Hello, World

I'm Abdullah Nauman.,a programmer.,a thinker.

I like to make things; and break things.

View Cv

About Me

I build things, do AI research, and bring communities together.

I immigrated from Pakistan in 2011. In America, I was exposed to computers for the first time and fell in love with them. I enjoy building Science Olympiad devices and participating in hackathons.

Two months after I moved here, a major terrorist was found in my home country. Islamophobia has been on the rise for several years now, but after events Christchurch, NZ, I realized that we could no longer stand by the sidelines.

Teaming up with other students at Ward Melville H.S, I've started ComeTogether. We build and use models to understand where Islamophobia stems from and use these insights to disengage misconceptions in our community. Since March, we've positively affected over 120 people. ComeTogether is growing fast because we're not just a movement; we're an actionable idea for positive change.

My Research

I had the amazing opportunity to work under Dr. Abid Malik at Brookhaven National Laboratory

Mititgating Comunication Bottlenecks in MPI-Based Distributed Learning

July 2019 Brookhaven National Laboratory

Presented at the Supercomputing Conference in November.

Profiled and analyzed existing distribution framework for neural networks used by researchers and found significant performance bottlenecks due to communication. Proposed solution to mitigate bottlenecks by sending parameter updates in parallel with loading data for the next training batch.

Read Full Abstract

Distributed Generative Adversarial Networks

Sept. 2019 Brookhaven National Laboratory

Generative Adversarial Networks can help develop valuable models on complex data. However, as data volume grows, model training times increase significantly. To help tackle this problem in the long term, we successfully distribute the process of training a Generative Adversarial Model across several GPUs and demonstrate 68% parallel effenciency.

Full Abstract Coming Soon

A docking system for permanent ASV installation

Ongoing Independent Research Project

ASVs serve a range of functions from collecting data to monitoring systems. However, endurance poses a key bottleneck: batteries run out and fuel tanks dry up. We present a novel system for permanent ASV installation using computer vision based navigation to charging dock and wireless charging.

Read Full Abstract | See Video