Aurelia Butler

When the stakes are high, machine-learning models are sometimes used to aid human decision-makers. For instance, a model could predict which law school applicants are most likely to pass the bar exam to help an admissions officer determine which students should be accepted.
These models often have millions of parameters, so how they make predictions is nearly impossible for researchers to fully understand, let alone an admissions officer with no machine-learning experience. Researchers sometimes employ explanation methods that mimic a larger model by creating simple approximations of its predictions. These approximations, which are far easier to understand, help users determine whether to trust the model’s predictions.
But are these explanation methods fair? If an explanation method provides better approximations for men than for women, or for white people than for Black people, it may encourage users to trust the model’s predictions for some people but not for others.
MIT researchers took a hard look at the fairness of some widely used explanation methods. They found that the approximation quality of these explanations can vary dramatically between subgroups and that the quality is often significantly lower for minoritized subgroups.
In practice, this means that if the approximation quality is lower for female applicants, there is a mismatch between the explanations and the model’s predictions that could lead the admissions officer to wrongly reject more women than men.
Once the MIT researchers saw how pervasive these fairness gaps are, they tried several techniques to level the playing field. They were able to shrink some gaps, but couldn’t eradicate them. More

Interactive visualizations have changed the way we understand our lives. For example, they can showcase the number of coronavirus infections in each state.
But these graphics often are not accessible to people who use screen readers, software programs that scan the contents of a computer screen and make the contents available via a synthesized voice or Braille. Millions of Americans use screen readers for a variety of reasons, including complete or partial blindness, learning disabilities or motion sensitivity.
University of Washington researchers worked with screen-reader users to design VoxLens, a JavaScript plugin that — with one additional line of code — allows people to interact with visualizations. VoxLens users can gain a high-level summary of the information described in a graph, listen to a graph translated into sound or use voice-activated commands to ask specific questions about the data, such as the mean or the minimum value.
The team presented this project May 3 at CHI 2022 in New Orleans.
“If I’m looking at a graph, I can pull out whatever information I am interested in, maybe it’s the overall trend or maybe it’s the maximum,” said lead author Ather Sharif, a UW doctoral student in the Paul G. Allen School of Computer Science & Engineering. “Right now, screen-reader users either get very little or no information about online visualizations, which, in light of the COVID-19 pandemic, can sometimes be a matter of life and death. The goal of our project is to give screen-reader users a platform where they can extract as much or as little information as they want.”
Screen readers can inform users about the text on a screen because it’s what researchers call “one-dimensional information.”
“There is a start and an end of a sentence and everything else comes in between,” said co-senior author Jacob O. Wobbrock, UW professor in the Information School. “But as soon as you move things into two dimensional spaces, such as visualizations, there’s no clear start and finish. It’s just not structured in the same way, which means there’s no obvious entry point or sequencing for screen readers.” More

Your virtual visit with your doctor is at 1:00 p.m. It’s now 1:20 p.m. and your physician has not yet logged in. Do you call the clinic? Hang up and log back in? Groan in frustration?
Being stuck in a virtual waiting room and staring at a blank computer or device screen is a huge dissatisfier among telemedicine patients. To respect patients’ time, and provide the optimal experience, UC San Diego Health conducted a 10-week quality improvement study to evaluate how text messaging a link to a patient when their doctor is ready provides a way to connect patients and doctors most efficiently, without relying on the virtual waiting room.
Results of the study published in the May 27 online issue of Quality Management in Health Care.
“Borrowing from the airline and restaurant industries, we tested whether we could contact patients via text to log into their appointment when their doctor is ready. The goal of the feasibility study was to determine if this flexibility lead to improved perception of waiting time and an enhanced experience, while assessing for time saving for both patients and providers,” said Brett C. Meyer, MD, neurologist, co-director of the UC San Diego Heath Stroke Center, and clinical director of telehealth at UC San Diego Health.
“We stepped back and asked, ‘Do we need a virtual waiting room at all? Can we let patients know when their provider is available instead of making them wait online?'” said Emily S. Perrinez, RN, MSN, MPH, study co-author and director of telehealth operations at UC San Diego Health. “The reality is that wait times and lack of timely communication both correlate with patient experience. Real-time text notification that the provider is ready improved patient satisfaction and this experience is the kind of feedback we love to see.”
Twenty-two patients at a stroke clinic participated in the two-and-a-half month study. Patients chose to either receive a text, which included a visit link when their provider was ready for their visit or the standard telehealth routine of logging in at a scheduled time and waiting in front of a camera in a virtual waiting room. More

