Data Science newsletter – July 26, 2017

GROUP CURATION: N/A

The number of data breaches in the U.S. jumped 29 percent in the first half of this year, hitting a record high of 791, according to a new report from the Identity Theft Resource Center and CyberScout, the data risk management company.

“Frankly, I was surprised at how significantly the number of breaches has grown,” said Eva Velasquez, ITRC’s president and CEO. “We knew this was a trend, we knew that the thieves would continue to find this lucrative, but the sheer volume of growth has been really surprising.”

The University of British Columbia (UBC) has developed a new program to help graduates tap into this growing market in the new age of data science.

“By tradition, our focus was more narrow; now it is becoming more broad,” said Harish Krishnan, head of the graduate business analytics program at UBC’s Sauder School of Business. “Before, it might be more focused on operational problems like capacity planning, process improvement or patient flow through a hospital.”

“The new program recognizes that while these things are important, there are other areas within various business disciplines that have a strong demand for analytical talent,” Krishnan said.

Six proposals have been selected to participate in a new partnership between two U.S. Department of Energy (DOE) user facilities through the “Facilities Integrating Collaborations for User Science” (FICUS) initiative. The expertise and capabilities available at the DOE Joint Genome Institute (JGI) and the National Energy Research Scientific Computing Center (NERSC) – both at the Lawrence Berkeley National Laboratory (Berkeley Lab) – will help researchers explore the wealth of genomic and metagenomic data generated worldwide through access to supercomputing resources and computational science experts to accelerate discoveries.

“As we bring researchers into the FICUS program, we are introducing a new user community to the power of supercomputers. Scientists will use whatever tools are readily available to investigate a hypothesis and to date, only a small set of biological tools have needed a supercomputer, but this is changing quickly,” says Kjiersten Fagnan, who serves a dual role as the DOE JGI’s Chief Informatics Officer and NERSC’s Data Science Engagement Group Lead.

Over the past three months, lobbying efforts by technology firms to influence US government policy have intensified.

The period from April 1 to June 30 saw record spending by giants like Amazon, Google and Microsoft on a myriad of issues, as well as increased contributions from data center players like Equinix and Iron Mountain, which are interested in federal data center consolidation initiatives and broader regulation of the sector.

In a recent survey of 704 principal investigators for National Science Foundation biology grants, the majority said their most important unmet data needs were not software or infrastructure, but training in data integration and data management.[1]

This lack of data skills is holding back progress toward more reproducible research by making it harder for researchers to share, review, and reanalyze one another’s data. In a recent survey by Nature, the top four solutions scientists identified for improving reproducibility related to better understanding of statistics and research design and improved mentoring, supervision, and teaching of researchers.[2] Data skills need to be an integral part of academic training in order to ensure that research is reliable, transparent, and reproducible.

My organization, Data Carpentry, and our sister organization, Software Carpentry, are among the groups filling this gap by training researchers in the latest technologies and techniques for cleaning, organizing, cataloging, analyzing, and managing their data. We see this training as an important part of a larger project to transform academic culture to make research more reproducible and transparent.

By now our future is clear: We are to be cared for, entertained, and monetized by artificial intelligence. Existing industries like healthcare and manufacturing will become much more efficient; new ones like augmented reality goggles and robot taxis will become possible.

But as the tech industry busies itself with building out this brave new artificially intelligent, and profit boosting, world, it’s hitting a speed bump: Computers aren’t powerful and efficient enough at the specific kind of math needed. While most attention to the AI boom is understandably focused on the latest exploits of algorithms beating humans at poker or piloting juggernauts, there’s a less obvious scramble going on to build a new breed of computer chip needed to power our AI future.

Missing data is a challenge for all studies; however, this is especially true for electronic health record (EHR) based analyses. Failure to appropriately consider missing data can lead to biased results. Here, we provide detailed procedures for when and how to conduct imputation of EHR data. We demonstrate how the mechanism of missingness can be assessed, evaluate the performance of a variety of imputation methods, and describe some of the most frequent problems that can be encountered. We analyzed clinical lab measures from 602,366 patients in the Geisinger Health System EHR. Using these data, we constructed a representative set of complete cases and assessed the performance of 12 different imputation methods for missing data that was simulated based on 4 mechanisms of missingness. Our results show that several methods including variations of Multivariate Imputation by Chained Equations (MICE) and softImpute consistently imputed missing values with low error; however, only a subset of the MICE methods were suitable for multiple imputation. The analyses described provide an outline of considerations for dealing with missing EHR data, steps that researchers can perform to characterize missingness within their own data, and an evaluation of methods that can be applied to impute clinical data. While the performance of methods may vary between datasets, the process we describe can be generalized to the majority of structured data types that exist in EHRs and all of our methods and code are publicly available.

