estimated from a limited subset of data. To address this, we introduce the Conditional Unbalanced Optimal Transport (CUOT) framework, which relaxes conditional distribution-matching constraints through Csiszár divergence penalties while strictly preserving the conditioning marginals. We establish a rigorous formulation of the CUOT problem and derive its dual and semi-dual formulations. Based on the semi-dual form, we propose Conditional Unbalanced Optimal Transport Maps (CUOTM), an outlier-robust conditional generative model built upon a triangular c c -transform parameterization. We theoretically justify the validity of this parameterization by proving that the optimal triangular map satisfies the c c -transform relationships. Our experiments on 2D synthetic and image-scale datasets demonstrate that CUOTM achieves superior outlier robustness and competitive distribution-matching performance compared to existing COT-based baselines, while maintaining high sampling efficiency. I Introduction Conditional generative modeling aims to approximate a conditional distribution p_{\mathrm{data}}(\cdot\mid y) , where y denotes a conditioning variable, such as a class label [ 28 ] , a text prompt [ 12 ] , or noisy measurements in an inverse problem [ 38 ] . Recent advances in generative modeling, including GANs [ 28 , 3 , 20 ] , diffusion models [ 35 ] , and flow matching models [ 27 ] , have achieved remarkable success in modeling such conditional distributions. Among existing approaches, Conditional Optimal Transport (COT) has emerged as a principled framework for conditional generation due to its rigorous theoretical formulation and sampling efficiency [ 17 , 21 , 5 ] . While standard Optimal Transport (OT) investigates the transport problem between two marginal distributions (the source \eta and the target \nu ), the COT problem focuses on the transport between conditional distributions . Specifically, for each fixed condition y , the objective is to learn a transport map (or transport plan) that transforms the source conditional distribution \eta(\cdot|y) into the target conditional distribution \nu(\cdot|y) . This provides a natural principled foundation for conditional generative modeling by setting \nu=p_{\mathrm{data}} . However, classical OT formulations (both marginal and conditional) are known to be sensitive to outliers [ 2 , 34 ] . This sensitivity originates from the hard distribution-matching constraints, which force the transport plan to account for every empirical sample [ 8 , 34 ] . When the data contains noise or corrupted samples, the optimal transport map can be significantly distorted, resulting in unstable estimation depending on sampling variation. This vulnerability is further amplified in the conditional setting . In conditional generative modeling, the dataset is partitioned according to the conditioning variable. As a result, each conditional distribution is estimated from a smaller subset of data. In these data-sparse regimes, even a few outliers can exert a disproportionately large influence on the learned conditional transport map, thereby amplifying the instability inherent in classical OT. In practical applications, where imperfect or contaminated data are common [ 24 , 18 ] , this limitation can severely affect conditional generation quality. To overcome this limitation, we introduce the Conditional Unbalanced Optimal Transport (CUOT) framework. Our key idea is to relax the rigid distribution-matching constraints of COT while preserving the structural constraint on the conditioning variable. Specifically, we incorporate Csiszár divergence penalties to allow controlled deviations within each conditional distribution, while strictly enforcing the marginal constraint on the conditioning variable y . This formulation preserves conditional alignment while introducing robustness against outliers. Building upon this formulation, we propose Conditional Unbalanced Optimal Transport Maps (CUOTM) , a conditional generative model derived from the semi-dual formulation of the CUOT problem. CUOTM parameterizes the conditional c -transform using a triangular transport map and enables efficient adversarial-style training. We theoretically justify this parameterization by proving that the optimal conditional transport map satisfies this relationship. In addition, we establish bounds that quantify how the relaxed conditional marginals deviate from the original distributions, explicitly characterizing the robustness–accuracy trade-off. Experimental results on 2D synthetic datasets and class-conditional CIFAR-10 demonstrate that CUOTM achieves strong robustness to outliers while maintaining competitive distribution-matching performance. In particular, our model outperforms existing dynamic approaches that require 100 NFEs, while employing only one NFE for CIFAR-10 conditional generation task. Our contributions can be summarized as follows: • We introduce the first mathematical formulation of the Conditional Unbalanced Optimal Transport problem, which relaxes conditional distribution-matching constraints via divergence penalties while preserving the conditioning marginals. • We establish the dual and semi-dual formulations of CUOT, extending classical Unbalanced Optimal Transport (UOT) theory to the conditional setting. • We propose Conditional Unbalanced Optimal Transport Maps (CUOTM), a conditional generative model derived from the semi-dual formulation of CUOT through a triangular c -transform parameterization. We further provide theoretical justification for the validity of this parameterization. • CUOTM demonstrates superior robustness to outliers while maintaining competitive distribution-matching performance on conditional generation benchmarks. II Preliminaries 1) Notations and Assumptions Let \mathcal{V},\mathcal{U}\subseteq\mathbb{R}^{d} and \mathcal{Y}\subseteq\mathbb{R}^{c} be Polish spaces, where \mathcal{V} and \mathcal{U} are closures of connected open sets in \mathbb{R}^{d} , and \mathcal{Y} is the closure of a connected open set in \mathbb{R}^{c} . We define \mathcal{V} and \mathcal{U} as the source and target data spaces , respectively, and \mathcal{Y} as the conditioning variable space . The set of Borel probability measures on a space (\cdot) is denoted by \mathcal{P}(\cdot) . For a measure \eta\in\mathbb{P}(\mathcal{V}) and a measurable map T:\mathcal{V}\to\mathcal{U} , the pushforward measure of \eta through T is denoted by T_{\#}\eta , i.e., T_{\#}\eta(A):=\eta(T^{-1}(A)) for any Borel set A\subseteq\mathcal{U} . We also denote the space of continuous functions on a given domain by \mathcal{C}(\cdot) . Throughout this paper, we consider the quadratic cost function c(z,v;y,u)=c(v,u)=\tau\|v-u\|_{2}^{2} for positive constant \tau>0 . 2) Optimal Transport Optimal Transport (OT) theory investigates the problem of transforming a source probability distribution \mu\in\mathbb{P}(\mathcal{V}) into a target distribution \nu\in\mathbb{P}(\mathcal{U}) with minimal transport cost [ 37 , 33 ] . This transformation can be formulated as a deterministic transport map ( Monge’s OT problem ) or a stochastic transport plan ( Kantorovich’s OT problem ). Formally, in Monge’s OT formulation, the objective is to find a deterministic transport map T:\mathcal{V}\to\mathcal{U} that minimizes the total transport cost, given by a cost function c:\mathcal{V}\times\mathcal{U}\to\mathbb{R} : \inf_{T_{\#}\mu=\nu}M(T),\quad\text{where}\quad M(T):=\int_{\mathcal{V}}c(v,T(v))d\mu(v). (1) The optimization is conducted under the constraint T_{\#}\mu=\nu , which requires that the transport map T exactly transforms \mu to \nu . Because T is a measurable function, it defines a deterministic coupling between the source sample v\sim\mu and the target T(v) . However, due to this restriction to deterministic maps, the existence of an optimal transport map T^{\star} , i.e., the minimizer of M(T) , is often not guaranteed depending on the selection of \mu and \nu . For instance, a solution may not exist if \mu is a Dirac mass and \nu is a continuous