< Back to All Events
|
Shankari Subramanyam Impact Grant (SSIG) seminar - Inside Graph Neural Networks From Message Passing to Predicting the Critical Point of Molecules
Speaker Name
Dr. Ghanshyam Pilania, Principal Engineer and Technology Manager, GE Aerospace Research Center in Niskayuna, New York
November 5, 2026
02:00 PM
|SPEAKER BIO
Dr. Ghanshyam serves as a Principal Engineer and Technology Manager for Probabilistic Design and Materials Informatics Team within the Intelligent Systems Organization at the GE Aerospace Research Center in Niskayuna, New York.
|ABSTRACT
This session will be of hands-on workshop on development of Graph neural networks for materials informatics.
PART 1- The Machinery
Graphs, adjacency matrices & node/edge embeddings
The message-passing template: GCN,GraphSAGE, GAT, edge-aware MPNNs
Readout, training & multi-task learning
Permutation invariance: why every architecture looks the way it does
PART 2 - Predicting Tc, Pc and Tb
Why three numbers unlock the whole vapor-liquid phase diagram
A curated, provenance-tracked dataset of 10,503 molecules
Build a directed message-passing network in ~60 lines of PyTorch
Every training knob, tested in controlled experiments
