cs.LGNov 9, 2026

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

Authors: Anran Xu

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.

Figures & tables

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Nov 12, 2026

Same-State Pairs Make One JEPA Proof of the Platonic Representation Hypothesis Vacuous

In a Joint-Embedding Predictive Architecture (JEPA), the JEPA proof of the Platonic Representation Hypothesis (PRH) assumes that every observation determines the world state and that a spectral gap makes minimizers unique up to rotation. Under the hard-whitening population objective, same-state positive pairs make the induced world kernel the identity: every centered whitened state function is a global minimizer, and, when dz−1≥2kd_z-1\ge 2k, two minimizers can have best-rotation squared residual 2k2k. The uniqueness claim is conditional; for k<dz−1k<d_z-1, where dzd_z is the number of world states, this limit makes its premise unsatisfiable, so the claim has no content there. The experiments that motivate it leave the assumption: a caption sees only part of an image, and the paper states no experiment meets its hypotheses exactly. The theorem does not reach those experiments, and for k<dz−1k<d_z-1 its idealized same-state limit is vacuous. One of the paper's four quantitative results survives: a gap-free lower bound. Its bound vanishes only at perfect scene-explained fraction; it bounds alignment error but cannot certify good-view alignment. A canonical frame is determined only up to a rotation and a latent permutation. Paired samples fix the relative frame at effective rank kk; a global rotation remains. Component semantics fix the relative permutation only up to signs; value anchors fix it once the mixing frame is pinned. A resource--gauge ledger records which resource fixes which gauge.
May 19, 2026

Beyond Isotropy in JEPAs: Hamiltonian Geometry and Symplectic Prediction

JEPAs often regularize one-view embeddings toward an isotropic Gaussian, implicitly baking Euclidean symmetry into the representation. We show that this is not merely a benign default. For a known structured downstream geometry H≻0H\succ0, the minimax and maximum-entropy covariance under a Hamiltonian energy budget is (c/d)H−1(c/d)H^{-1}, and Euclidean isotropy incurs a closed-form price of isotropy. More importantly, when the downstream geometry is unknown, no geometry-independent fixed marginal target is canonical: every fixed covariance shape can be maximally misaligned for some structured geometry. We further show that even oracle one-view marginals do not identify the JEPA view-to-view predictive coupling. Within continuous linear phase-space flows, stationarity of the lifted oracle law, momentum reversal, and fixed-unit kinematics uniquely select Hamiltonian dynamics. These results suggest that the structural bias in JEPAs should enter the cross-view coupling rather than a fixed encoder marginal. We instantiate this principle with HamJEPA, which encodes each view as a phase-space state (q,p)(q,p) and predicts view-to-view transitions with a learned Hamiltonian leapfrog map, while non-isotropic scale and spectral floors discourage collapse. In a deliberately headless token protocol, HamJEPA improves over SIGReg on CIFAR-100 by +4.89+4.89 kNN@20 and +3.52+3.52 linear-probe points at 30 epochs, and, over two training seeds at 80 epochs, by +7.24+7.24 kNN@20 and +9.67+9.67 linear-probe points relative to the full 2048-dimensional SIGReg representation. Matched non-symplectic and non-reversible controls retain the gain, so the experiments do not identify either property as individually necessary. On ImageNet-100, HamJEPA-qq improves by +4.82+4.82 kNN@20 and +7.52+7.52 linear-probe points at 45 epochs.
Oct 7, 2026

Shared Gaussianization: What Gaussian Regularizers Certify About Contrastive Learning, and What They Miss

What can a distribution-matching regularizer such as SIGReg in LeJEPA certify about contrastive learning? We study shared Gaussianization (SG), a characteristic-function Gaussianity test on the average of two normalized views, scaled by an independent χdχ_d radius. Because disagreeing views shorten the average, one test detects both misalignment and non-uniformity. SG vanishes exactly at the aligned, uniform minimizers of population InfoNCE, and under equal marginals it bounds the InfoNCE excess by 4⋅33/4β4\cdot 3^{3/4}β times the square root of the SG loss, plus a term linear in the loss. The square-root rate and this dimension-free constant are sharp, and no squared mean-embedding distance on view pairs achieves a faster rate. With an explicit alignment term, a rotation-invariant uniformity test gives a linear bound if and only if its spectrum dominates that of InfoNCE's kernel eβu⊤ve^{βu^\top v}; SG's own test does, Gaussian kernels e−γ∥u−v∥2e^{-γ\|u-v\|^2} qualify exactly when γ≥β/2γ\ge β/2, and moment matching never does. Away from the optimum, the objectives differ. Along an isotropic nuisance channel, pure SG lowers its loss by adding per-view nuisance whenever the shared code is non-uniform. An alignment weight above the channel's gain makes the nuisance-free solution a strict local minimizer; for LeJEPA, the same rule gives a critical SIGReg weight that decreases with the batch size. At finite batch size, an off-diagonal U-statistic removes a plug-in bias toward misalignment. In controlled latent-variable models, pure SG retains per-view style, an alignment weight above the measured gain removes it, and for LeJEPA at three batch sizes the measured gain separates the encoders that retain style from those that do not. InfoNCE training also reaches a lower SG0.2_{0.2} loss than SG0.2_{0.2} training from scratch, which points to an optimization gap.