The task of comparing brains and clusters of galaxies is a difficult one. For one thing it requires dealing with data obtained in drastically different ways: telescopes and numerical simulations on the one hand, electron microscopy, immunohistochemistry, and functional magnetic resonance on the other.

,
It also requires us to consider enormously different scales: The entirety of the cosmic web—the large-scale structure traced out by all of the universe’s galaxies—extends over at least a few tens of billions of light-years. This is 27 orders of magnitude larger than the human brain. Plus, one of these galaxies is home to billions of actual brains. If the cosmic web is at least as complex as any of its constituent parts, we might naively conclude that it must be at least as complex as the brain.

Artificial intelligence (AI) enables Siri to recognise your question, Google to correct your spelling, and tools such as Kinect to track you as you move around the room.

Data big and small have come to education, from creating online platforms to increasing standardised assessments. But how can AI help us use and improve it?

Computer science offers a large set of tools for prototyping, writing, running, testing, validating, sharing and reproducing results, however computational science lags behind. In the best case, authors may provide their source code as a compressed archive and they may feel confident their research is reproducible. But this is not exactly true. James Buckheit and David Donoho proposed more than two decades ago that an article about computational results is advertising, not scholarship. The actual scholarship is the full software environment, code, and data that produced the result. This implies new workflows, in particular in peer-reviews. Existing journals have been slow to adapt: source codes are rarely requested, hardly ever actually executed to check that they produce the results advertised in the article. ReScience is a peer-reviewed journal that targets computational research and encourages the explicit replication of already published research, promoting new and open-source implementations in order to ensure that the original research can be replicated from its description. To achieve this goal, the whole publishing chain is radically different from other traditional scientific journals. ReScience resides on GitHub where each new implementation of a computational study is made available together with comments, explanations, and software tests.

To connect companies with data scientists, local startup Experfy is working on both creating a marketplace of available experts as well as training current employees or people who want to start a career in data science.

First, the company created an online network of over 30,000 data experts, but also machine learning, Artificial Intelligence and Internet of Things experts, including academics and full-time consultants. By using the marketplace, companies can post the projects they need help with, connect with the right people and hire on-demand data scientists.

“On an average, I would say that every project gets about 10 to 15 proposals,” Jyotsna Khan, a senior manager at Experfy, during an interview. “The client would select like, say, three people and … pick times to interview them. Everything happens through the platform.”

Copenhagen, Denmark Empirical Methods in Natural Language Processing, September 7-11. [$$$]

Stanford, CA Stanford Medicine X is a multifaceted program that represents a new way of solving health care’s most pressing problems, September 15-17. [$$$$]

“Combining datasets and performing large aggregate analyses are a powerful new way to improve service across large populations. Critically important in this task is the deduplication of identities across multiple data sets that were rarely designed to work together. Inconsistent data entry, typographical errors, and real world identity changes pose significant challenges to this process. To help, we have built a tool called pgdedupe.”

Australia released today a vast trove of underwater data gathered during the search for the missing Boeing 777 airliner, which disappeared in March 2014 while en route from Kuala Lumpur to Beijing with 239 people on board. To find the plane, search teams armed with high-tech equipment and relatively large budgets painstakingly surveyed about 120,000 sq km (46,000 sq miles) of seafloor, turning an otherwise remote part of the Indian Ocean into one of the world’s most thoroughly mapped regions of deep ocean.

Where do you start checking if your model is outputting garbage (for example predicting the mean of all outputs, or it has really poor accuracy)?

A network might not be training for a number of reasons. Over the course of many debugging sessions, I would often find myself doing the same checks. I’ve compiled my experience along with the best ideas around in this handy list. I hope they would be of use to you, too.

Boosted decision trees are responsible for more than half of the winning solutions in machine learning challenges hosted at Kaggle, according to KDNuggets. In addition to superior performance, these algorithms have practical appeal as they require minimal tuning. In this post, we evaluate two popular tree boosting software packages: XGBoost and LightGBM, including their GPU implementations.