https://github.com/aurelienperez/sciphy-rl-for-doctr-l

Offline Reinforcement Learning with neural PDEs.
https://github.com/aurelienperez/sciphy-rl-for-doctr-l

deep-learning neural-pde reinforcement-learning stochastic-control

Last synced: 18 days ago
JSON representation

Offline Reinforcement Learning with neural PDEs.

• Host: GitHub
• URL: https://github.com/aurelienperez/sciphy-rl-for-doctr-l
• Owner: aurelienperez
• License: mit
• Created: 2025-03-18T16:56:09.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2025-03-26T14:40:59.000Z (about 1 year ago)
• Last Synced: 2025-04-09T14:37:21.390Z (about 1 year ago)
• Topics: deep-learning, neural-pde, reinforcement-learning, stochastic-control
• Language: Python
• Homepage:
• Size: 8.1 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Distributional Offline Continuous-Time Reinforcement Learning with Neural Physics-Informed PDEs

![DOCTRL_pres_image-2](https://github.com/user-attachments/assets/7ed7a90a-42e0-4ac2-b7d7-e95779c9981a)

This repository provides a full implementation of the **DEEP DOCTR-L** algorithm, based on:

**“Distributional Offline Continuous-Time Reinforcement Learning with Neural Physics-Informed PDEs” – Igor Halperin (2023)**

It was developed as part of the course *Machine Learning and Stochastic Control* in the Master **Probabilités et Finance** at Sorbonne Université.

The repository includes:

- The complete code to reproduce the original experiments described in the paper

- A main script to run the full pipeline: data generation, model training, and evaluation

## 📄 Resources

- **[Original Paper](paper/Halperin_2023_DOCTR-L.pdf)**

- **[Project Report](report/DOCTR-L_Report.pdf)** — presents the method, key derivations, and additional experiments comparing DOCTR-L to a semi-closed-form Riccati solution

> 🔧 The Riccati-based experiments from the report are **not included** in this repository.