Search papers, labs, and topics across Lattice.
The paper addresses the problem of hubness in high-dimensional embedding spaces used for evaluating generative models, which distorts nearest neighbor relationships and biases distance-based metrics. They introduce Generative ICDM (GICDM), an adaptation of Iterative Contextual Dissimilarity Measure (ICDM), to correct neighborhood estimation for both real and generated data by mitigating hubness. Experiments on synthetic and real datasets demonstrate that GICDM resolves hubness-induced failures, leading to more reliable metric behavior and better alignment with human judgment.
Distance-based evaluations of generative models are biased by the "hubness" phenomenon in high-dimensional embeddings, but GICDM offers a fix that aligns better with human judgment.
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.
