Data Science newsletter – December 26, 2018

GROUP CURATION: N/A

Facebook Inc. tumbled on Wednesday, with shares extending their decline throughout the session after the social-media company was sued by the District of Columbia over a privacy breach.

The news followed a report from the New York Times that Facebook had allowed more than 150 companies to access more personal data from users than it had disclosed, the latest in a series of controversies that have weighed on shares in 2018.

The stock fell as much as 7.3 percent, putting it on track for its biggest one-day percentage drop since its historic collapse in late July. Wednesday’s decline extends a sell-off that has erased nearly 40 percent in value.

First-timers to NeurIPS — the world’s most prestigious machine learning conference — can explore and enjoy a tech environment full of energy, innovations, and mind-blowing discussions among people studying and working in AI. However, with almost 9,000 attendees packing a single venue for a week, NeurIPS can also be a bit overwhelming. Synced asked NeurIPS 2018 attendees to share their first-timer experiences.

A machine learning researcher at Virginia Tech proposes a way to vet papers in AI by the look of the text and graphics in the document. Is its ability to gauge the “gestalt” of winning research enough to speed up the peer-review process?

Diversifying into new industries is vital to an economy’s ability to grow and generate wealth. But to branch out into new industrial activities, a city, region or country must first have a pool of people with the right mix of knowledge and experience to make those pioneering firms a success.

So how do local economies ensure they have the right mix of experience to allow new ventures to thrive?

In a paper published this week in the Proceedings of the National Academy of Sciences, a team led by César A. Hidalgo, director of the Collective Learning Group in the MIT Media Lab, studied the effects of occupation-specific, industry-specific, and location-specific knowledge on the success of pioneer firms. These are firms operating in an industry that has not previously been present in a region.

On December 22, 2018, the OPEN Government Data Act passed Congress. It now heads to the President’s desk. The OPEN Government Data Act was included in the Foundations for Evidence-Based Policymaking Act (H.R.4174) as Title II. The open data proposal will require federal agencies to publish their information online, using machine-readable data formats.

Lu Zhang, a finance professor at Ohio State University, has something to say about your hot new index funds, and it may not be flattering.

Not long ago, the typical investment portfolio was a grab bag of stocks, bonds and actively managed mutual funds. Today, it is more likely an assortment of index funds. And not just any index funds. Indexes are no longer content to simply track the market. A growing number of them are attempting to replicate traditional styles of active management, also known as “factors.” I counted roughly 900 mutual funds and exchange-traded funds in the U.S. that track factor indexes, and that number is likely to grow.

The pivot to indexing may be new, but it was cultivated by decades of research in economics and finance, which gives it the imprimatur of science, or at least robust inquiry. But a new generation of academics, Zhang prominently among them, are re-examining the research and finding much of it questionable. Their work could derail the indexing revolution and, as might be expected, index providers and fund companies aren’t likely to be happy about it.

A recent survey we did of commercial AI startups provides some intriguing insights into how they work, and how and where they compete with larger and more established firms. The picture that emerges is one of a robust, competitive market for startups, but also a market that is restricted in some important ways.

Large incumbent tech firms have several advantages over startups: they can invest huge sums into R&D, they have access to large amounts of data, and they have complementary assets and established markets. Some commentators point to the vast amounts of data available to big technology companies such as Google and Amazon and argue that startups may not be able to compete.

But when it comes to datasets, bigger is not always better. Google constantly experiments with its search engine to improve its results rather than simply relying on its volume of search data, and Amazon has found that the accuracy of its predictions in e-commerce does not continue to improve by simply adding more products. In other words, it is not the quantity of the data, but how the company uses it, that provides the competitive advantage. And thus, AI startups can still compete with larger companies even without enormous datasets to work with.

Let’s take a moment to pity the Internal Revenue Service. Yes, to many Americans, it’s a money-grabbing ogre siphoning hard-earned cash to the faceless federal bureaucracy.

But the nation’s tax collector today is an enfeebled enforcer. Its budget has been bled dry by a Republican Congress in service to wealthy donors and businesses aggressively pursuing tax avoidance, leaving uncollected 18 percent to 20 percent of potential tax revenues annually. That’s the conclusion in articles by the journalism site ProPublica, co-published by The Atlantic and The Times.

Artificial intelligence is shaking up the art world; earlier this month Miami was at the forefront of the AI revolution when a presentation of 12 works created by an AI artist called Aican went on show at the Scope Miami Beach fair. This arty technological innovation comes courtesy of Dr Ahmed Elgammal from Rutgers University who devised the algorithm that underpins Aican. His intriguing AI creation draws from works made in the last 500 years. “To Aican, all the works are equally important and represent the canon of western art. It learns the aesthetics from the whole collection,” Elgammal says. And the $100m question: has Aican really learnt how to paint? “We can see good choice of complementary colours for the foreground and background, warm colours for foreground and cool colours for background. We can see a variety of textures, painterly contours and linear contours,” Elgammal adds. The artists Tim Bengel and Devin Gharakhanian also showed works developed in collaboration with Aican.

Do you believe that people’s eyes emit an invisible beam of force?

According to a rather fun paper in PNAS, you probably do, on some level, believe that. The paper is called Implicit model of other people’s visual attention as an invisible, force-carrying beam projecting from the eyes.

Mountain View, CA March 27-28, 2019. “Two Turing Award Winners, the creators of TensorFlow, PyTorch, Spark, Caffe, TensorRT, OpenAI, Matroid, and others will lead discussions about running and scaling machine learning algorithms on a variety of computing platforms, such as GPUs, CPUs, TPUs, & the nascent AI chip industry.” [$$$]

Social media data is widely analyzed in computational social science. Twitter, one of the largest social media platforms, is used for research, journalism, business, and government to analyze human behavior at scale. Twitter offers data via three different Application Programming Interfaces (APIs). One of which, Twitter’s Sample API, provides a freely available 1% and a costly 10% sample of all Tweets. These data are supposedly random samples of all platform activity. However, we demonstrate that, due to the nature of Twitter’s sampling mechanism, it is possible to deliberately influence these samples, the extent and content of any topic, and consequently to manipulate the analyses of researchers, journalists, as well as market and political analysts trusting these data sources. Our analysis also reveals that technical artifacts can accidentally skew Twitter’s samples. Samples should therefore not be regarded as random. Our findings illustrate the critical limitations and general issues of big data sampling, especially in the context of proprietary data and undisclosed details about data handling. [full text]