Personal Page
Projects
Meta-Explaining in Benchmarking Contrastive Categories of Explicability Methods for Graph Neural Networks
I started my Ph.D. by doing a benchmarking on explainability methods of graph neural networks on five downstream tasks: Graph Classification, Node Classification, Entity Resolution, Link Prediction, and Question Answering. The project encompasses state-of-the-art explainability methods from various domains. A comprehensive study is conducted on contrastive categories of explainability methods, while also proposing a solution for aggregating evaluation metrics to identify the superior method for a given problem.
Estimate and Reduce Uncertainty in Uncertain Graphs
We target to advance explainability methods to uncertain graph data and hypergraph-level data. For the case of uncertain data we aim to cover graph data where edges follow probabilistic approaches, and our intention is to use explainability on top of such uncertain data. Such an approach is to extend explainability from deterministic graph data (traditional graph data) to uncertain graph data (complex graph data).
Cyclic Adversarial Framework with Implicit Auto-Encoders and Wasserstein Loss
This was my Master's project which is also is submitted to IEEE TNNLS. In this project a new hybrid model was proposed to address the most important problems of GANs that is Mode Collapsing and Missing Mode. The experimental results of the proposed methods by different evaluation metrics demonstrates promising improvements on the state-of-the-art.
Business Intelligence in Knowledge Management Approaches
In this project, based on the datasets provided by East Azarbaijan Technology Park, a new framework was proposed to improve business intelligence of the companies envisaging knowledge management approaches.
E-Commerce Status Analysis
This project was related to analyzing e-commerce status with respect to economical growth. In this project, the focus was mainly on analyzing any consequential impact of new e-commerce on economical growth, or how economical condition reacts to introduction of new e-commerces, and what are the main challenges for them.
Benchmarking on Evaluational Metrics of XAI on Graph Neural Networks
While doing a benchmarking on explicability methods of graph neural networks, it was felt that a benchmarking on the evaluation metrics of explainability methods is also missing. In this project, the goal is to not only gather all the evaluation metrics of explainability methods together, but also to analyze which evaluation metric/s work best for which explainability method.
