I recently spoke by phone with Romer, who is a professor at N.Y.U. During our conversation, which has been edited for length and clarity, we also discussed why Americans may be resistant to digital contact tracing, the need for states to administer tests, and what lessons economists have learned from the crisis.
I am especially concerned about three trends among social scientists during the Covid-19 pandemic, the first of which is that many of them appear to be rushing their work. Good science takes time. Researchers often spend months collecting, organizing and double-checking their data. They spend more months presenting their findings and gathering feedback from colleagues before they publicly release their results. But many social scientists are already releasing and publicizing studies using Covid-19 data that was collected just days ago, and they are often failing to apply the same level of rigor that they normally would.
One of their more sober tasks is predicting the number of Americans who will die due to COVID-19. FiveThirtyEight — with the help of the Reich Lab at the University of Massachusetts Amherst — has assembled six models published by infectious disease researchers to illustrate possible trajectories of the pandemic’s death toll. In doing so, we hope to make them more accessible, as well as highlight how the assumptions underlying the models can lead to vastly different estimates. Here are the models’ U.S. fatality projections for the coming weeks.
The number of new papers on the COVID-19 pandemic is doubling every two weeks, and shows no sign of slowing. Many of these papers are published first on preprint servers, which means they are made public before having undergone peer review. This makes it all the harder to judge their merit. Now, one start-up company says that its platform — called Scite.ai — can automatically tell readers whether papers have been supported or contradicted by later academic work.
Unlike conventional citation-metrics tools, Scite.ai tells users how often a paper has been supported or contradicted by the studies that cite it, as well as how many times it has simply been mentioned. The resulting reports display citations in the context in which they are mentioned, allowing users to assess for themselves how the paper is being cited.
So far, Scite.ai has analysed more than 16 million full-text scientific articles from publishers such as BMJ Publishing Group in London and Karger in Basle, Switzerland. But that is just a fraction of the scientific literature. “They’re limited by the literature they can get hold of and the machine-learning algorithms,” notes Jodi Schneider, an information scientist at the University of Illinois at Urbana–Champaign.
One analysis of the IHME model found that its next-day death predictions for each state were outside its 95 percent confidence interval 70 percent of the time — meaning the actual death numbers fell outside the range it projected 70 percent of the time. That’s not great! (A recent revision by IHME fixed that issue; more on this below.)
This track record has led some experts to criticize the model. “It’s not a model that most of us in the infectious disease epidemiology field think is well suited” to making projections about Covid-19, Harvard epidemiologist Marc Lipsitch told reporters.
But if that’s the case, how has it risen to such prominence among policymakers? Other models have done better than IHME’s at predicting the course of the epidemic, and many of them use approaches epidemiologists believe are more promising. Yet it’s the IHME model that has generally guided policymakers, for the most part, in the direction of focusing on a return to normal.
A team of engineers, epidemiologists, and physicians from Johns Hopkins’ Whiting School of Engineering, Bloomberg School of Public Health, and School of Medicine today launched a new smartphone app that analyzes users’ body temperatures in a study to predict geographical areas at risk for outbreaks of the novel coronavirus, giving public health experts and government officials critical information to inform decisions on mitigation, resource allocation, and deconfinement.
The study relies on users recording their body temperatures, as well as other symptoms, daily. The free app is available under the name “COVID Control” on Google Play and in the Apple App Store.
In a recent paper, [Xi] Chen and his colleagues with the Samsung AI Center utilized the WiFi signals to establish a submeter-level localization system that employs WiFi propagation characteristics as users’ location fingerprints. The researchers also propose a WiFi-based Domain-adaptive system (FiDo) , which is able to localize new users without labelling their data.
“Indoor localization systems like FiDo can be very useful for developing intrusion detection and fall detection devices, or to add new features such as presence and activity detection for existing smart devices,” Chen told Synced.
Over the weekend, for the first time, the anonymous creator of Vocal Synthesis received a copyright claim on YouTube, taking two of his videos offline with deepfaked audio of Jay-Z reciting the “To Be or Not To Be” soliloquy from Hamlet and Billy Joel’s “We Didn’t Start the Fire.”
According to the creator, the copyright claims were filed by Roc Nation LLC with an unusual reason for removal: “This content unlawfully uses an AI to impersonate our client’s voice.”
On March 4, Hawaii had no confirmed cases of COVID-19, but officials had started to take action in anticipation of an outbreak. Gov. David Ige declared a state of emergency, giving him the authority to “suspend any law that impedes … emergency functions.” By the 16th, the outbreak had arrived: The state had 10 confirmed cases, and Ige began to act on that declaration.
Among the statutes he suspended was Chapter 92F of something called “the uniform information practices act.” It was easy for a layperson to miss, but the change effectively blocked requests for public records in the state for the duration of the emergency.
Hawaii is among several jurisdictions around the country that have amended or suspended access to public records as the coronavirus spreads.
The pandemic “has unmasked how big and powerful these companies are,” one antitrust advocate says. And that could make them an even bigger target in Washington.
