Data Science newsletter – June 2, 2021

… Inspired by the way brains work, deep learning models with neural networks methods were overwhelmingly more accurate than signal processing-based approaches when we had massive amounts of data (in this case ~40,000 sets of sensor reads). But they performed similarly to signal processing methods when the data size was an order of magnitude smaller. Of course, data sets of sufficient size are incredibly rare and prohibitive to collect for most groups. Democratizing data access can facilitate better biomarker development.

Access to data is not the only barrier to the development of quality digital biomarkers. In an attempt to publish or perish, researchers’ own self-interests can lead them to use subtle tactics such as selective choice of data and metrics to improve the perception of their model accuracy relative to competing methods. Unfortunately, the peer-review process is often not sufficient to detect these subtle manipulations. This also makes it difficult to truly understand the relative performance of competing methods in a truly unbiased fashion.

/2. Details of the methodology have not been disclosed, but the idea is to develop models describing the interrelationships of all the variables in the ACS, and then construct a simulated population consistent with those models.

/3. Such modeled data captures relationships between variables only if they have been intentionally included in the model. Accordingly, synthetic data are poorly suited to studying unanticipated relationships, which impedes new discovery.

“I have to credit our content strategy team. In early January (2020) they began researching the coronavirus. My senior content person came to me and said ‘I think we should write an article on coronavirus’ and so we had an article up on the site in late January,” Aaron Watkins, senior director of internet strategy at Johns Hopkins Medicine told Fierce Healthcare.

Watkins and his team are responsible for marketing, social media, search engine positioning and broad digital transformation strategies at the organization.

The teams at Johns Hopkins had to shift quickly to create a content model that was more nimble and could provide access to accurate coronavirus information in a rapidly changing environment. For example, the organization leveraged its medical expertise and content to develop an online COVID-19 self-checker.

As Mitsu Ogihara, the director of education at the University of Miami Institute for Data Science and Computing (IDSC) notes, such smart data must be collected, analyzed, and conveyed instantly because “two days—or even two seconds—later is too late.”

Now, with a prestigious grant from the National Science Foundation (NSF), the University has become the newest academic partner of the NSF-sponsored Center for Accelerated Real Time Analytics (CARTA), which helps industry and government members drive innovations by extracting and analyzing real-time information from massive and moving data sets, such as video, voice, social networking, and the “internet of things.”

In a new study, [Naftali] Kaminski’s team and a multinational team of investigators and the Human Cell Atlas project, an international initiative to describe all cells in the human body, examined the diverse characteristics of these relatively understudied cells.

In analysis of healthy lung tissues in 15,000 endothelial cells obtained from 73 individuals, they established a reference map of endothelial cells in the lung which will help researchers identify specific abnormalities that occur in a host of lung diseases, including COVID-19.

If someone handed you an art history book and asked you to show them the famous painting of a woman in a brack dress who is not quite smiling, you could probably identify Da Vinci’s Mona Lisa from the pictures within. Humans learn this ability to match images and objects to a written or oral description over many years. AI has proven capable of this as well but not quite as capable as humans. This is especially true when the AI is asked to identify an object outside of its current vocabulary or when it only has a textual description of the object to go off of.

A team of scientists at NYU and Facebook, including CDS founder Yann LeCun, CDS Deepmind Fellow Aishwarya Kamath and CIMS Post-Doctoral Fellow Nicolas Carion came together to try to create a solution to this problem. What they came up with is MDETR or Modulated Detection for End-to-End Multi-Modal Understanding which they outlined in a paper available on arXiv.

Conservation and agricultural organizations have advanced the business case for in-field conservation practices in recent years, but there remains a need for scalable data-gathering mechanisms that easily integrate into farmers’ and lenders’ most trusted sources of financial benchmarking information.

Leveraging the insights from a recently published guide from Environmental Defense Fund and partners, there are clear opportunities to address this data gap at regionally specific scales, ultimately reducing farmer uncertainty in the profitability of conservation practices and helping farmer advisers support wise conservation investments.

‘As with many hidden criminal syndicates, you don’t always know what’s happening,’ says Retraction Watch’s Ivan Oransky about paper mills. They are the biggest organised fraud perpetrated on scientific journals ever, eroding scientists’ trust in the publishing system – and in each other.

While plagiarism and fraud isn’t new – individual researchers have been caught photoshopping electron microscopy images or inventing elemental analysis data – paper mills serve up professional fakery for their customers on an industrial scale. Buyers can apparently purchase a paper, or authorship of one, on any topic based on phony results to submit to a journal. This makes them not only harder to detect and crack down on, but also exponentially increases the damage they could do.

The extent of their operations became apparent in early 2020. Two independent groups of image detectives came across a number of manuscripts, all from different authors at different institutions working on different biomedical topics, that seemed to share strange inconsistencies – as if they had all used the same stock images. The set now contains almost 600 manuscripts. Another set of 125 was discovered only a few months later. And there could be 10 times as many professionally manipulated papers that have not yet been – and might never be – found, estimates science integrity consultant Elisabeth Bik.

Seatrec, a five-year-old ‘blue tech’ start-up based in Vista, California is one company rising to that challenge, with power generating technology that it hopes will bring a level of sustainability to oceanographic exploration not possible until recently.

Dead batteries on the seabed

‘Profiling floats’ – unmanned, robotic data gathering devices that monitor the sea’s physical, chemical and geological characteristics – is an established approach. The Argo international research project, for example, has more than 3,000 active floats dotted around the world.

Laden with sensors, these battery-fuelled floats dive to predetermined depths to carry out their programmed tasks, re-emerge on completion and transmit their data harvest to the research stations via satellite.

