PALATE: Peculiar Application of the Law of Total Expectation to Enhance the Evaluation of Deep Generative Models

Name: PALATE: Peculiar Application of the Law of Total Expectation to Enhance the Evaluation of Deep Generative Models
Uploaded: 2026-10-22T15:23:42.460151+00:00
Duration: 221 s
Description: Deep generative models need better evaluation for fidelity, diversity, and novelty. PALATE enhances existing metrics by using the law of total expectation combined with DINOv2 features and an MMD baseline. This new framework efficiently detects sample memorization and overfitting in generated data.

Deep generative models need better evaluation for fidelity, diversity, and novelty. PALATE enhances existing metrics by using the law of total expectation combined with DINOv2 features and an MMD baseline. This new framework efficiently detects sample memorization and overfitting in generated data.

Open the full episode