Saturday, May 14, 2016

Sunsets, Fraternities, and Deep Learning

Some projects I’ve been working on recently:

Sunset Nerd: a beautiful sunset detector. It’s a robot that monitors the number of posts tagged #sunset on Instagram (a picture sharing website) and sends you a message when there is a beautiful sunset happening in your area so you don’t miss it. If you live in a city -- currently the app works for San Francisco, NYC, Philadelphia, Los Angeles, Palo Alto, Washington DC, Chicago, Houston, San Diego, Miami, Indianapolis, Jacksonville, Boston, and Seattle -- click here to try it out on Facebook -- just send a message! (It will only message you once a week on average, and you can unsubscribe at any time.) For example, Boston on February 5 had a gorgeous pink sunset as a snowstorm cleared, and the detector picked it up; the red line shows the number of posts on that day, and the grey lines show the normal number of posts.

Fraternities: My New York Times piece, a shorter version of the piece originally posted on this blog; Total Frat Move’s response.

Computational biology: I’ve gotten to work on two papers, one with Pang Wei Koh and Anshul Kundaje on using convolutional neural networks to denoise epigenetic data, to be presented as a spotlight talk at ICML’s Computational Biology Workshop, and one with Bo Wang, Junjie Zhu, and Serafim Batzoglou on a method for clustering single-cell RNA-seq data, which has been presented at Cold Spring Harbor.

Relationship abuse dataset: if this is a research interest of yours, I have a dataset of more than 100,000 Twitter posts under the hashtags #MaybeHeDoesntHitYou and #MaybeSheDoesntHitYou. Filtering on tweets with form "#MaybeHeDoesntHitYou but" yields a large dataset of people's experiences with non-physical relationship abuse. Shoot me an email (emmap1 “at” if you have project ideas -- no time to pursue this project on my own, but happy to discuss collaboration.

1 comment:

  1. One of my data science students did a cool final project this term using sunset forecasts from SunsetWx: