Scientist Yi Nian is sharing his machine-learning expertise with the world in his latest co-authored publication, “Globally Interpretable Graph Learning via Distribution Matching.” Published by 2024 ACM Web Conference, the paper explores the emergence of graph neural networks (GNNs) as a powerful model to capture critical graph patterns and answers the important question of how to provide a global interpretation for the graph learning procedure.
Nian’s newest publication dives into an actual explanation of the model behavior of GNNs, instead of treating them as black boxes in an end-to-end fashion. This is contrary to existing works that have primarily focused on local interpretation to reveal the discriminative pattern for each individual instance, which does not directly reflect the high-level model behavior across instances.
For the sake of gaining global insights into graph learning, Nian and his collaborators focused on high-level and human-intelligible patterns that dominate the GNN learning procedure. Their preliminary analysis shows that interpretative patterns generated by existing global methods fail to recover the model training procedure, leading to their proposed solution — Graph Distribution Matching (GDM).
GDM synthesizes interpretive graphs by matching the distribution of the original and interpretive graphs in the GNN’s feature space as its training proceeds, capturing the most informative patterns the model learns during training. Extensive experiments on graph classification datasets demonstrated multiple advantages of the proposed method, including high model fidelity, predictive accuracy and time efficiency, as well as the ability to reveal class-relevant structure.
“This latest publication is a reflection of my drive to further understand and educate about the complexities of GNNs, and provide a global perspective that has not yet been discussed in other academic papers,” said Nian. “Much like my other machine-learning research and resulting publications, this latest endeavor reveals new insights and understanding that will help propel our industry forward.”
Nian has worked as an Applied Scientist with Amazon since 2022, In this role, he created science-driven solutions using large language models (LLM) and integrate new feature into AWS Transcribe with engineerings. Prior to his role at Amazon, Nian was a Data Science Research Assistant for UTHealth Science Center in Houston, Texas, where he constructed knowledge graphs with NLP tools from Alzheimer’s Disease literature. He conducted research in state-of-the-art graph neural network methods for biomedical knowledge graphs and designed innovative graph mining methods for drug repurposing and drug discovery.
Nian holds a Master’s degree in computer science from the University of Chicago, and a Bachelor’s degree in mathematics from Ohio State University.
The full publication, “Globally Interpret HYPERLINK “https://arxiv.org/abs/2306.10447″a HYPERLINK “https://arxiv.org/abs/2306.10447″ble Graph Learning via Distribution Matching HYPERLINK “https://arxiv.org/abs/2306.10447”, HYPERLINK “https://arxiv.org/abs/2306.10447″” can be read online in full detail.