New research from the Department of Sociology in Trinity College Dublin has found further evidence of a relationship between online engagement and mental wellbeing in teenagers. The study, published recently in the journal ‘Computers in Human Behaviour’, contributes to mounting international evidence on the dangers of high levels of digital media use.
Additionally, the researchers found that in today’s connected world low engagement with digital media is also associated with poor mental health outcomes for adolescents who spend less time online than their peers. This finding supports the ‘goldilocks’ hypothesis — that digital media use at moderate levels is not intrinsically harmful and there is a point between low and high use that is ‘just right’ for young people.
This is the first time the ‘goldilocks’ theory has been examined in Irish teenagers/young adults. It is also the first study to attempt the integration of both time and online behaviours when examining associations between digital media and mental wellbeing.
Professor Richard Layte, Professor of Sociology and co-author on the paper, said:
“Evidence is mounting internationally that online engagement among adolescents may be damaging for mental well-being but the evidence is mixed. Our work provides fresh insights on the impact of digital engagement at the age of 17/18 and the results provide worrying evidence of real harms that require urgent action.”
“There is a simple narrative out there that more is worse. It is important to emphasise that online engagement is now a normal channel of social participation and non-use has consequences. Our findings also raise the possibility that moderate use is important in today’s digital world and that low levels of online engagement carries its own risks. Now the questions for researchers are how much is too much and how little is too little?”
The research, drawing on longitudinal data from the Growing Up in Ireland study, looked at the association between adolescent use of online engagement and mental wellbeing in over 6,000 young people between the age of 13 and again at the age of 17/18. More

A team led by University of Minnesota Twin Cities researchers has discovered how subtle structural changes in strontium titanate, a metal oxide semiconductor, can alter the material’s electrical resistance and affect its superconducting properties.
The research can help guide future experiments and materials design related to superconductivity and the creation of more efficient semiconductors for various electronic device applications.
The study is published in Science Advances, a peer-reviewed, multidisciplinary scientific journal published by the American Association for the Advancement of Science.
Strontium titanate has been on scientists’ radar for the past 60 years because it displays many interesting properties. For one, it becomes a superconductor, i.e. conducts electricity smoothly without resistance, at low temperatures and low concentrations of electrons. It also undergoes a structure change at 110 Kelvin (-262 degrees Fahrenheit), meaning the atoms in its crystalline structure change their arrangement. However, scientists are still debating what exactly causes superconductivity in this material on the microscopic level or what happens when its structure changes.
In this study, the University of Minnesota-led team was able to shine some light on these issues.
Using a combination of materials synthesis, analysis, and theoretical modeling, the researchers found that the structural change within strontium titanate directly affects how electric current flows through the material. They also revealed how small changes in the concentrations of electrons in the material affect its superconductivity. These insights will ultimately inform future research on this material, including research on its unique superconducting properties. More

One of the most tedious, daunting tasks for undergraduate assistants in university research labs involves looking hours on end through a microscope at samples of material, trying to find monolayers.
These two-dimensional materials — less than 1/100,000th the width of a human hair — are highly sought for use in electronics, photonics, and optoelectronic devices because of their unique properties.
“Research labs hire armies of undergraduates to do nothing but look for monolayers,” says Jaime Cardenas, an assistant professor of optics at the University of Rochester. “It’s very tedious, and if you get tired, you might miss some of the monolayers or you might start making misidentifications.”
Even after all that work, the labs then must doublecheck the materials with expensive Raman spectroscopy or atomic force microscopy.
Jesús Sánchez Juárez, a PhD student in the Cardenas Lab, has made life a whole lot easier for those undergraduates, their research labs, and companies that encounter similar difficulties in detecting monolayers.
The breakthrough technology, an automated scanning device described in Optical Materials Express, can detect monolayers with 99.9 percent accuracy — surpassing any other method to date. More

