Research map / Label-efficient and robust learning / Domain adaptation and generalization

SimProF: A Simple Probabilistic Framework for Unsupervised Domain Adaptation

AAAI 2025 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Feature alignment for unsupervised adaptation · 1 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes SimProF, a probabilistic UDA framework that models per-class normalized features as von Mises-Fisher distributions, derives a closed-form expected supervised contrastive loss over infinitely many sampled pairs, and augments source features toward the target domain.

Problem

Unsupervised domain adaptation: transferring a classifier from a labeled source domain to an unlabeled target domain under domain shift, where contrastive methods usually need large batches and classifiers biased toward source.

Approach

Model normalized features with mixtures of vMF distributions on the unit sphere, estimate class means online via inter-domain mean differences, and optimize a closed-form upper bound of the expected contrastive loss (equivalent to infinite contrastive pairs), plus a mutual-information term on target predictions.

Key results

• SimProF reaches 93.5% average accuracy on Office-31, versus 90.1% for T2SA.
• SimProF reaches 76.7% average accuracy on Office-Home, versus 74.4% for T2SA.
• On VisDA-2017 SimProF improves over T2SA by 4.7 points; on DomainNet the average gain over T2SA is 1.6 points.

Measured effects reported in the paper

Limitations

• Results averaged over three random trials; only ResNet-50 backbone reported.
• Gains over T2SA on DomainNet are modest (1.6 points).

Experimental setup

Datasets: Office-31 Office-Home VisDA-2017 DomainNet

Benchmarks: Office-31 Office-Home VisDA-2017 DomainNet

Metrics: recognition accuracy proxy A-distance average accuracy

Baselines: ResNet-50 DDC DAN DANN SimNet CyCADA CDAN TADA BSP MCD SAFN DSAN

Builds on: T2SA supervised contrastive learning CDAN DANN

Closest papers by meaning

• Connect, Not Collapse: Explaining Contrastive Learning for Unsupervised Domain Adaptation (ICML 2022 · similarity 0.90)
• Contrastive Adaptation Network for Unsupervised Domain Adaptation (CVPR 2019 · similarity 0.90)
• Divide and Contrast: Source-free Domain Adaptation via Adaptive Contrastive Learning (NeurIPS 2022 · similarity 0.90)
• A Prototype-Oriented Framework for Unsupervised Domain Adaptation (NeurIPS 2021 · similarity 0.89)
• FixBi: Bridging Domain Spaces for Unsupervised Domain Adaptation (CVPR 2021 · similarity 0.89)
• CLDA: Contrastive Learning for Semi-Supervised Domain Adaptation (NeurIPS 2021 · similarity 0.89)