# On the Hardness of Unsupervised Domain Adaptation: Optimal Learners and Information-Theoretic Perspective

arXiv: https://arxiv.org/abs/2507.06552 · HTML: https://mlsanity.com/paper/2507.06552/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 13 of its 35 references that are in the map.
- First posted on arXiv 9 Jul 2025.

## Builds on (most cited first, 13 of 13)

- [Unsupervised Domain Adaptation by Backpropagation](https://mlsanity.com/paper/1409.7495/) (2014)
- [Bridging Theory and Algorithm for Domain Adaptation](https://mlsanity.com/paper/1904.05801/) (2019)
- [Coupled Generative Adversarial Networks](https://mlsanity.com/paper/1606.07536/) (2016)
- [Statistical Aspects of Wasserstein Distances](https://mlsanity.com/paper/1806.05500/) (2018)
- [On the Value of Target Data in Transfer Learning](https://mlsanity.com/paper/2002.04747/) (2020)
- [Algorithms and Theory for Multiple-Source Adaptation](https://mlsanity.com/paper/1805.08727/) (2018)
- [f-Domain-Adversarial Learning: Theory and Algorithms](https://mlsanity.com/paper/2106.11344/) (2021)
- [KL Guided Domain Adaptation](https://mlsanity.com/paper/2106.07780/) (2021)
- [Bridging the Gap Between f-GANs and Wasserstein GANs](https://mlsanity.com/paper/1910.09779/) (2019)
- [Unsupervised Domain Adaptation Based on Source-guided Discrepancy](https://mlsanity.com/paper/1809.03839/) (2018)
- [Information-Theoretic Analysis of Unsupervised Domain Adaptation](https://mlsanity.com/paper/2210.00706/) (2022)
- [The Value of Out-of-Distribution Data](https://mlsanity.com/paper/2208.10967/) (2022)
- [On f-Divergence Principled Domain Adaptation: An Improved Framework](https://mlsanity.com/paper/2402.01887/) (2024)

## Built on by (most cited first, 1 of 1)

- [When is Test-Time Adaptation Identifiable From Unlabeled Evidence?](https://mlsanity.com/paper/2609.11235/) (2026)

## Abstract (arXiv)

This paper studies the hardness of unsupervised domain adaptation (UDA) under covariate shift. We model the uncertainty that the learner faces by a distribution π in the ground-truth triples (p, q, f) -- which we call a UDA class -- where (p, q) is the source -- target distribution pair and f is the classifier. We define the performance of a learner as the overall target domain risk, averaged over the randomness of the ground-truth triple. This formulation couples the source distribution, the target distribution and the classifier in the ground truth, and deviates from the classical worst-case analyses, which pessimistically emphasize the impact of hard but rare UDA instances. In this formulation, we precisely characterize the optimal learner. The performance of the optimal learner then allows us to define the learning difficulty for the UDA class and for the observed sample. To quantify this difficulty, we introduce an information-theoretic quantity -- Posterior Target Label Uncertainty (PTLU) -- along with its empirical estimate (EPTLU) from the sample , which capture the uncertainty in the prediction for the target domain. Briefly, PTLU is the entropy of the predicted label in the target domain under the posterior distribution of ground-truth classifier given the observed source and target samples. By proving that such a quantity serves to lower-bound the risk of any learner, we suggest that these quantities can be used as proxies for evaluating the hardness of UDA learning. We provide several examples to demonstrate the advantage of PTLU, relative to the existing measures, in evaluating the difficulty of UDA learning.