Convex relaxations are effective for training and certifying neural networks against norm-bounded adversarial attacks, but they leave a large gap between certifiable and empirical robustness. In principle, convex relaxation can provide tight bounds if the solution to the relaxed problem is feasible for the original non-convex problem. We propose two regularizers that can be used to train neural networks that yield tighter convex relaxation bounds for robustness. In all of our experiments, the proposed regularizers result in higher certified accuracy than non-regularized baselines.

This work presents a non-intrusive model reduction method to learn low-dimensional models of dynamical systems with non-polynomial nonlinear terms that are spatially local and that are given in analytic form. In contrast to state-of-the-art model reduction methods that are intrusive and thus require full knowledge of the governing equations and the operators of a full model of the discretized dynamical system, the proposed approach requires only the non-polynomial terms in analytic form and learns the rest of the dynamics from snapshots computed with a potentially black-box full-model solver. The proposed method learns operators for the linear and polynomially nonlinear dynamics via a least-squares problem, where the given non-polynomial terms are incorporated in the right-hand side. The least-squares problem is linear and thus can be solved efficiently in practice. The proposed method is demonstrated on three problems governed by partial differential equations, namely the diffusion-reaction Chafee-Infante model, a tubular reactor model for reactive flows, and a batch-chromatography model that describes a chemical separation process. The numerical results provide evidence that the proposed approach learns reduced models that achieve comparable accuracy as models constructed with state-of-the-art intrusive model reduction methods that require full knowledge of the governing equations.

We aim to improve question answering (QA) by decomposing hard questions into easier sub-questions that existing QA systems can answer. Since collecting labeled decompositions is cumbersome, we propose an unsupervised approach to produce sub-questions. Specifically, by leveraging >10M questions from Common Crawl, we learn to map from the distribution of multi-hop questions to the distribution of single-hop sub-questions. We answer sub-questions with an off-the-shelf QA model and incorporate the resulting answers in a downstream, multi-hop QA system. On a popular multi-hop QA dataset, HotpotQA, we show large improvements over a strong baseline, especially on adversarial and out-of-domain questions. Our method is generally applicable and automatically learns to decompose questions of different classes, while matching the performance of decomposition methods that rely heavily on hand-engineering and annotation.

Using a machine-learning algorithm, MIT researchers have identified a powerful new antibiotic compound. In laboratory tests, the drug killed many of the world's most problematic disease-causing bacteria, including some strains that are resistant to all known antibiotics. It also cleared infections in two different mouse models. The computer model, which can screen more than a hundred million chemical compounds in a matter of days, is designed to pick out potential antibiotics that kill bacteria using different mechanisms than those of existing drugs. "We wanted to develop a platform that would allow us to harness the power of artificial intelligence to usher in a new age of antibiotic drug discovery," says James Collins, the Termeer Professor of Medical Engineering and Science in MIT's Institute for Medical Engineering and Science (IMES) and Department of Biological Engineering.

Researchers announced the breakthrough discovery of a new type of antibiotic compound that is capable of killing many types of harmful bacteria, including deadly antibiotic-resistant strains, and published their findings in Cell on February 20. What makes this remarkable is that the researchers, from the Massachusetts Institute of Technology (MIT), Harvard, and McMaster University, used machine learning (a form of artificial intelligence) to discover the new antibiotic--an achievement that heralds the disruption of traditional research and drug development processes deployed by pharmaceutical industry behemoths. Antibiotic resistance is a global threat that is exacerbated by the overuse of antibiotics in livestock, the proliferation of antimicrobials in consumer products, and over-prescription in health care. Though estimating the future impact is challenging, one report predicted that by 2050, 10 million deaths per year could result from antimicrobial-resistant (AMR) infections. Combating the problem of antimicrobial resistance requires bringing novel compounds to market.

NASA is running out of options in its mission to get its InSight lander's probe back on track. According to the agency, it will attempt to use a robotic arm attached to its InSight Lander to push down on a probe meant to drill into Martian soil which has struggled to achieve its mission throughout the past year. NASA says the goal is to stop the probe from popping out of its partially dug hole which it has done twice in recent months in addition to almost burying itself. While the act of pushing down on the probe with the arm should be relatively easy, NASA acknowledges that choosing to do so could create problems for the instrument if too much force is applied. The worry is that pushing down with the arm may damage a ribbon-like stretch of wires that attaches to InSight.

Being a cybersecurity analyst at a large company today is a bit like looking for a needle in a haystack -- if that haystack were hurtling toward you at fiber optic speed. Every day, employees and customers generate loads of data that establish a normal set of behaviors. An attacker will also generate data while using any number of techniques to infiltrate the system; the goal is to find that "needle" and stop it before it does any damage. The data-heavy nature of that task lends itself well to the number-crunching prowess of machine learning, and an influx of AI-powered systems have indeed flooded the cybersecurity market over the years. But such systems can come with their own problems, namely a never-ending stream of false positives that can make them more of a time suck than a time saver for security analysts.

Local councils in England must be given powers to regulate Airbnb and other short-term letting sites in order to alleviate the "intolerable" pressure they put on the availability of local housing, the Green party MP, Caroline Lucas, has said. Her intervention followed a Guardian investigation that found one Airbnb listing for every four residential properties in some hotspots across Britain. Airbnb has disputed the finding. Meanwhile, an organisation representing landlords has warned that imminent tax changes will drive an increasing number of landlords towards Airbnb and its rivals, depriving renters of long-term, stable tenancies. Last month Lucas asked the government to make it easier for councils to impose a 90-day cap on homes let out on Airbnb and other online platforms.

M1A2 Abrams Tanks from A Company, 2-116th Cavalry Brigade Combat Team (CBCT), Idaho Army National Guard run through field exercises on Orchard Combat Training Center - file photo. Should U.S. forces be facing a massive armored enemy ground vehicle assault, they would need their own heavily armored vehicles -- such as infantry carriers, ground forces, unmanned attack vehicles and, perhaps of greatest significance, Abrams tanks. Large numbers of heavily armed, integrated and ready Abrams tanks would be needed for any kind of major ground offensive and "ready for war." Achieving this is not always as easy as it may sound; Abrams tanks are complex war machines that rely upon a wide range of properly functioning systems and technologies, including ammunition, mounted weapons, armor, sensors and electronics. Abrams parts often need to be repaired, upgraded and effectively maintained.

The European Commission released long-awaited position papers Wednesday on several key digital issues, including how to treat the continent's digital data and how best to regulate artificial intelligence. Why it matters: Europe has traditionally trailed the U.S. in creating giant tech companies that gobble up consumer data, but it has led in issuing rules and policies to govern such practices. The European Data Strategy and the AI recommendations have themes will be familiar. What they're saying: Not surprisingly, many trade groups released statements praising the goals of the proposals, while urging restraint in regulation. Meanwhile: Cornell business professor Thomas Jungbauer argues the proposals aren't what's needed to help Europe catch up.