GICDM: Mitigating Hubness for Reliable Distance-Based Generative Model Evaluation

2026-02-18

AI Summary

This work addresses the hubness phenomenon in high-dimensional embedding spaces, which distorts nearest-neighbor relationships and introduces bias into distance-based evaluation metrics for generative models. To mitigate this issue, the paper proposes the Generative Iterative Contextual Dissimilarity Measure (GICDM), the first evaluation framework to incorporate a hubness correction mechanism into generative model assessment. GICDM refines the adjacency structures between real and generated data through multi-scale neighborhood adjustments, effectively alleviating hubness-induced distortions. Experimental results on both synthetic and real-world benchmarks demonstrate that GICDM significantly improves the reliability of evaluation metrics and enhances their alignment with human judgment.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Generative model evaluation commonly relies on high-dimensional embedding spaces to compute distances between samples. We show that dataset representations in these spaces are affected by the hubness phenomenon, which distorts nearest neighbor relationships and biases distance-based metrics. Building on the classical Iterative Contextual Dissimilarity Measure (ICDM), we introduce Generative ICDM (GICDM), a method to correct neighborhood estimation for both real and generated data. We introduce a multi-scale extension to improve empirical behavior. Extensive experiments on synthetic and real benchmarks demonstrate that GICDM resolves hubness-induced failures, restores reliable metric behavior, and improves alignment with human judgment.

Problem

• hubness
• generative model evaluation
• distance-based metrics
• high-dimensional embedding
• nearest neighbor

Innovation

• hubness
• generative model evaluation
• ICDM
• multi-scale
• distance-based metrics