Back to AI News
arxiv

Good Views Are Not Enough: Global Counterexamples for Gaussian-Regularized JEPAs

The paper presents globally certified counterexamples showing that good views alone do not guarantee scene invariance or uniqueness up to rotation at global optima of Gaussian-regularized JEPAs. In a four-state model, every global minimizer splits a scene and has an equal-value partner outside its rotation orbit; a 512-dimensional model shows pooled Epps-Pulley regularization can force retention of nuisance and positive cross-modal alignment error. The authors also report that a whitened prediction JEPA aligns better to the scene than next-token prediction but has 0.2872 nats/token higher linear-readout test NLL under matched readout budgets and 1% observation noise. A Bayes-risk decomposition separates information lost through the observation channel from readout difficulty.

AI Brief

It provides theoretical counterexamples and concrete evidence that Gaussian regularization in JEPAs can undermine the assumed identifiability and prediction benefits.

arXiv:2611.12137v1 Announce Type: cross Abstract: Good views do not suffice to guarantee scene invariance or uniqueness up to rotation at the global optima of Gaussian-regularized JEPAs. We give globally certified counterexamples showing that the uniqueness premise of the Gaussian extension in Good Views fails for specified soft objectives. Gaussian marginal matching can reward retaining independently resampled observation variation, and this benefit can exceed its prediction cost. In a four-state model, comparison against the entire scene-only class proves that every global minimizer splits a scene and admits an equal-value partner outside its rotation orbit. In a 512-dimensional model with a complete scene basis, a spectral gap, and exact population whitening, pooled Epps-Pulley regularization still forces every global minimizer to retain nuisance and yields positive cross-modal alignment error even after the best rotation when modality nuisances are independent. Better scene alignment can also accompany worse prediction. With matched 512-label readout budgets and 1% observation noise, whitened prediction JEPA aligns better to the scene than next-token prediction but incurs 0.2872 nats/token higher linear-readout test NLL. A Bayes-risk decomposition distinguishes information lost through the observation channel from the fitted readout's difficulty using retained information. These results connect three determinants of learned representations: what views share, what regularization favors, and what a prediction task can use.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!