Deep learning models are known to be vulnerable not only to input-dependent adversarial attacks but also to input-agnostic or universal adversarial attacks. Dezfooli et al. \cite{Dezfooli17,Dezfooli17anal} construct universal adversarial attack on a given model by looking at a large number of training data points and the geometry of the decision boundary near them. Subsequent work \cite{Khrulkov18} constructs universal attack by looking only at test examples and intermediate layers of the given model. In this paper, we propose a simple universalization technique to take any input-dependent adversarial attack and construct a universal attack by only looking at very few adversarial test examples. We do not require details of the given model and have negligible computational overhead for universalization. We theoretically justify our universalization technique by a spectral property common to many input-dependent adversarial perturbations, e.g., gradients, Fast Gradient Sign Method (FGSM) and DeepFool. Using matrix concentration inequalities and spectral perturbation bounds, we show that the top singular vector of input-dependent adversarial directions on a small test sample gives an effective and simple universal adversarial attack. For VGG16 and VGG19 models trained on ImageNet, our simple universalization of Gradient, FGSM, and DeepFool perturbations using a test sample of 64 images gives fooling rates comparable to state-of-the-art universal attacks \cite{Dezfooli17,Khrulkov18} for reasonable norms of perturbation.

Rule based classifiers that use the presence and absence of key sub-strings to make classification decisions have a natural mechanism for quantifying the uncertainty of their precision. For a binary classifier, the key insight is to treat partitions of the sub-string set induced by the documents as Bernoulli random variables. The mean value of each random variable is an estimate of the classifier's precision when presented with a document inducing that partition. These means can be compared, using standard statistical tests, to a desired or expected classifier precision. A set of binary classifiers can be combined into a single, multi-label classifier by an application of the Dempster-Shafer theory of evidence. The utility of this approach is demonstrated with a benchmark problem.

Iran almost got into a full-blown conflict with the United States when a U.S. drone strike killed top Iranian military commander Qassem Soleimani in Baghdad on Jan. 3, prompting Tehran to retaliate with a missile barrage against a U.S. base in Iraq days later.

The United States Patent and Trademark Office officially selected a new partner to support its increasing adoption of artificial intelligence and machine learning capabilities.General Dynamics Information Technology on Monday announced it was awarded a contract worth up to $50 million through its Intelligent Automation and Innovation Support Services blanket purchase agreement. GDIT is the latest of more than a dozen companies the agency tapped under the future-facing BPA. Other businesses who've made their own recent announcements detailing partnerships via the agreement include Octo and Steampunk.In the announcement, Vice President & General Manager Christopher Hegedus for GDIT's Diplomacy, Commerce and Government Operations business area noted the company's supported the agency for nearly two decades, and through this "new work, [aims to bring its] AI, ML and robotic process automation expertise to help USPTO develop solutions that accelerate the patent and trademark process to benefit American innovators." Charged with issuing patents for inventions and registering trademarks for product and intellectual property identification, USPTO is making deliberate moves to "propel" itself into the next decade technologically, the agency's chief information officer recently told Nextgov. And it appears the BPA is one avenue helping it to do exactly that.

Science's COVID-19 reporting is supported by the Pulitzer Center. Timothy Sheahan, a virologist studying COVID-19, wishes he could keep pace with the growing torrent of new scientific papers about the disease and the novel coronavirus that causes it. But there are just too many--more than 4000 alone last week. "I'm not keeping up," says Sheahan, who works at the University of North Carolina, Chapel Hill. A loose-knit army of data scientists, software developers, and journal publishers is pressing hard to change that.

NASA's Artemis program will eventually need robots to help live off the lunar soil, and it's enlisting help from the public to make those robots viable. The space agency has picked winners from a design challenge that tasked people with improving the bucket drums RASSOR (Regolith Advanced Surface Systems Operations Robot) will use to dig on the Moon. The victors all had clever designs that should capture lunar regolith with little effort -- important when any long-term presence might depend on bots like this. The winner was a trap from Caleb Clausing that uses a passive door to grab large amounts of soil while remaining dust-tolerant. Others included a simple-yet-effective drum from Michael R, another from Kyle St. Thomas that uses narrow drums, an efficient double-helix design from Stephan Weiβenböck and a model from Clix that uses both gravity and weight to help movement.

A quick web search for "chatbots and security" brings up results warning you about the security risks of using these virtual agents. Dig a little deeper, however, and you'll find that this artificial intelligence (AI) technology could actually help address many work-from-home cybersecurity challenges -- such as secure end-to-end encryption and user authentication -- and ensure that your organization continues to prove its data privacy compliance with less direct oversight. While many companies rely on chatbots to answer customer questions or step through a process, that same service can be used to help employees connect with security professionals as they work remotely, allowing many security problems to be resolved as efficiently as they would be if the security team were able to come directly to their colleagues' desks. Between 2005 and 2018, the number of remote workers grew by 173 percent, 11 percent faster than the rest of the workforce, according to Global Workplace Analytics. And as more employees and management experience the benefits of working from home, more people will demand the opportunity.

Like crystal balls for the universe's deeper mysteries, galaxies and other massive space objects can serve as lenses to more distant objects and phenomena along the same path, bending light in revelatory ways. Gravitational lensing was first theorized by Albert Einstein more than 100 years ago to describe how light bends when it travels past massive objects like galaxies and galaxy clusters. These lensing effects are typically described as weak or strong, and the strength of a lens relates to an object's position and mass and distance from the light source that is lensed. Strong lenses can have 100 billion times more mass than our sun, causing light from more distant objects in the same path to magnify and split, for example, into multiple images, or to appear as dramatic arcs or rings. The major limitation of strong gravitational lenses has been their scarcity, with only several hundred confirmed since the first observation in 1979, but that's changing, and fast.

As Japan's second female astronaut to fly up in the Space Shuttle Discovery, Naoko Yamazaki didn't expect to spend a quarter of her time dusting, feeding mice and doing other menial jobs. It can cost more than $430 million a year to keep an astronaut in orbit, according to three-year-old startup called Gitai Inc. It's only possible to keep humans alive in outer space because of the money and effort poured into ensuring their safety. One way to bring down the cost and risks is to send an avatar -- a remotely controlled robot. "There's a need for robots that can help us," Yamazaki, 49, said.

The aim of this Foresight Brief is therefore merely to initiate a debate.