Data Science newsletter – May 4, 2017

GROUP CURATION: N/A

The now annual event was launched in 2010 by the Obama administration as a hackathon-style program where attendees were challenged to develop prototype applications in 30 days from 30 data sets. Today, Health Datapalooza includes presentations from government and private sector healthcare experts, breakout panel sessions that dive into specific areas of interest, and an exhibit hall. The common thread of discussion throughout these activities is how health data is, can, or will eventually be used to improve health outcomes. Over the next two days, we’ll be giving you a snapshot of the presentations and topics being covered at this year’s event with a few specific highlights as we find out what’s on the horizon for health data technology.

We are pleased to present the 4th edition of the Global Open Data Index (GODI), a global assessment of open government data publication. GODI compares national government in 94 places across the 15 key datasets that have been assessed by our community as the most useful for solving social challenges.

For this edition, we received 1410 submitted of datasets, but only 10% of these are open according to the Open Definition. At Open Knowledge International (OKI), we believe it’s important to look further than just the numbers. GODI is not just a benchmark, it can be and should be used as a tool to improve open data publication and make data findable, useful and impactful. This is why we include a new phase, the dialogue phase for this edition.

By embedding the brain into a gel that swells up when pumped with water, the team blew up mouse brain tissue to roughly 20 times its original size, while preserving the normal structure and connections of neurons and their dendrites.

Using this method, aptly dubbed expansion microscopy (ExM), the team reconstructed a tiny piece of the mouse brain in 3D. Normally, dendrites entangle into a jumbled mess, making it hard to tease apart individual synaptic connections with a conventional light microscope.

A consortium of eight UK universities, led by the University of Oxford, has been awarded £3 million by the Engineering and Physical Sciences Research Council (EPSRC) to establish a national high-performance computing facility to support machine learning.

Across many scientific domains, there is a common need to automatically extract a simplified view or coarse-graining of how a complex system’s components interact. This general task is called community detection in networks and is analogous to searching for clusters in independent vector data. It is common to evaluate the performance of community detection algorithms by their ability to find so-called ground truth communities. This works well in synthetic networks with planted communities because these networks’ links are formed explicitly based on those known communities. However, there are no planted communities in real-world networks. Instead, it is standard practice to treat some observed discrete-valued node attributes, or metadata, as ground truth. We show that metadata are not the same as ground truth and that treating them as such induces severe theoretical and practical problems. We prove that no algorithm can uniquely solve community detection, and we prove a general No Free Lunch theorem for community detection, which implies that there can be no algorithm that is optimal for all possible community detection tasks. However, community detection remains a powerful tool and node metadata still have value, so a careful exploration of their relationship with network structure can yield insights of genuine worth. We illustrate this point by introducing two statistical techniques that can quantify the relationship between metadata and community structure for a broad class of models. We demonstrate these techniques using both synthetic and real-world networks, and for multiple types of metadata and community structures.

Tropical forests dominate global terrestrial carbon exchange, but long-term climate variability might affect their ability to uptake and store carbon dioxide. To better understand tropical forest carbon dynamics, the Tropical Nutrient Limitation Working Group assembled published datasets to determine how temperature and rainfall interact to control carbon cycling in tropical forests. Their results are featured in a recent Ecology Letters publication.

New computer simulations, described this week in the journal Physics of Fluids can better probe the underlying physics, potentially leading to more efficient extraction of oil and gas.

The general theme of the presentation is the future of programming and computer science and Peter touches on numerous aspects of these topics.


Peter envisions how our interaction with Technology will evolve as intelligent agents become increasingly prevailing in our lives and how Computer Science is gradually becoming an empirical science. [video, 1:37:13]

There is currently increased attention to the statistical (and replication) crisis in science. Biomedicine and social psychology have been at the heart of this crisis, but similar problems are evident in a wide range of fields. We discuss three examples of replication challenges from the field of social psychology and some proposed solutions, and then consider the applicability of these ideas to clinical neuropsychology. In addition to procedural developments such as preregistration and open data and criticism, we recommend that data be collected and analyzed with more recognition that each new study is a part of a learning process. The goal of improving neuropsychological assessment, care, and cure is too important to not take good scientific practice seriously.

NuTonomy will integrate its software and sensors into the Peugeot 3008, a gasoline-powered crossover SUV. Converted vehicles will undergo real-world testing on the streets of Singapore, where nuTonomy last year launched the world’s first taxi service built around self-driving vehicles.

[Karl] Iagnemma said Peugeot and nuTonomy plan to “get a couple of cars on the road quickly in Singapore this summer, and then scale up rapidly from there.”

Last week, Satya Nadella, the chief executive of Microsoft, slipped on a glove made of cardboard and clenched his hand into a fist, causing a robotic hand with fingers made of drinking straws to mimic his movements.

The glove was one of several engineering projects built in a makeshift laboratory on Microsoft’s campus. The company spent the last year talking to thousands of teachers and designing high-tech experiments that require mostly low-cost parts. It will give the designs to schools for free so teachers can use them in their lesson plans.

The projects are part of a major push the company announced Tuesday at an event in New York to make its products more attractive to school administrators, students and teachers.

We don’t know how quickly machines will displace people’s jobs, or how many they’ll take, but we know it’s happening — not just to factory workers but also to money managers, dermatologists and retail workers.

The logical response seems to be to educate people differently, so they’re prepared to work alongside the robots or do the jobs that machines can’t. But how to do that, and whether training can outpace automation, are open questions.

In the not-too-distant future, a “planetary” computer will be able to create a computational model of your body, with the ability to run simulations of your health and to anticipate chronic disease before you show any symptoms.

This is the direction we’re headed, according to Larry Smarr, founding director of the California Institute of Telecommunications & Information Technology at UC San Diego. While his expertise lies in computer networks and infrastructure, Smarr has emerged as a de facto leader in quantified health—largely due to his relentless curiosity about his own health. In 2011, Smarr diagnosed his own Crohn’s disease long before he showed any symptoms.

Places that used to be ad-free—from the living room to our friendships—are now becoming sites for ads or surveillance designed to make them more effective.

Urbana, IL May 16 at NCSA. [free, registration required]

New York, NY June 2-4 at Galvanize NYC. [free, registration required]

New York, NY A night of PyCon warmup talks! Thursday, May 11 at 7 p.m., 2U Office (601 W 26th St, Suite 1255) [free, rsvp required]

San Francisco, CA June 2. The Artificial Intelligence Conference is an annual event where leading AI researchers and top industry practitioners meet and collaborate. [$$$]

“We’re continuously adding new features to our API to make it easier for developers to integrate with our data and share great local businesses through their apps. Today, we’re releasing access to query our data via GraphQL, a graph query language. This is available immediately through our developer beta program.”

A metric-based evaluation give an NLG system a score by computing how similar its output text is to “gold-standard” reference texts. There are a number of different metrics (including BLEU, METEOR, and ROUGE), which are based on different scoring functions.

I am not a great fan of metric-based evaluation, for reasons I explain below, and would be very dubious if, for example, I was asked to review a paper on NLG which only presented a metric-based evaluation. Nevertheless, I will also below give some advice on best practice for such evaluations.

Over the course of this blog post, I will first contrast transfer learning with machine learning’s most pervasive and successful paradigm, supervised learning. I will then outline reasons why transfer learning warrants our attention. Subsequently, I will give a more technical definition and detail different transfer learning scenarios. I will then provide examples of applications of transfer learning before delving into practical methods that can be used to transfer knowledge. Finally, I will give an overview of related directions and provide an outlook into the future.

A resilient Data Science Platform is a necessity to every centralized data science team within a large corporation. It helps them centralize, reuse, and productionize their models at peta scale. We’ve built Algorithmia Enterprise for that purpose.