NYU Data Science newsletter – September 23, 2016

GROUP CURATION: N/A

A study of hundreds of new genomes from across the globe has yielded insights into modern human genetic diversity and ancient population dynamics, including compelling evidence that essentially all non-Africans today descend from a single migration out of Africa.

The multinational research effort, led by Harvard Medical School geneticists and published Sept. 21 in Nature, also suggests that no single genetic change or small group of changes is likely to explain the substantial transformations in human culture and cognition seen in the last 50,000 years.

The anatomy of an architecture to bring data science into production.

President Obama’s chief economist argues that, with the right policies, artificial intelligence can be boon to the labor market, not a threat.

The company’s platform is built around a number of core founding principles. At the root is a desire to remove friction from the insurance process, cutting bureaucracy, time needed, and costs. But Lemonade is also setting out to combat existing models through an annual “giveback,” where it donates unclaimed money to good causes. Through the app, users select a cause that they care about, and this cause-creation process generates virtual groups of like-minded people — or “peers.”

The popular handmade-goods site Etsy announced it has acquired a startup called Blackbird Technologies, which developed algorithms for natural language processing, image recognition and analytics — similar to those used by Amazon and Google for product and other searches — and then “democratized” them to be used by any company of any size. At Etsy, the tech will be used to improve its own search features.

The National Institutes of Health today announced that the Johns Hopkins Bloomberg School of Public Health, along with the research firm RTI International, will receive a seven-year, $95 million grant to analyze the data from its new Environmental Influences on Child Health Outcomes (ECHO) program, an initiative designed to understand how the environment influences health beginning in the womb.

Open source release “contains significant improvements to the computer vision component of the captioning system, is much faster to train, and produces more detailed and accurate descriptions” Git repo at the bottom of the blog linked here.

San Jose, CA The deadline for speaker proposals is Friday, 30 September 2016.

Deadline for paper submissions is 24 October 2016.

“From 80,000+ stations in 180 countries, spans decades and has been quality-checked to ensure that it’s temporally and spatially consistent.”

We applied a neural network to the task of predicting numerical ratings for text-based consumer reviews, training the model to learn the ratings directly from the words in each review. It turns out that our model had decent prediction accuracy, but this wasn’t our objective, especially since we already knew the ratings for these reviews. Instead, we wanted to understand why those particular ratings were assigned. Interpreting what a neural network has learned from data is a separate endeavor from applying it to make predictions. In this post, we’ll explore some strategies for bringing the inside of a neural network to light, using our ratings prediction model to demonstrate.