Resources
Everything in one place: every method with paper and code links, plus the datasets, benchmarks and frameworks that matter for world modeling — and for a world model of the cell.
Foundations
| Method | Type | Year | Paper | Code |
|---|---|---|---|---|
| A Path Towards Autonomous Machine Intelligence ★ | Theory | 2022 | paper | — |
| Introduction to Latent Variable Energy-Based Models | Theory | 2023 | paper | — |
Core Architectures
Theory & Analysis
| Method | Type | Year | Paper | Code |
|---|---|---|---|---|
| LeJEPA ★ | Theory | 2025 | paper | code |
| EB-JEPA | Theory | 2026 | paper | code |
| JEPAs Focus on Slow Features | Theory | 2022 | paper | — |
| How JEPA Avoids Noisy Features | Theory | 2024 | paper | — |
| Connecting JEPA with Contrastive SSL | Theory | 2024 | paper | — |
| Why and How Auxiliary Tasks Improve JEPA | Theory | 2025 | paper | — |
| Image World Models (IWM) | ImageTheory | 2024 | paper | — |
| LiDAR: Sensing Linear Probing Performance | Theory | 2023 | paper | — |
| VICReg | Theory | 2021 | paper | — |
| Understanding SSL Dynamics without Contrastive Pairs | Theory | 2021 | paper | — |
| Var-JEPA | Theory | 2026 | paper | — |
| Gaussian Joint Embeddings | Theory | 2026 | paper | — |
World Models, Robotics & Planning
| Method | Type | Year | Paper | Code |
|---|---|---|---|---|
| V-JEPA 2 ★ | VideoAction | 2025 | paper | code |
| What Drives Success in Physical Planning with JEPA World Models? | ActionTheory | 2025 | paper | code |
| ACT-JEPA ★ | Action | 2025 | paper | — |
| Value-Guided Action Planning with JEPA World Models | Action | 2025 | paper | — |
| VLA-JEPA | ActionLanguage | 2026 | paper | — |
| Causal-JEPA ★ | ActionTheory | 2026 | paper | — |
| Learning Invariant Visual Representations for Planning with JEPA World Models | Action | 2026 | paper | — |
| LeWorldModel ★ | ActionTheory | 2026 | paper | code |
| Hierarchical Planning with Latent World Models | Action | 2026 | paper | code |
| stable-worldmodel | Action | 2026 | paper | code |
| When Does LeJEPA Learn a World Model? ★ | Theory | 2026 | paper | — |
| ThinkJEPA | Theory | 2026 | paper | — |
Biology & Drug Discovery
| Method | Type | Year | Paper | Code |
|---|---|---|---|---|
| BioJEPA-AC ★ | BiologyAction | 2026 | — | code |
| Cell-JEPA ★ | Biology | 2026 | paper | — |
| JEPA-DNA ★ | Biology | 2026 | paper | — |
| ProteinJEPA ★ | Biology | 2026 | paper | — |
Graphs & Molecules
| Method | Type | Year | Paper | Code |
|---|---|---|---|---|
| Graph-JEPA | Graph | 2023 | paper | — |
| Polymer-JEPA ★ | Molecule | 2025 | paper | code |
Medical Imaging & Biosignals
| Method | Type | Year | Paper | Code |
|---|---|---|---|---|
| S-JEPA (Signal-JEPA) | Biosignal | 2024 | paper | — |
| Brain-JEPA | Biosignal | 2024 | paper | — |
| JEPA for ECG Classification | Biosignal | 2024 | paper | — |
| From Video to EEG: Adapting JEPA to Brain Signals | Biosignal | 2025 | paper | — |
| Multimodal JEPA for Imaging and Clinical Signatures ★ | MedImaging | 2025 | paper | — |
| RadJEPA | MedImaging | 2026 | paper | — |
| EchoJEPA ★ | MedImaging | 2026 | paper | — |
| US-JEPA ★ | MedImaging | 2026 | paper | — |
Audio & Speech
3D & Point Clouds
Time Series & Tabular
| Method | Type | Year | Paper | Code |
|---|---|---|---|---|
| LaT-PFN | TimeSeries | 2024 | paper | — |
| T-JEPA (Trajectory Similarity) | TimeSeries | 2024 | paper | — |
| T-JEPA (Tabular) | Tabular | 2024 | paper | — |
| Joint Embeddings Go Temporal | TimeSeries | 2025 | paper | — |
| Koopman Invariants in JEPAs | TimeSeriesTheory | 2025 | paper | — |
| MTS-JEPA | TimeSeries | 2026 | paper | — |
| Giving Sensors a Voice: Multimodal JEPA for Time Series | TimeSeries | 2026 | paper | — |
Earth Observation
Language & Multimodal
Generative Modeling
| Method | Type | Year | Paper | Code |
|---|---|---|---|---|
| D-JEPA (Denoising with JEPA) | Generative | 2024 | paper | — |
| Improving JEPA with Diffusion Noise | Generative | 2025 | paper | — |
| JEPA-T | Generative | 2025 | paper | — |
Datasets & Benchmarks
| Resource | Use | Link |
|---|---|---|
| scPerturb ★ | Harmonized single-cell perturbation datasets (genetic + chemical) for action-conditioned cell world models | sanderlab.org/scPerturb |
| CMap / L1000 ★ | ~1.5M expression profiles across ~5,000 compounds and ~3,000 genetic reagents | clue.io |
| DepMap ★ | Cancer Dependency Map — gene essentiality across 2,000+ models | depmap.org |
| DROID | Large in-the-wild robot manipulation dataset for JEPA world-model planning | droid-dataset |
| Kinetics | Human-action video used to pretrain V-JEPA | paper |
| IntPhys 2 / MVPBench / CausalVQA | Physical-understanding benchmarks for video world models | IntPhys2 |
Code & Frameworks
| Repo | What |
|---|---|
| facebookresearch/ijepa | Official I-JEPA |
| facebookresearch/jepa | Official V-JEPA |
| facebookresearch/vjepa2 | V-JEPA 2 / 2.1 |
| facebookresearch/eb_jepa | Lightweight EB-JEPA library |
| galilai-group/lejepa | LeJEPA + SIGReg |
| galilai-group/stable-worldmodel | Reproducible world-model research stack |
| GPTomics/biojepa ★ | Action-conditioned cell world model |
| AbdelStark/awesome-jepa | Curated JEPA resource index (CC0) |