An interactive, from-scratch implementation of LeJEPA in JAX/Equinox.
Heuristics-free self-supervised representation learning for time series with SIGReg (LeJEPA). Disentangles time-axis collapse, positional structure, and representation richness across PatchTST, TCN, and bag-of-patches encoders. Reproducible, seeded, significance-tested.
From-scratch JEPA world-model engine in PyTorch (FSDP, bf16, SIGReg/VICReg): data curation, distributed training and inference optimisation, evaluated on CAMELS cosmology against control floors with seed noise bands.
A governed SRE agent that plans inside a JEPA (LeJEPA/SIGReg) latent world model, with a pre-registered test of planning as telemetry turns into noise. Delta Lake, MLflow, STERNWATCH audit. CPU only.
Implementing LeJEPA's SIGReg for an action-conditioned world model (Super Mario Bros 1-1): the calibration the objective needs outside image SSL, and per-checkpoint representation gating
LeWorldModel (JEPA world model, SIGReg) reimplemented in plain PyTorch, matching the official model on its own weights, and retrained from scratch on free Kaggle T4s to the released model's planning success.
To associate your repository with the lejepa topic, visit your repo's landing page and select "manage topics."