Fixed isotropic marginals in JEPAs can be maximally misaligned with unknown structured geometries, and HamJEPA using symplectic Hamiltonian leapfrog maps improves kNN and linear-probe performance on CIFAR-100 and ImageNet-100.
Simplifying dino via coding rate regularization
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
years
2026 3verdicts
UNVERDICTED 3representative citing papers
A visualization protocol using unsupervised semantic segmentation outputs reveals positional biases, scaling behaviors, and boundary artifacts in self-supervised ViTs and distinguishes them from locality bias.
Randomly initialized networks trained solely via peer-to-peer self-distillation learn useful representations that outperform random baselines on downstream tasks.
citing papers explorer
-
Beyond Isotropy in JEPAs: Hamiltonian Geometry and Symplectic Prediction
Fixed isotropic marginals in JEPAs can be maximally misaligned with unknown structured geometries, and HamJEPA using symplectic Hamiltonian leapfrog maps improves kNN and linear-probe performance on CIFAR-100 and ImageNet-100.
-
Unsupervised Semantic Segmentation Facilitates Model Understanding
A visualization protocol using unsupervised semantic segmentation outputs reveals positional biases, scaling behaviors, and boundary artifacts in self-supervised ViTs and distinguishes them from locality bias.
-
Randomly Initialized Networks Can Learn from Peer-to-Peer Consensus
Randomly initialized networks trained solely via peer-to-peer self-distillation learn useful representations that outperform random baselines on downstream tasks.