A View Of The Sky And A NeurIPS Paper

In 2017, nearing the end of my first year at Harvard Business School, I had the opportunity to travel to the Philippines for a class. While we were put up in a lavish hotel in an affluent part of Manila, we traveled to various neighborhoods in and around Manila. I remember visiting parts of the city that were so prim and pristine and other parts that were so impoverished, it seemed jarring to think that these neighborhoods were a few miles away apart.

Some photos of the gorgeous Manila sky taken during my trip

This is nothing new. The chasmic disparity in wealth and quality of life between the affluent and the not so affluent is apparent in virtually every country of the world. But something about experiencing it in a foreign country, where I was merely a hired observer and not a direct participant in society, made me see minute details that had so far escaped me. Among other reflections, I wrote the following to myself.

"Have you ever thought about the person who lives in the 30th floor penthouse apartment has a much better view of the sky, of the sun, the clouds, the moon and stars, than the person who lives in a plastic tent on the road, who only sees a little bit of sky among the tops of large buildings. How strange that it has become a privilege to enjoy a uncluttered view of the sky. People less fortunate get less and less of everything in society but they are also losing something as basic and essential to the human spirit as the ability to watch clouds float by."

I later shared this with Saquib, my husband, who thought it would be interesting to try and quantify my qualitative observation. He and one of his mentees, Halima, used Google Street View and computer vision techniques to compare the visibility of the sky and the level of greenery in different neighborhoods in Dhaka, Bangladesh. They also classified the affluence of the neighborhoods by mining for rent data.

The research findings were intriguing. While the visibility of the sky from the street level was indeed higher in affluent neighborhoods, these areas tended to have less greenery compared to less well-to-do areas. More importantly, this work showed that such automated means of quantifying landscape features could be a useful tool in urban planning.

This work was accepted in NeurIPS 2018. Besides contributing the idea for this project, I also partook in writing parts of this paper with Saquib and Halima, earning myself co-authorship in the process.

Feel free to check out the paper here and let me know what you think.