Var-JEPA: A Variational Formulation of the Joint-Embedding Predictive Architecture -- Bridging Predictive and Generative Self-Supervised Learning
About
The Joint-Embedding Predictive Architecture (JEPA) is often seen as a non-generative alternative to likelihood-based self-supervised learning, emphasizing prediction in representation space rather than reconstruction in observation space. We argue that the resulting separation from probabilistic generative modeling is largely rhetorical rather than structural: the canonical JEPA design, coupled encoders with a context-to-target predictor, mirrors the variational posteriors and learned conditional priors obtained when variational inference is applied to a particular class of coupled latent-variable models, and standard JEPA can be viewed as a deterministic specialization in which regularization is imposed via architectural and training heuristics rather than an explicit likelihood. Building on this view, we derive the Variational JEPA (Var-JEPA), which makes the latent generative structure explicit by optimizing a single Evidence Lower Bound (ELBO). This yields meaningful representations without ad-hoc anti-collapse regularizers and allows principled uncertainty quantification in the latent space. We instantiate the framework for tabular data (Var-T-JEPA) and achieve strong representation learning and downstream performance, consistently improving over T-JEPA while remaining competitive with strong raw-feature baselines.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Binary Classification | Credit Card (test) | Macro F1 Score79.8 | 40 | |
| Classification | adult (AD) (test) | Macro F1 Score92.3 | 36 | |
| Classification | Covertype (CO) (test) | Macro F1-score81.6 | 36 | |
| Classification | Bank Marketing (BM) (test) | Macro F1 Score91.2 | 36 | |
| Classification | MNIST (test) | Macro F193.6 | 36 | |
| Classification | Electricity (EL) (test) | Macro F1-score89.1 | 36 | |
| Classification | SIM (test) | Macro F1-score95.5 | 36 |