Mohammad Tariqul Islam
MIT-Novo Nordisk Artificial Intelligence Fellow

I am a Postdoctoral Fellow at Massachussetts Institute of Technology. Currently working with Deblina Sarkar at the Nano Cybernetic Biotrek lab. Previously, I was PhD Candidate at the Department of Electrical and Computer Engineering in Princeton University. I was mentored by Prof. Jason Fleischer in the Imaging Physics Lab and worked on artificial intelligence, machine learning, imaging, and data analysis for phenotyping disease and characterizing large image repositories. Prior to that, I completed bachelors and masters at Bangladesh University of Engineering and Technology (BUET), where I worked on deep learning, training image to image translation networks for image denoising, multimodal learning for emotion detection, and physiological signal processing for hand wearable.

Email  /  CV  /  Google Scholar  /  Github  /  Hobby

My primary research interests include artificial intelligence, machine learning, signal processing, and imaging.

The research topic of my PhD involves unsupervised nonlinear dimensionality reduciton (UMAP, t-SNE) methods and their applications in biomedical data analysis.

In this paper, we analyze the forces to reveal their effects on cluster formations and visualization. Repulsion emphasizes differences, controlling cluster boundaries and inter-cluster distance. Attraction is more subtle, as attractive tension between points can manifest simultaneously as attraction and repulsion in the lower-dimensional mapping. This explains the need for learning rate annealing and motivates the different treatments between attractive and repulsive terms. Moreover, by modifying attraction, we improve the consistency of cluster formation under random initialization. Overall, our analysis makes UMAP and similar embedding methods more interpretable, more robust, and more accurate.

We propose Manifold-approximated Kernel Alignment (MKA) that incorporates manifold geometry into the alignment task. We derive a theoretical framework for MKA, perform empirical evaluations on synthetic datasets and real-world examples to characterize and compare MKA to a popular alternative Centered Kernel Alingment (CKA). Our findings suggest that manifold-aware kernel alignment provides a more robust foundation for measuring representations, with potential applications in representation learning.

We show that neighbor embedding algorithms can find anomalies from large chest X-ray datasets, essentially by forming independent clusters that are distinct from the main (“good”) data but similar to other points with the same error type.

We explore performance of gradient descent algorithm for two-point super-resolution under abberations in the optical path. We developed PySuperRes, a python package, to model PSFs under noise and abberation, and to estiamte PSF positions.

We extend UMAP for manifold alignment.

Egocentric detection of fingertips from videos recorded in a VR headset.

We constructed a dataset for predicting advertisement likebality by recording user reaction while watching ads. Then, we employ regression analysis (MSE and concordance correlation) using LSTM and estimate the viewers' affective response.

Using UMAP, we show that x-ray features reveal two distinct phenotype in COVID-19 response.

We analyze COVID-19 chest x-ray features for signatures of L and H phenotypes of COVID-19.

Modified from Jon Barron's website. Images in Misc sections are taken from xkcd.