Scientific American, Observations, Eszter Hargittai and Elissa Redmiles
from
In early April, as a professor of communication at the University of Zurich (Eszter Hargittai) and a researcher of security and privacy at Microsoft Research (Elissa M. Redmiles), we surveyed 1,374 adults across the U.S. about their willingness to install a coronavirus tracking app. Two thirds of Americans reported that they are willing to install an app that would help slow the spread of the virus and reduce the lockdown period, even if that app would collect information about their location data and health status.
People whom the Centers for Disease Control has identified as being higher risk, those who are younger and those who are more technologically savvy were more likely to be willing to install such an app. Those with different levels of education, genders, races and incomes were all equally likely to be willing to install.
The National Academy of Sciences announced today the election of 120 members and 26 international members in recognition of their distinguished and continuing achievements in original research.
Those elected today bring the total number of active members to 2,403 and the total number of international members to 501. International members are nonvoting members of the Academy, with citizenship outside the United States.
For its May 2020 print article “The Laser at 60,” OPN interviewed a range of OSA Fellows to get their insights on some particularly interesting insights in laser research today. We’re presenting a selection of those interviews online. Below is an edited version of our interview with Robert Byer of Stanford University, USA.
Scientists and drug companies searching for a coronavirus treatment have launched so many clinical trials that some now fear they will run out of patients to enroll, trial sites or personnel to carry out the tests.
There are more than 70 coronavirus drug and vaccine trials now registered with the Food and Drug Administration. Many, but not all, are taking place in the U.S. Although the number of new coronavirus infections nationwide is still climbing, the frenetic pace at which trials are launching — and the number that are potentially redundant or don’t involve enough patients to reach accurate conclusions — could prevent some of these studies from producing useful results.
Ms Merkel herself was able to deftly explain how the coronavirus spreads so quickly, in a way that few other leaders have been able to.
At a press conference this month, Ms Merkel — who has a doctorate in quantum chemistry — didn’t need a medical officer or chief scientist to communicate the mathematical modelling.
Online May 11-16. “A week of both fun and informative online activities for the healthtech community, including coffee mornings, webinars, Tweetchats, virtual panels, yoga and much much more!”
Online July 6-12. “Recorded SciPy talks will be released each evening throughout the week. We will gather together each day for live tutorials in the morning, a live keynote or plenary session and then Q&A and moderated discussion for each of our tracks and mini-symposia. We rounded out the week with birds of a feather gatherings, lightning talks and networking events. The week concludes with two days or remote developer sprints.” [$$]
Online May 13, starting at 1 p.m. EDT. The conference “will look at what’s already been accomplished and how to handle the hurdles that lie ahead for this shift in focus toward digital health. Discover how you can be part of this growing community, how your past successes can be applied to the evolving market, and what resources in your backyard you can utilize.” [$$$]
Online June 15, starting at 12 p.m. EDT. “Presentations by government leaders will discuss quantum initiatives and innovations revolutionizing U.S. Air Force operations.” [free, registration required]
“We promise this won’t take too long. We would like to ask you a couple of questions about making mistakes, learning & building resilience . This will help us write up a crowd sourced a community blog based on your responses.” Deadline for responses is May 11.
“We’re introducing Jukebox, a neural net that generates music, including rudimentary singing, as raw audio in a variety of genres and artist styles. We’re releasing the model weights and code, along with a tool to explore the generated samples.”
As more artificial intelligence applications move to smartphones, deep learning models are getting smaller to allow apps to run faster and save battery power. Now, MIT researchers have a new and better way to compress models.
It’s so simple that they unveiled it in a tweet last month: Train the model, prune its weakest connections, retrain the model at its fast, early training rate, and repeat, until the model is as tiny as you want.
“That’s it,” says Alex Renda, a PhD student at MIT. “The standard things people do to prune their models are crazy complicated.”
Michael Laskin, Kimin Lee, Adam Stooke, Lerrel Pinto, Pieter Abbeel, Aravind Srinivas
from
“Learning from visual observations is a fundamental yet challenging problem in reinforcement learning (RL). Although algorithmic advancements combined with convolutional neural networks have proved to be a recipe for success, current methods are still lacking on two fronts: (a) sample efficiency of learning and (b) generalization to new environments. To this end, we present RAD: Reinforcement Learning with Augmented Data, a simple plug-and-play module that can enhance any RL algorithm. We show that data augmentations such as random crop, color jitter, patch cutout, and random convolutions can enable simple RL algorithms to match and even outperform complex state-of-the-art methods across common benchmarks in terms of data-efficiency, generalization, and wall-clock speed. We find that data diversity alone can make agents focus on meaningful information from high-dimensional observations without any changes to the reinforcement learning method. On the DeepMind Control Suite, we show that RAD is state-of-the-art in terms of data-efficiency and performance across 15 environments. We further demonstrate that RAD can significantly improve the test-time generalization on several OpenAI ProcGen benchmarks. Finally, our customized data augmentation modules enable faster wall-clock speed compared to competing RL techniques.”