What Inception Net Doesn't See

Written on February 12, 2021
Deep learning-based comuter vision models like Inception Net have achieved state-of-the-art performance on image recognition. However, that doesn't mean that they don't have blindspots and biases. Here's a few of them, along with interactive aplications for you to try it out yourself. Read More

12 Atomic Experiments in Deep Learning [Notebook]

Written on April 8, 2019
Deep learning remains somewhat of a mysterious art even for frequent practitioners, because we usually run complex experiments on large datasets, which obscures basic relationships between dataset, hyperparameters, and performance. The goal of this notebook is to provide some basic intuition of deep neural networks by running very simple experiments on small datasets that help understand trends that occur generally... Read More

Which Classifier Is the Best on 18 UCI Datasets? [Notebook]

Written on April 1, 2018
We all know that deep neural networks (DNNs) are great for image recognition and speech processing. What about good ol' numerical datasets? I compared DNNs to other standard ML algorithms on many public classification datasets from the UCI ML repository, and here are the results. Read More

Gently Building Up The EM Algorithm

Written on March 1, 2018
Today, I spent some understanding the Expectation-Maximization (EM) algorithm. It was the 8th or 9th time I had tried to understand it, because even though there are many nice tutorials about it online, about three quarters of the way through them, my eyes start to glaze over the math, and I end up leaving with only a “high-level understanding” of... Read More

Moral Machine Learning

Written on December 12, 2017
Is machine learning research making the world a better place? I think about this sometimes when I attend conferences and see poster after poster proving theoretical guarantees for machine learning algorithms that already work well, or algorithm after algorithm that claims to achieve records results on standard image recognition datasets. Read More

Stop Doing Fragile Research

Written on October 18, 2017
Many of us, consciously or subconsciously, treat our research as fragile, when in reality, we should be stress-testing it. Here, I share 3 ways to make the methods we develop more robust. Read More

Can a Machine Learn to Classify Meccan and Medinan Surahs?

Written on January 16, 2017
A basic concept in Quranic studies is the difference between its meccan surahs/chapters (revealed before the Prophet Muhammad migrated from trading city of Mecca to oasis town of Medina) and its medinan chapters (revealed after the migration). Read More

An Introduction to Deep Learning in Medicine and Biology

Written on November 3, 2016
Over the last few years, researchers have applied deep learning to all sorts of datasets in biology and medicine. On almost every problem – from identifying protein-DNA interactions to diagnosing Alzheimer’s disease from brain scans – deep learning techniques have performed remarkably well, leaving traditional machine learning in the dust. Read More

Interview Questions for the Soros Fellowship

Written on July 16, 2016
I was recently selected to be one of the 30 students that received the Paul and Daisy Soros Fellowship for graduate students. Going in, the interview for the fellowship was quite the black box for me – I had no idea what questions to expect, so I spent about 12 hours over the span of two weeks, practicing all sorts... Read More