Imbalance-Robust and Sampling-Efficient Continuous Conditional GANs via Adaptive Vicinity and Auxiliary Regularization

2025-08-03

AI Summary

To address data imbalance and inefficient sampling in continuous conditional generative models (e.g., CcGAN, CCDM) for high-dimensional data, this paper proposes Adaptive-CCGAN: an integrated generative framework leveraging adaptive neighborhood modeling and a multi-task discriminator. Our method dynamically adjusts the conditional neighborhood radius and jointly enforces regression-based auxiliary supervision and density-ratio regularization to mitigate conditional distribution skew and eliminate iterative sampling. It enables single-step forward generation while achieving state-of-the-art (SOTA) fidelity across four benchmark datasets. Inference is 300–2000× faster than diffusion-based methods, with robust performance under diverse imbalance conditions. The core innovations lie in (i) adaptive neighborhood construction that tailors conditional support per input, and (ii) a multi-objective discriminator design that unifies adversarial, regression, and density-ratio estimation tasks for improved conditional consistency and sample efficiency.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Recent advances in conditional generative modeling have introduced Continuous conditional Generative Adversarial Network (CcGAN) and Continuous Conditional Diffusion Model (CCDM) for estimating high-dimensional data distributions conditioned on scalar, continuous regression labels (e.g., angles, ages, or temperatures). However, these approaches face fundamental limitations: CcGAN suffers from data imbalance due to fixed-size vicinity constraints, while CCDM requires computationally expensive iterative sampling. We present CcGAN-AVAR, an enhanced CcGAN framework that addresses both challenges: (1) leveraging the GAN framework's native one-step generation to overcome CCDMs' sampling bottleneck (achieving 300x-2000x faster inference), while (2) two novel components specifically target data imbalance - an adaptive vicinity mechanism that dynamically adjusts vicinity's size, and a multi-task discriminator that constructs two regularization terms (through auxiliary regression and density ratio estimation) to significantly improve generator training. Extensive experiments on four benchmark datasets (64x64 to 192x192 resolution) across eight challenging imbalanced settings demonstrate that CcGAN-AVAR achieves state-of-the-art generation quality while maintaining sampling efficiency.

Problem

• Addresses data imbalance in CcGAN via adaptive vicinity
• Improves sampling efficiency over CCDM with one-step generation
• Enhances generator training via multi-task discriminator regularization

Innovation

• Adaptive vicinity mechanism for imbalance
• Multi-task discriminator with regularization
• One-step GAN generation for speed