SGN: A Similarity-based Generative Network for Data Generation under Distribution Shift

2026-07-20

AI Summary

This work addresses the challenge of distribution shift, where generative models trained on a source domain produce mismatched samples when applied to a target domain. To overcome this without fine-tuning model parameters, the authors propose the Similarity-Guided Generation Network (SGN)—a reusable framework that adapts to arbitrary target domains using only a few labeled target samples. SGN constructs a structured latent space during a single source-domain training phase by leveraging label-guided pairwise similarity, enabling simultaneous preservation of target-specific characteristics and class consistency. Built upon an encoder–decoder architecture with a latent space composition mechanism and dimension feasibility analysis, SGN demonstrates effective target-guided data augmentation under distribution shifts across both image and tabular data, marking the first parameter-free, reusable generative approach for such scenarios.

Abstract

Generative models trained on a source domain often produce samples that are poorly aligned with shifted target domains, limiting their effectiveness for target-domain data augmentation. Although target-specific adaptation can reduce this mismatch, it typically requires additional optimization and domain-specific parameters. We propose a Similarity-based Generative Network (SGN), a reusable framework that is trained once on labeled source data and applied to new target domains without parameter updates. SGN learns a latent space structured by label-induced pairwise similarities while preserving reconstructive information through an encoder-decoder architecture. At generation time, a small labeled representative set from the target domain is encoded and combined in the learned latent space, allowing the generated samples to inherit target-specific characteristics while maintaining class consistency. We further analyze the realizability and dimensionality requirements of the proposed similarity structure. Experiments on image and tabular datasets demonstrate the effectiveness of SGN for target-guided data augmentation under source-to-target distribution shifts.

Problem

• distribution shift
• data generation
• domain adaptation
• generative models
• data augmentation

Innovation

• Similarity-based Generative Network
• distribution shift
• target-guided data augmentation
• latent space alignment
• parameter-free adaptation