NYU Data Science newsletter – February 5, 2016

GROUP CURATION: N/A

A new $12 million dollar project at Carnegie Mellon University could make machine learning even more powerful by uncovering ways to teach computers more efficiently while using much less data.

The five-year effort will use a newish technique, called 2-photon calcium imaging microscopy, to study the way visual information is processed in the brain. The funding comes through President Obama’s BRAIN Initiative, and it is a good example of one of the near-term benefits that powerful new brain imaging techniques could have.

The White House has denied a security clearance to a member of its technology team who previously helped report on documents leaked by Edward Snowden.

Ashkan Soltani, a Pulitzer prize-winning journalist and recent staffer at the Federal Trade Commission, recently began working with the White House on privacy, data ethics and technical outreach. The partnership raised eyebrows when it was announced in December because of Soltani’s previous work with the Washington Post, where he helped analyze and protect a cache of National Security Agency documents leaked by Snowden.

How predictable is success in complex social systems? In spite of a recent profusion of prediction studies that exploit online social and information network data, this question remains unanswered, in part because it has not been adequately specified. In this paper we attempt to clarify the question by presenting a simple stylized model of success that attributes prediction error to one of two generic sources: insufficiency of available data and/or models on the one hand; and inherent unpredictability of complex social systems on the other. We then use this model to motivate an illustrative empirical study of information cascade size prediction on Twitter.

Huddled in a coffee shop one drizzly Seattle morning six years ago, the astrobiologist Shawn Domagal-Goldman stared blankly at his laptop screen, paralyzed. He had been running a simulation of an evolving planet, when suddenly oxygen started accumulating in the virtual planet’s atmosphere. Up the concentration ticked, from 0 to 5 to 10 percent.

“Is something wrong?” his wife asked.

“Yeah.”

The rise of oxygen was bad news for the search for extraterrestrial life.

… machine learning is rapidly changing that landscape. “By building learning systems, we don’t have to write these rules anymore,” John Giannandrea told a room full of reporters inside Google headquarters this fall. “Increasingly, we’re discovering that if we can learn things rather than writing code, we can scale these things much better.”

Here’s a recap of all the great presentations & discussions from the annual, sell-out event!

MIT engineers recently presented a chip designed to use run sophisticated image-processing neural network software on a smartphone’s power budget.

The great performance of neural networks doesn’t come free. In image processing, for example, neural networks like AlexNet work so well because they put an image through a huge number of filters, first finding image edges, then identifying objects, then figuring out what’s happening in a scene. All that requires moving data around a computer again and again, which takes a lot of energy, says Vivienne Sze, an electrical engineering professor at MIT. Sze collaborated with MIT computer science professor Joel Emer, who is also a senior research scientist at GPU-maker Nvidia.

… Over the last five years, Jennings [Missouri] precinct commander Jeff Fuesting has tried to improve relations between officers — nearly all white — and residents — nearly all black — by going door to door for “Walk and Talks.” Fuesting had expressed interest in predictive policing years before, so when the department heads brought in HunchLab, they asked his precinct to roll it out first. They believed that data could help their officers police better and more objectively. By identifying and aggressively patrolling “hot spots,” as determined by the software, the police wanted to deter crime before it ever happened.

HunchLab, produced by Philadelphia-based startup Azavea, represents the newest iteration of predictive policing, a method of analyzing crime data and identifying patterns that may repeat into the future. HunchLab primarily surveys past crimes, but also digs into dozens of other factors like population density; census data; the locations of bars, churches, schools, and transportation hubs; schedules for home games — even moon phases.

The Lane Fellow Program recognizes and supports scientists of outstanding intellect who are dedicated to a career at the interface of computational and biological sciences so that they can pursue postdoctoral research in the rich computational environment at Carnegie Mellon. Absent special circumstances, applicants must have received their doctoral degree after August 1, 2014 or be expected to receive their degree by August 1, 2016. Candidates who have done prior Ph.D. or postdoctoral work at Carnegie Mellon are not normally considered.

Deadline for applications is Tuesday, March 15.

… [Jesse] Toth is an engineer at GitHub—the company at the heart of the modern software world—and today, she and her fellow GitHub engineers officially released a tool designed to ensure that your new code is ready before you disconnect your old code—in some cases, very old code. The tool is called Scientist, and as open source software, it’s freely available to all. Toth and others believe it could potentially help anyone upgrade even the largest of online services.

“I feel like the scope for this is huge. As soon as you write code, it becomes legacy code. Somebody has to maintain it, and eventually, you will need to change it,” Toth says. “It’s hard for people to make these changes and feel confident in them.”

This repository contains the original models (ResNet-50, ResNet-101, and ResNet-152) described in the paper “Deep Residual Learning for Image Recognition” (http://arxiv.org/abs/1512.03385). These models are those used in ILSVRC and COCO 2015 competitions, which won the 1st places in: ImageNet classification, ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.