Skip to content

This work implements "SigMA: Path Signatures and Multi-head Attention for Learning Parameters in fBm-driven SDEs" by Xianglin Wu, Cornelis W. Oosterlee, and Chiheb Ben Hammouda.

Notifications You must be signed in to change notification settings

changanluoxue/SigMA

Repository files navigation

SigMA

This work implements "SigMA: Path Signatures and Multi-head Attention for Learning Parameters in fBm-driven SDEs" by Xianglin Wu, Chiheb Ben Hammouda, and Cornelis W. Oosterlee. https://arxiv.org/abs/2512.15088

🚀 Prerequisites

All experiments were run on a system with CUDA12.8, utilizing GPU acceleration. The specific versions of the key libraries were: PyTorch2.8.0, Python3.10.12.

🏗️ Project Structure

The "example" files correspond to the seven different experiments in the article. Specifically, "example1" to "example5" use synthetic data, and "example6" and "example7" use real data. The "paths_simulation" file is used to generate the synthetic data for "example1" to "example5". The "market_data" folder contains the real data used by "example6" and "example7". Within the "example" files, setting name = "training" is for training the neural networks, and setting name = "results" or name = "calibration" is for outputting the experimental results. The "layers" and "NNmodels" files together define the neural network model structures required for all experiments. The "plots" file is used for generating the figures in the paper. The "utils" file defines the trainer functions.

About

This work implements "SigMA: Path Signatures and Multi-head Attention for Learning Parameters in fBm-driven SDEs" by Xianglin Wu, Cornelis W. Oosterlee, and Chiheb Ben Hammouda.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages