Problem
Graph structure learning asks how to infer useful relational structure from data rather than assuming that the graph is already known.
Back to research
NSysS 2025
ChainGraph-DL: Data-Driven Causal Inference for Graph Structure Learning
A graph-learning paper by Soily Ghosh Sneha, Md. Tanzeem Rahat, and Md. Manzurul Hasan, accepted at the 12th International Conference on Next Generation Computing, Communication, Systems and Security.
Why It Matters
Causal and graph-based learning can make AI systems more interpretable, especially when relationships among variables are as important as individual predictions.
Core Idea
ChainGraph-DL connects data-driven causal inference with graph structure learning, placing the work at the boundary of graph AI, causality, and explainable modeling.
Main Result
The paper is listed as an accepted full paper for NSysS 2025 and appears on the Scholar publication list supplied for this site.
What Was Difficult
The main modeling challenge is learning meaningful structure from data while avoiding spurious relationships that look predictive but are not causally useful.
Resources
Open Directions
- How can graph structure learning distinguish causality from correlation?
- Which evaluation protocols best measure learned graph usefulness?
- Can graph explanations be made actionable for domain users?