Learning Interpretable Features by Tensor Decomposition

Date: 
Thursday, February 6, 2020 - 11:00
Location: 
Blakeslee Room
Presenter: 
Dr. Shah Muhammad Hamdi

Representation learning of the nodes in a graph has facilitated many downstream machine learning applications such as classification, clustering, and visualization. Existing algorithms generate less interpretable feature space for the nodes, where the roles of the features are not understandable. This talk covers the use of multi-dimensional arrays or tensors in node embedding. I will explain how tensor decomposition-based node embedding algorithms consider local and global structural similarities of the nodes, learn the proximity itself, require less number of tunable hyperparameters, and generate a feature space where the feature roles are understandable, while working on different types of static networks. In addition to the social networks, I will show another application in the neuroscience domain, more specifically on the brain networks found from the resting-state fMRI data of healthy and disabled subjects, where nodes represent brain regions, and edges represent functional correlation among them. I will discuss the use of tensor decomposition in the representation learning of the biomarkers of the neurological diseases, which are the discriminative nodes and edges of the brain networks that can distinguish the healthy population from the disabled population. I will demonstrate some experimental findings on social networks and brain networks, and the potentials of this approach in one research problem of solar physics, which is the multi-variate time series-based solar flare prediction.

 




Biography


Shah Muhammad Hamdi is a PhD candidate in the Department of Computer Science of Georgia State University. His research interests are machine learning, data mining and deep learning, more specifically, finding interesting patterns from real-life graphs and time-series data. His research finds applications in the fields of social networks, neuroscience, and solar physics. He has publications in top data mining conferences such as IEEE ICDM, ACM CIKM, and IEEE Big Data. He worked as a data scientist intern at Amazon Web Services Inc. (AWS) and LexisNexis Risk Solutions. Before starting his PhD, he worked as a Lecturer in Computer Science in Northern University Bangladesh, Dhaka, Bangladesh. He received his Bachelor's degree in Computer Science in 2014 from Rajshahi University of Engineering and Technology (RUET), Rajshahi, Bangladesh.