Over and over, the pandemic has reinforced the reality of racial disparities in the U.S. health system. But that story remains difficult to see in the data, which is still inconsistently collected and reported across the country.

On Wednesday, a coalition of researchers and advocates launched a tool they hope will fill some of those gaps: the Health Equity Tracker, a portal that collects, analyzes, and makes visible data on some of the inequities entrenched in U.S. medicine.

“For far too long it’s been ‘no data, no problem,’” said Nelson Dunlap, chief of staff at the Satcher Health Leadership Institute at Morehouse School of Medicine, which developed the tool with funding and resources from Google.org, Gilead Sciences, Annie E. Casey Foundation, and CDC Foundation. By making data that do exist on racial health disparities accessible, the tracker aims to empower local advocates to drive change in their communities — and inspire action to fill in holes in data that are themselves reinforced by structural racism. In the tracker’s display, 38% of federally-collected Covid-19 cases report unknown race and ethnicity.

To help reveal the pandemic’s impacts on economies and people’s daily lives at local, regional, and global scales around the world, scientists, policymakers, and the public have looked to information from many different sources, including various public health and population data sets as well as in situ and satellite observations of environmental conditions. Satellite data typically cover wider areas than in situ observations; in situ observations are better at identifying small-scale phenomena, and they provide ground truth for satellite observations. Thus, integrating satellite environmental data with data from multiple on-the-ground sources can provide a more holistic look at the causes and spread of disease outbreaks, as well as the effects these outbreaks have on the environment and society.

Assembling all this information into forms that allow researchers to uncover links between health and environmental factors and that decisionmakers can use to enact effective policies is far from simple. Nonetheless, this sort of geohealth data integration is crucial in facilitating research and the search for solutions during public health crises like disease outbreaks, which are often influenced by environmental conditions. This point has been demonstrated not only during the ongoing COVID-19 pandemic but also by studies of other disease outbreaks in the past.

Here we discuss the need for effective integration of data from different disciplines (e.g., health, geosciences, economy, population) and sources (satellite versus on the ground), barriers to such integration, and tools and opportunities to overcome these barriers.

What we’re doing at Numenta is starting with deep learning networks, and how far we can go by modifying them. Instead of throwing everything away, how do we start moving from where we are?

We have a whole theory on sparsity, and we’ve now been able to speed up existing deep neural networks by anywhere from five to over a hundred times, depending on the network architecture. They become more robust, and they’re less prone to adversarial attacks.

We’re now adding dendrite theory that I wrote about in the book. We’re showing that we can solve another big problem in deep learning, which is continuous learning. Deep learning networks can’t be incrementally trained; you have to start over and train the whole thing again.

A new center at the UW School of Medicine and Public Health seeks to examine how a person’s environment and social conditions impact their health down to the molecular level.

The new research effort, called the UW Center for Health Disparities Research (CHDR), launched in April. As its name implies, center investigators will focus on how physical environment and social conditions intersect to influence an individual’s health. Researchers will aim to identify new therapies, precision medicine approaches and other interventions, according to Amy Kind, MD, PhD, center director, and professor of medicine at the school.

It all started with a problem many geoscientists faced in 2020.

Alex Steiner, a doctoral student at Michigan State University, had research to do working on thin sections—slivers of geological materials that are usually analyzed under a microscope. But he and the two undergraduate students on the project were not allowed to access the lab or the geological samples they were working on. Because, well, pandemic.

It was out of this necessity that Steiner helped develop a new tool that could automatically take pictures of entire thin sections and stitch them into digital panoramic microscope images that could be analyzed anywhere.

The technical report on the device, named PiAutoStage, was recently published

A new master’s degree in computer vision will launch this fall at the University of Central Florida, the first public university in the country to offer a degree in this rapidly expanding field.

Chicago November 9-11. “Rev 3 is dedicated to providing attendees continuous learning that improves MLOps so they can better accomplish their missions.” [save the date]

Online July 12-18. [$$$]

Online July 13-14. “Technologies and procedures that would have been science fiction are now the reality of medicine. Our 2021 STAT Breakthrough Science Summit will take you inside these innovations.” [$$$]

In the past decade, the incredible progress in word embeddings and deep learning has fueled an interest in neural information retrieval. An increasing number of folks believe that it’s time to retire the traditional inverted indexes (aka posting lists) that search engines use for retrieval and ranking.

In its place, they advocate a model where search engines use neural networks to represent documents and queries as vectors, and then use nearest neighbor search — or more sophisticated ranking models — to retrieve and rank results.

This revolutionary approach is tempting, but — in my view — misdirected. As I argue below, the right place to focus AI efforts is query understanding.

AI models are increasingly applied in high-stakes domains like health and conservation. Data quality carries an elevated significance in high-stakes AI due to its heightened downstream impact, impacting predictions like cancer detection, wildlife poaching, and loan allocations. Paradoxically, data is the most under-valued and de-glamorised aspect of AI. In this paper, we report on data practices in high-stakes AI, from interviews with 53 AI practitioners in India, East and West African countries, and USA. We define, identify, and present empirical evidence on Data Cascades—compounding events causing negative, downstream effects from data issues—triggered by conventional AI/ML practices that undervalue data quality. Data cascades are pervasive (92% prevalence), invisible, delayed, but often avoidable. We discuss HCI opportunities in designing and incentivizing data excellence as a first-class citizen of AI, resulting in safer and more robust systems for all.

This is a very active area with many startups rolling out new offerings

2/ Models interacting with the real world and failing to make sense of it can have serious consequences. Unfortunately, models begin to degrade the moment they get deployed. We begin by describing the many catalysts that can result in model degradation