Data Science newsletter – November 23, 2016

GROUP CURATION: N/A

… It’s a world that flows in one direction: data comes from us, but it rarely returns to us. The systems that we’ve created are designed to be unidirectional: data is gathered from people, it’s processed by an assembly line of algorithmic machinery, and spit out to an audience of different people — surveillors and investors and academics and data scientists. Data is not collected for high school students, but for people who want to know how high school students feel. This new data reality is from us, but it isn’t for us.

So how can we turn data around? How can we build new data systems that start as two-way streets, and consider the individuals from whom the data comes as first-class citizens?

A new study suggests that despite 50 years of research, science is still not very good at predicting who will kill themselves.

Dr. Joseph Franklin, a Florida State University researcher, made the assertion after an exhaustive examination of hundreds of suicide prediction studies. Franklin is now testing a “machine-learning” method employing algorithms to identify risk factors for suicidal behavior.

In the study, Franklin and his colleagues found traditional risk factors — such as depression, substance abuse, stress, or previous suicide attempts — were not good predictors of suicide.

As artificial intelligence advances, the goal for modern tech companies is to build AI software that thinks for itself without human intervention.

Towards that end, Amazon Web Services just picked MXNet, as its favored deep-learning framework to facilitate that work, according to a blog post Tuesday by Amazon chief technology officer Werner Vogels.

Back in August, Intel (NASDAQ:INTC) announced that it would purchase deep-learning chip startup Nervana systems. Re/code, citing a “source with knowledge of the deal,” says that Intel paid around $408 million for the company.

Although this acquisition is quite small for Intel, which raked in $3.4 billion in net income last quarter, the company seems to be quite bullish on Nervana’s technology for its deep-learning and artificial-intelligence efforts.

Let’s take a closer look at what Intel’s near-term plans are for this technology, and then I’ll finish off with what I expect to see Intel do with the technology out in time.

Serkan Piantino, a longtime Facebook engineer who helped to create its artificial intelligence research lab, has left the social networking firm to start a company focused on making it easier for developers to access the best AI processing hardware.

His new startup, called Top 1 Networks, will offer customers access to the latest Nvidia graphics processing units (or GPUs) as a cloud service, much like Amazon and others offer cloud computing services. Nvidia’s GPUs have taken off in deep learning, a flavor of AI where the computer teaches itself how to do specific tasks.

Before a medical product can be widely used in the United States, it generally must first be approved or cleared for marketing by the US Food and Drug Administration (FDA). Then, payers such as the Centers for Medicare & Medicaid Services (CMS) must decide whether the product merits coverage and payment. Because the statutes governing these agencies evolved to meet the exigencies of particular moments in the history of medical product development, the degree of convergence in standards and in the underlying evidence needed to support regulatory and payment decisions is not always immediately obvious. The resulting fragmentation—perceived or real—has led to questions about whether FDA approval or clearance for marketing will necessarily result in approval for coverage and payment.

Despite these challenges, changes in the organization of health care and in the larger information ecosystem should allow the FDA and CMS to move increasingly toward use of shared sources of evidence while still applying the most appropriate criteria to their decision making. Such a move should help reduce current gaps in evidence that create uncertainty surrounding the approval or clearance of new therapies and their subsequent use in practice. It should also enable greater efficiency in medical product development and provide the higher-quality evidence needed in the emerging era of precision medicine.

Artificial intelligence, once relegated to the realm of science fiction, is now found in everything from translation services to virtual assistants to video games.

And as companies race to develop self-driving cars and offer increasingly personalized online experiences, they’re building on research that was largely pioneered by a group of Canadian researchers who are still attracting plenty of attention and investment dollars.

Montreal, in particular, has developed a concentration of expertise in the area of AI, largely thanks to the efforts of Université de Montréal professor Yoshua Bengio, head of the Montreal Institute for Learning Algorithms (MILA).

The integration of the Merritt repository with Amazon’s S3 and Glacier cloud storage services, previously described in an August 16 post on the Data Pub blog, is now mostly complete. The new Amazon storage supplements Merritt’s longstanding reliance on UC private cloud offerings at UCLA and UCSD. Content tagged for public access is now routed to S3 for primary storage, with automatic replication to UCSD and UCLA. Private content is routed first to UCSD, and then replicated to UCLA and Glacier. Content is served for retrieval from the primary storage location; in the unlikely event of a failure, Merritt automatically retries from secondary UCSD or UCLA storage. Glacier, which provides near-line storage with four hour retrieval latency, is not used to respond to user-initiated retrieval requests.

The premise of Princeton University sociologist Matthew Salganik’s forthcoming book, Bit by Bit: Social Science in the Digital Age, is that technological innovation creates new opportunities for social science researchers.

For example, the proliferation of the internet and other technological advances has opened the door for researchers to use huge caches of data on user behavior collected by companies such as Facebook and Google. While such “found data” has drawn the interest of many researchers, it is far different from the “designed data” social scientists have generally collected under controlled conditions, said Salganik, a professor of sociology.

