Geometric Deep Learning
I built a temporal graph convolutional network (T-GCN) model to detect seizures in electroencephalogram (EEG) data. T-GCN models can detect the time of a seizure and its region in the brain by combining graph convolutional network (GCN) and recurrent neural network (RNN) architectures. Unlike convolutional neural networks (CNNs), T-GCNs can handle the non-Euclidean signals in the EEG data. Professor Weiqing Gu and I are currently working to integrate EEG and gait data with geometric deep learning methods to diagnose Parkinson’s disease and analyze its progression.
Our work focused on generalization bounds in statistical learning. Because the paper is currently under review, I have removed detailed information about it. The paper is currently under review at the 33rd International Conference on Algorithmic Learning Theory (ALT 2022), and I am the second of four authors on the submission.
We worked on a probabilistic theory of abductive reasoning. Specifically, we developed a model that unifies selective and creative abduction by focusing on common cause abduction. Our model incorporates principles of causation by modeling abductive reasoning through a Bayesian network. I integrated selective and creative abduction with causal principles by developing two algorithms, which allow the model to compute novel and common-cause explanations for observations. I also developed one of the two similarity metrics we used, derived from the Jaccard index and edit-distance, in order to compute the similarity of graph nodes. I am the first of four authors on the paper published at the 13th International Conference on Agents and Artificial Intelligence (ICAART 2021).