Among several infectious disease terms to enter the public lexicon, superspreading events continue to make headlines years after the first cases of the COVID-19 pandemic. How features of the SARS-CoV2 virus lead to some events becoming superspreading events while leaving others relatively benign remains unresolved.
In Physics of Fluids, by AIP Publishing, researchers in Canada and the United States created a model to connect what biologists have learned about COVID-19 superspreading with how such events have occurred in the real world. They use real-world occupancy data from more than 100,000 places where people gather across 10 U.S. cities to test several features ranging from viral loads to the occupancy and ventilation of social contact settings.
They found that 80% of infections occurring at superspreading events arose from only 4% of those who were carrying the virus into the event, called index cases. The top feature driving the wide variability in superspreading events was the number of viral particles found in index cases, followed by the overall occupancy in social contact settings.
The researchers’ methods take aim at the curious observations that the variability between infection events is higher than one would expect, a situation called overdispersion.
“It is now well known that COVID-19 is airborne, and that is probably the dominant pathway of transmission,” said author Swetaprovo Chaudhuri. “This paper connects indoor airborne transmission to the evolution of the infection distribution on a population scale and shows the physics of airborne transmission is consistent with the mathematics of overdispersion.”
The group’s model draws on numerical simulations and research by others on viral loads and the number of virus-laden aerosols ejected by people, as well as data on the occupancy of a restaurant or area from SafeGraph, a company that generates such data from anonymized cell phone signals.
“While there are uncertainties and unknowns, it appears it is rather hard to prevent a superspreading event if the person carrying high viral load happens to be in a crowded place,” Chaudhuri said.
Chaudhuri said the findings not only underscore the importance of efforts to curb the spread of the virus but also help describe how integral properly planning can be for each situation.
“To mitigate such superspreading events, vaccination, ventilation, filtration, mask wearing, reduced occupancy — all are required,” he said. “However, putting them in place is not enough, knowing what size, type, parameters can mitigate risk to certain acceptable levels is important.”
Story Source:
Materials provided by American Institute of Physics. Note: Content may be edited for style and length. More

A new technique developed by the National Center for Atmospheric Research (NCAR) uses artificial intelligence to efficiently update the vegetation maps that are relied on by wildfire computer models to accurately predict fire behavior and spread.
In a recent study, scientists demonstrated the method using the 2020 East Troublesome Fire in Colorado, which burned through land that was mischaracterized in fuel inventories as being healthy forest. In fact the fire, which grew explosively, scorched a landscape that had recently been ravaged by pine beetles and windstorms, leaving significant swaths of dead and downed timber.
The research team compared simulations of the fire generated by a state-of-the-art wildfire behavior model developed at NCAR using both the standard fuel inventory for the area and one that was updated with artificial intelligence (AI). The simulations that used the AI-updated fuels did a significantly better job of predicting the area burned by the fire, which ultimately grew to more than 190,000 acres of land on both sides of the continental divide.
“One of our main challenges in wildfire modeling has been to get accurate input, including fuel data,” said NCAR scientist and lead author Amy DeCastro. “In this study, we show that the combined use of machine learning and satellite imagery provides a viable solution.”
The research was funded by the U.S. National Science Foundation, which is NCAR’s sponsor. The modeling simulations were run at the NCAR-Wyoming Supercomputing Center on the Cheyenne system.
Using satellites to account for pine beetle damage
For a model to accurately simulate a wildfire, it requires detailed information about the current conditions. This includes the local weather and terrain as well as the characteristics of the plant matter that provides fuel for the flames — what’s actually available to burn and what condition it’s in. Is it dead or alive? Is it moist or dry? What type of vegetation is it? How much is there? How deep is the fuel layered on the ground? More