Microsoft is putting its considerable financial and engineering muscle into the experimental field of quantum computing as it works to build a machine that could tackle problems beyond the reach of today’s digital computers.

There is a growing optimism in the tech world that quantum computers, superpowerful devices that were once the stuff of science fiction, are possible — and may even be practical. If these machines work, they will have an impact on work in areas such as drug design and artificial intelligence, as well as offer a better understanding of the foundations of modern physics.

Information doesn’t flow through the internet as freely as it seems. Depending on which country you live in, you may see different content on a web page—or no content at all. As the internet has become the most important public space for everything from protest movements to pornography, governments around the world have started locking it down. And that has given rise to a new field of research: the science of censorship. ScienceInsider caught up with Phillipa Gill, a computer scientist at the University of Massachusetts in Amherst, to find out what’s cooking in this online cat and mouse game. She spoke to us yesterday from the Internet Measurement Conference in Santa Monica, California, which she co-chaired.

During a hackathon at Princeton University, four college students created one in the form of a Chrome browser extension in just 36 hours. They named their project “FiB: Stop living a lie.”

The students are Nabanita De, a second-year master’s student in computer science student at the University of Massachusetts at Amherst; Anant Goel, a freshman at Purdue University; Mark Craft, a sophomore at the University of Illinois at Urbana-Champaign; and Qinglin Chen, a sophomore also at the University of Illinois at Urbana-Champaign.

Boston-based Cogito, a behavioral analytics company that spun out of the Massachusetts Institute of Technology, has raised $15 million in Series B funding in a round led by OpenView. Existing investors Romulus Capital and Salesforce Ventures also contributed to the round, and OpenView Founder and Managing Partner Scott Maxwell joined Cogito’s board of directors.

The funding will be used to expand Cogito’s customer base and further develop the company’s deep learning and other technological capabilities, which analyze voice and speech to detect emotions. I

DeepMind Health’s inaugural collaboration with the U.K.’s National Health Service (NHS), initially focused on building an app for helping early detection of Acute Kidney Injury (AKI), was relaunched earlier today — under a new information-sharing agreement with the Royal Free NHS Trust, and a broader scope for the collaboration.

You may remember that back in September, Google announced that its Neural Machine Translation system had gone live. It uses deep learning to produce better, more natural translations between languages. Cool!

Following on this success, GNMT’s creators were curious about something. If you teach the translation system to translate English to Korean and vice versa, and also English to Japanese and vice versa… could it translate Korean to Japanese, without resorting to English as a bridge between them? They made this helpful gif to illustrate the idea of what they call “zero-shot translation.”

It’s an intricate process through which massive stars lose their gas as they evolve. And a more complete understanding could be just calculations away, if only those calculations didn’t take several millennia to run on normal computers.

But astrophysicists Matteo Cantiello and Yan-Fei Jiang of UC Santa Barbara’s Kavli Institute for Theoretical Physics (KITP) may find a way around that problem.

The pair have been awarded 120 million CPU hours over two years on the supercomputer Mira — the sixth-fastest computer in the world — through the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program, an initiative of the U.S. Department of Energy Office of Science. INCITE aims to accelerate scientific discoveries and technological innovations by awarding, on a competitive basis, time on supercomputers to researchers with large-scale, computationally intensive projects that address “grand challenges” in science and engineering.

Over the last decade physicists and mathematicians have been exploring a surprising correspondence that has the potential to breathe new life into the venerable Feynman diagram and generate far-reaching insights in both fields. It has to do with the strange fact that the values calculated from Feynman diagrams seem to exactly match some of the most important numbers that crop up in a branch of mathematics known as algebraic geometry. These values are called “periods of motives,” and there’s no obvious reason why the same numbers should appear in both settings. Indeed, it’s as strange as it would be if every time you measured a cup of rice, you observed that the number of grains was prime.

“There is a connection from nature to algebraic geometry and periods, and with hindsight, it’s not a coincidence,” said Dirk Kreimer, a physicist at Humboldt University in Berlin.

In most circles, email marketing is highly underestimated as a revenue driver for businesses and even when the potentials are understood, it’s frequently done wrong. So I put together a simple list of tricks you can act on now without spending anything to improve your email marketing strategy.

#TIP 1: The gist about sender names and email address

Today at BlizzCon 2016 in Anaheim, California, we announced our collaboration with Blizzard Entertainment to open up StarCraft II to AI and Machine Learning researchers around the world.

Two Sigma research scientist Vinod Valsalam provides an overview of some of the most interesting research presented at ICML 2016, covering recurrent neural networks, unsupervised learning, supervised training methods, and deep reinforcement methods.

For financial services companies, the arrival and rapid development of Hadoop and Big Data analytics over this same past decade couldn’t be more timely. The ability of Hadoop platforms to store gigantic volumes of disparate data matches perfectly with new incumbent data streams. Meanwhile Big Data analytics solutions offer unprecedented opportunities to actually profit from compliance while keeping fraud at bay and enabling new revenue streams. Below are several use cases for Hadoop and Big Data analytics already in full swing.