Skip to content

himanshuvnm/GIFT_KASTL_POSTER

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

37 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

2025 SciFM Conference Research Poster GIFT-KASTL

Graph Foundation Models for Fracture Network Learning

Scientific Machine Learning • Graph AI • Foundation Models

Research Poster Scientific ML Graph AI License


Overview

This repository hosts the research poster for the GIFT-KASTL project, exploring the use of graph foundation models for fracture network learning in scientific machine learning.

Fracture systems arise in many scientific and engineering domains including:

• subsurface modeling
• geophysics
• material science
• fluid flow systems

Modeling these systems presents unique challenges due to their complex topology, high dimensional geometry, and large-scale simulation costs.

The GIFT-KASTL framework investigates how graph-based machine learning architectures can capture the structure of fracture networks and enable data-driven surrogate modeling of physical systems.


Poster

Download the full poster


Poster Preview


Motivation

Traditional physics-based simulation methods for fracture networks often require expensive numerical solvers.

Machine learning offers an alternative paradigm:

learning representations directly from structured data.

Graph-based representations are particularly well-suited for fracture systems because they preserve:

• connectivity
• topology
• spatial relationships

Recent advances in graph neural networks and foundation models suggest new possibilities for scalable scientific learning.


Approach

The GIFT-KASTL framework investigates a pipeline consisting of:

Graph Representation

Fracture systems are encoded as graphs where

• nodes represent fracture elements
• edges represent interactions and connectivity

This representation naturally captures structural relationships.


Learning Architecture

The framework explores graph-based neural architectures capable of modeling complex interactions in fracture systems.

Key elements include:

• graph representation learning
• scalable neural architectures
• structured scientific datasets


Scientific Machine Learning

The system aims to support:

• surrogate modeling
• predictive simulation
• data-driven physical modeling


Applications

Potential application areas include

• fracture propagation prediction
• subsurface simulation
• material failure modeling
• accelerated scientific simulation


Author

Himanshu Singh

Research interests include:

• scientific machine learning
• graph foundation models
• sparse learning
• mechanistic interpretability


Related Research Projects

You may also find these related repositories interesting:

• Mechanistic AI Interpretability
• Sparse Neural Networks
• Fourier Neural Operator Learning

Explore more at:

https://github.com/himanshuvnm


Citation

If you use or reference this work, please cite the repository:


Contact

GitHub
https://github.com/himanshuvnm

LinkedIn
https://linkedin.com/in/himanshuvnm


Scientific Machine Learning • Graph AI • Foundation Models

About

https://himanshuvnm.github.io/gift-kastl/ 2025 SciFM Conference Research Poster on the graph foundation model on the Discrete Fracture Networks dataset. We introduced a novel neural net layer to achieve high-end approximation based on the Kolmogorov-Arnold Superposition Theorem which now is also used in constructing KAN.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors