See the documentation at https://micdoh.github.io/XLRON/
As presented at Optical Fibre Communication Conference (OFC) 2024 - see the paper here
@INPROCEEDINGS{10526884,
author={Doherty, Michael and Beghelli, Alejandra},
booktitle={2024 Optical Fiber Communications Conference and Exhibition (OFC)},
title={XLRON: Accelerated Reinforcement Learning Environments for Optical Networks},
year={2024},
volume={},
number={},
pages={1-3},
keywords={Training;Graphics processing units;Reinforcement learning;Optical fiber networks;Resource management},
doi={}}
XLRON ("ex-el-er-on") is an open-source project that provides a suite of gym-style environments for simulating resource allocation problems in optical networks and applying reinforcement learning (RL) techniques. Unlike similar libraries, it is built on the JAX machine learning framework, enabling accelerated training on GPU/TPU/XPU hardware. This gives orders of magnitude faster training than other optical network simulation libraries (e.g. optical-rl-gym, DeepRMSA, RSA-RL) due to:
- JIT compilation of the entire training loop
- Massive parallelism (1000s of parallel environments) on accelerator hardware
- Co-location of environment and agent on GPU avoids CPU-XPU data transfer bottleneck
- Bypasses the Python Global Interpreter Lock (GIL)
XLRON is a product of my PhD research, which focuses on a set of combinatorial optimisation problems related to resource allocation in optical networks. XLRON aims to overcome the computational bottleneck in applying RL to these problems. The project is in active development.
This work was supported by the Engineering and Physical Sciences Research Council (EPSRC) grant EP/S022139/1 - the Centre for Doctoral Training in Connected Electronic and Photonic Systems - and EPSRC Programme Grant TRANSNET (EP/R035342/1)
Copyright (c) Michael Doherty 2023. This project is licensed under the MIT License - see LICENSE file for details.