On Training Sample Memorization: Lessons from Benchmarking Generative Modeling with a Large-scale Competition
This work addresses the benchmarking challenges in generative modeling for researchers and practitioners, highlighting vulnerabilities in current evaluation methods.
The authors tackled the problem of gameable metrics in generative modeling by running a large-scale competition with over 11,000 submitted models, revealing that unintentional memorization is a serious and common issue in popular models. They introduced the Memorization-Informed Fréchet Inception Distance (MiFID) as a new metric to detect memorization and ensure genuine perceptual improvements.
Many recent developments on generative models for natural images have relied on heuristically-motivated metrics that can be easily gamed by memorizing a small sample from the true distribution or training a model directly to improve the metric. In this work, we critically evaluate the gameability of these metrics by designing and deploying a generative modeling competition. Our competition received over 11000 submitted models. The competitiveness between participants allowed us to investigate both intentional and unintentional memorization in generative modeling. To detect intentional memorization, we propose the ``Memorization-Informed Fréchet Inception Distance'' (MiFID) as a new memorization-aware metric and design benchmark procedures to ensure that winning submissions made genuine improvements in perceptual quality. Furthermore, we manually inspect the code for the 1000 top-performing models to understand and label different forms of memorization. Our analysis reveals that unintentional memorization is a serious and common issue in popular generative models. The generated images and our memorization labels of those models as well as code to compute MiFID are released to facilitate future studies on benchmarking generative models.