We find a surprising connection between multitask learning and robustness to neuron failures. Our experiments show that bilingual language models retain higher performance under various neuron perturbations, such as random deletions, magnitude pruning and weight noise. Our study is motivated by research in cognitive science showing that symptoms of dementia and cognitive decline appear later in bilingual speakers compared to monolingual patients with similar brain damage, a phenomenon called bilingual cognitive reserve. Our language model experiments replicate this phenomenon on bilingual GPT-2 and other models.We provide a theoretical justification of this robustness by mathematically analyzing linear representation learning and showing that multitasking creates more robust representations.

We find a surprising connection between multitask learning and robustness to neuron failures. Our experiments show that bilingual language models retain higher performance under various neuron perturbations, such as random deletions, magnitude pruning and weight noise. Our study is motivated by research in cognitive science showing that symptoms of dementia and cognitive decline appear later in bilingual speakers compared to monolingual patients with similar brain damage, a phenomenon called bilingual cognitive reserve. Our language model experiments replicate this phenomenon on bilingual GPT-2 and other models.We provide a theoretical justification of this robustness by mathematically analyzing linear representation learning and showing that multitasking creates more robust representations.

On January 28, 2022, at 6.39 a.m. EST, the Fern Hollow Bridge in Pittsburgh, Pennsylvania collapsed. Due to the timing of the failure, thankfully, fewer vehicles were on the bridge and only ten people were injured with no fatalities. Pittsburgh, also known as the City of Bridges, was getting ready for a visit from President Biden that day. Biden visited the collapse site and assured federal assistance to rebuild the bridge on the spot. This infrastructural failure, coinciding with a high-profile political visit and a push towards passing the Build Back Better infrastructure bill, attracted considerable media attention to the flailing infrastructural health in the US. As we were sifting through the social web discussions surrounding this issue, broad themes such as words of compassion for the victims and typical responses in social web political discourse such as political name-calling, conspiracy theories, and partisan mud-slinging emerged. However, apart from these expected social web reactions, we noticed a small minority of interactions that talked about anticipatory failures of other bridges in the US.

If the world is ever going to enjoy the upsides of a sky filled with drones, the unmanned aircraft must be able to behave at least as well as human pilots. They must know how to react to other aircraft coming right for them, how to manage sudden weather changes, and what to do when their vehicle goes haywire. That's why researchers at NASA's Langley Research Center in Virginia have developed a system that can help with one slice of drone troubleshooting: enabling small UAVs to determine on their own when they're not working properly, and then find a safe place to land. Safe2Ditch, invented by Langley's Trish and Lou Glaab, is designed for fully autonomous aircraft without human pilots at the controls. It uses software algorithms to detect battery or motor problems, control-surface or structural failures, or even shifting cargo that can disrupt the aircraft's balance.

The National Transportation Safety Board is looking into an accident involving an enormous experimental drone belonging to Facebook which crashed earlier this year. The drone was built to bring internet to far-flung parts of the planet and designed to hover 60,000 ft above the surface of the Earth. However, in June the drone known as Aquila, ended up plummeting to earth after suffering a'structural failure' as it was coming into land. No one was hurt in the incident. The drone, called Aquila, suffered from a'structural failure' right before it landed According to Bloomberg the crash is one of several hiccups that the social network has faced in recent months, following an explosion earlier this year that destroyed one of its satellites.