Graphic is from the paper “Maximizing Cohesion and Separation in Graph Representation Learning: A Distance-aware Negative Sampling Approach”

Having the opportunity to apply state-of-the-art machine learning models to bioinformatics problems as an undergraduate motivated M. Maruf to take a deep dive into machine learning and deep learning as a Ph.D. student in computer science at Virginia Tech which he chose because of its exemplary research and top-notch facilities. 

“Dr. Anuj Karpatne’s unique view towards solving real-world problems fascinated me to explore more knowledge-infused machine learning,” Maruf said of his advisor at the Sanghani Center.

Last April, Maruf presented their collaborative paper, “Maximizing Cohesion and Separation in Graph Representation Learning: A Distance-aware Negative Sampling Approach,” at the SIAM International Conference on Data Mining (SDM).

Maruf’s research interests lie in the broad domains of science-guided machine learning and its applications with a focus on integrating domain knowledge into machine learning models to obtain generalized solutions consistent with scientific knowledge. 

“In particular, I am developing new algorithms for graph neural networks that allow for better representation with coherent knowledge propagation,” he said.

A  black-box neural network model learns solely from training samples and requires a lot of annotated real-world observations to learn the underlying patterns accurately, said Maruf. Moreover, black-box Artificial Neural Networks (ANN) ignore external biological knowledge in the training phase, resulting in inconsistent outputs.

“I am currently addressing these challenges for the fish trait segmentation problem by incorporating biological knowledge into the state-of-the-art ANN model,” he said.

Maruf presented “Biology-Guided Neural Network for Fish Trait Discovery,” at the Society for Integrative and Comparative Biology Virtual Annual Meeting earlier this year.

The Sanghani Center environment, Maruf said, provides its students with multidisciplinary learning and research collaboration opportunites.

His additional work with faculty and other Ph.D. students at 2021 conferences includes

“PID-GAN: A GAN Framework based on a Physics-informed Discriminator for Uncertainty Quantification with Physics”in proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), held in August; and “Learning Compact Representations of Neural Networks using DiscriminAtive Masking (DAM),” which will be included in proceedings of the 35th Conference on Neural Information Processing Systems (NeurIPS) in December.

Maruf received a bachelor’s degree in computer science and engineering from Bangladesh University of Engineering and Technology.  

His projected graduation date is Spring 2021 and he plans to pursue an industrial research position.