csccm-iitd / stochastic-simulators

Overview

A deep learning based surrogate model for stochastic simulators
Akshay Thakur and Souvik Chakraborty

TensorFlow implementation of deep learning-based surrogate model for stochastic simulators. Generative neural network is used to approximate the stochastic response. A simple feed-forward neural network is used with a conditional maximum mean discrepancy (CMMD) loss-function. CMMD allows to capture the discrepancy between the true response of the stochastic simulator and the distribution predicted by the neural network.

Figure: Schematic representation of the proposed deep learning framework based on CGMMN.

Dependencies

• tensorflow 2.8.0
• python 3.x
• numpy
• matplotlib
• pandas
• scipy
• scikit-learn

Installation

• Install TensorFlow and other dependencies.
• Clone the repository using

git clone https://github.com/name_add/Deep-Learning-Based-Surrogate-Model-for-Stochastic-Simulator.git

Dataset for Training

The dataset for the problems of SDE without closed form solution and Stochastic SIR could be generated using the code in Data folder of this repository. For the remaining two problems of 1-D and 2-D Stochastic Simulator problems with closed form solution, the code for dataset generation is inside the respective python scripts.

Citation

If this respository helped you, please considering the citing the following work: