NYU Data Science newsletter – July 9, 2015

GROUP CURATION: N/A

These images are synthetically generated to maximally activate individual neurons in a Deep Neural Network (DNN). They show what each neuron “wants to see”, and thus what each neuron has learned to look for. The neurons selected for these images are the output neurons that a DNN uses to classify images as flamingos or school buses. Below we show that similar images can be made for all of the hidden neurons in a DNN. Our paper describes that the key to producing these images with optimization is a good natural image prior.

Metrics evoke a mixed reaction from the research community. A commitment to using data and evidence to inform decisions makes many of us sympathetic to, even enthusiastic about, the prospect of granular, real-time analysis of our own activities. If scientists cannot take full advantage of the possibilities of big data, then who can?

Yet we only have to look at the blunt use of metrics such as journal impact factors, h-indices and grant-income targets to be reminded of the pitfalls. Some of the most precious qualities of academic culture resist simple quantification, and individual indicators can struggle to do justice to the richness and plurality of our research. Too often, poorly designed evaluation criteria are distorting behaviour and determining careers. At their worst, metrics can contribute to what Rowan Williams, the former Archbishop of Canterbury, UK, calls a “new barbarity” in our universities. Metrics hold real power: they are constitutive of values, identities and livelihoods.

Genomics is a Big Data science and is going to get much bigger, very soon, but it is not known whether the needs of genomics will exceed other Big Data domains. Projecting to the year 2025, we compared genomics with three other major generators of Big Data: astronomy, YouTube, and Twitter. Our estimates show that genomics is a “four-headed beast”—it is either on par with or the most demanding of the domains analyzed here in terms of data acquisition, storage, distribution, and analysis. We discuss aspects of new technologies that will need to be developed to rise up and meet the computational challenges that genomics poses for the near future. Now is the time for concerted, community-wide planning for the “genomical” challenges of the next decade.

In recent months, government officials in the United States, the United Kingdom, and other countries have made repeated calls for law-enforcement agencies to be able to access, upon due authorization, encrypted data to help them solve crimes.

Beyond the ethical and political implications of such an approach, though, is a more practical question: If we want to maintain the security of user information, is this sort of access even technically possible?

That was the impetus for a report — titled Keys under doormats: Mandating insecurity by requiring government access to all data and communications — published today by security experts from MIT’s Computer Science and Artificial Intelligence Lab (CSAIL), alongside other leading researchers from the U.S. and the U.K.

This week, Lille, France hosts the 2015 International Conference on Machine Learning (ICML 2015), a premier annual Machine Learning event supported by the International Machine Learning Society (IMLS). As a leader in Machine Learning research, Google will have a strong presence at ICML 2015, with many Googlers publishing work and hosting workshops. If you’re attending, we hope you’ll visit the Google booth and talk with the Googlers to learn more about the hard work, creativity and fun that goes into solving interesting ML problems that impacts millions of people.

Lleyton Hewitt bowed out of Wimbledon on the first day of The Championships last Monday. The 2002 winner of The Championships took the match to five sets with a typically gutsy performance. In doing so, he achieved a milestone, hitting the 1,500th winner of his Wimbledon career. And to top it all off, his match featured Wimbledon’s first use of its new real-time notifications system.

While human brains are specialized for complex and variable real world tasks, most neuroscience studies reduce environmental complexity, which limits the range of behaviours that can be explored. Motivated to overcome this limitation, we conducted a large-scale experiment with electroencephalography (EEG) based brain-computer interface (BCI) technology as part of an immersive multi-media science-art installation. Data from 523 participants were collected in a single night. The exploratory experiment was designed as a collective computer game where players manipulated mental states of relaxation and concentration with neurofeedback targeting modulation of relative spectral power in alpha and beta frequency ranges. Besides validating robust time-of-night effects, gender differences and distinct spectral power patterns for the two mental states, our results also show differences in neurofeedback learning outcome. The unusually large sample size allowed us to detect unprecedented speed of learning changes in the power spectrum (~ 1 min). Moreover, we found that participants’ baseline brain activity predicted subsequent neurofeedback beta training, indicating state-dependent learning. Besides revealing these training effects, which are relevant for BCI applications, our results validate a novel platform engaging art and science and fostering the understanding of brains under natural conditions.

Major funding agencies should ensure that large biological data sets are stored in cloud services to enable easy access and fast analysis, say Lincoln D. Stein and colleagues.

It’s the height of summer in the Northern Hemisphere, and the oceans have never been hotter. What’s more, there’s increasing evidence that a warming feedback loop has been kicked off in recent days—which could quickly ramp up El Niño.

If you’re not a seasoned weather nerd, the feedback loop—warm water begets typhoons beget weird trade winds beget more warm water—is a bit tricky to follow in charts and maps. But its effects on El Niño could produce global ramifications.

Trees Atlanta has a new tool to convince property owners to plant more trees.

“We would drive up and down the street and say, ‘They don’t have any right of way trees, they don’t have any front-yard trees.’ And then we look for a link – someone to establish a project there,” said Alex Beasley of Trees Atlanta.

Beasley said now, convincing homeowners will be a lot easier thanks to a program at Georgia Tech. [audio, 1:12]

Advanced information technologies have revolutionized the way the world works and how people conceptualize it. Massive troves of information, known as “big data,” are aggregated and shared on a daily basis, recording an array of human behaviors and interactions at an unprecedented level of granularity. One form of such data is call data records, which, while preserving the anonymity of subscribers and the privacy of content, allow researchers to track the volume of traffic, timing, and location of calls. In combination with increasingly powerful computers, such data have shed light on important questions in the developed world on topics including marketing, health care, urban planning, and environmental policy.

Call data can also help us understand violent places in the developing world that are largely inaccessible. At a time when Yemen remains highly volatile, for example, anonymous Yemeni cell phone metadata from 2010 to 2013 that include over ten million users and several hundred million calls vividly capture patterns of Yemeni daily life, as well as celebrations, religious practices, involvement in politics, and reactions to violence.

Numenta today released Grok for Stocks, a new mobile application that tracks the trading patterns of companies in the stock market.

The new Google Play app can monitor stock price, stock volume, and Twitter activity for hundreds of publicly traded companies, and it uses artificial intelligence software based on Numenta’s research on how the brain works. The app is the latest example of Hierarchical Temporal Memory (HTM) technology, a machine-learning technology that is based on years of brain research.