Reproduce “Generative Uncertainty as a Self-supervised Signal for Semantic Similarity Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Enrico Pallotta, Sina Raoufi, Lars Doorenbos, Gianni Franchi, Juergen Gall
The paper adapts pretrained video encoders for semantic similarity by training a feature mask from repeated text-to-video generations: same-prompt pairs from general prompts are treated as stable, while specialized-prompt pairs are treated as divergent. For V-JEPA 2.1-L on DDPM-OOD, the binary mask reaches AUROC 89.65 versus 75.49 unmasked, in this setting, using observed variability as a proxy for semantic consistency.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.