| --- |
| tags: |
| - ml-intern |
| --- |
| # FinJEPA: Financial Joint-Embedding Predictive Architecture |
|
|
| FinJEPA is a JEPA-based world model for portfolio optimization over a **separated action space** consisting of: |
| - **Portfolio weights** (continuous, simplex-constrained) |
| - **Trading signals** (discrete: long/short/flat per asset) |
| - **Hedging signal** (binary: on/off) |
|
|
| ## Architecture (SOTA Blend) |
|
|
| | Component | Source | Key Innovation | |
| |-----------|--------|----------------| |
| | Time Series Encoder | TS-JEPA (Sennadir 2025) | 1D-CNN patch tokenizer + Transformer | |
| | Action Conditioning | JEPA-WMs (Terver 2025) | AdaLN + RoPE in predictor | |
| | Collapse Prevention | EB-JEPA (Terver 2026) | SIGReg + Inverse Dynamics Model | |
| | Multi-step Rollout | EB-JEPA | K-step autoregressive training | |
| | Planner | JEPA-WMs + EB-JEPA | CEM L2 cost / MPPI cumulative cost in latent space | |
| | TD Branch | TD-JEPA (Bagatella 2025) | Optional separate task encoder for zero-shot RL | |
|
|
| ## Model |
|
|
| ``` |
| Financial Time Series (T, F) |
| │ |
| ▼ |
| [TimeSeriesTokenizer] ── 1D-CNN patches + position encoding |
| │ |
| ├───► [Context Encoder] (student) |
| │ │ |
| │ ▼ |
| │ [Predictor] ◄─── Action embedding (weights + signals) |
| │ (AdaLN + RoPE) │ |
| │ │ │ |
| │ ▼ ▼ |
| │ Predicted target [ActionEmbedder] |
| │ embeddings ├── weights (continuous) |
| │ ├── signals (discrete) |
| │ └── hedge (binary) |
| │ |
| └───► [Target Encoder] (teacher, EMA frozen) |
| │ |
| ▼ |
| Ground truth target embeddings |
| ``` |
|
|
| ## Usage |
|
|
| ```bash |
| python finjepa/run_training_fast.py |
| ``` |
|
|
| Full training on real data: |
| ```bash |
| python finjepa/train.py --data_source hf \ |
| --dataset_name paperswithbacktest/Stocks-Daily-Price \ |
| --n_assets 5 --batch_size 128 --epochs 50 --push_to_hub |
| ``` |
|
|
| ## References |
|
|
| 1. TS-JEPA — Sennadir et al. (2025). arxiv:2509.25449 |
| 2. JEPA-WMs — Terver et al. (2025). arxiv:2512.24497 |
| 3. EB-JEPA — Terver et al. (2026). arxiv:2602.03604 |
| 4. V-JEPA 2 — Assran et al. (2025). arxiv:2506.09985 |
| 5. TD-JEPA — Bagatella et al. (2025). arxiv:2510.00739 |
|
|
| <!-- ml-intern-provenance --> |
| ## Generated by ML Intern |
|
|
| This model repository was generated by [ML Intern](https://github.com/huggingface/ml-intern), an agent for machine learning research and development on the Hugging Face Hub. |
|
|
| - Try ML Intern: https://smolagents-ml-intern.hf.space |
| - Source code: https://github.com/huggingface/ml-intern |
|
|
