Uncertainty quantification of neural networks is critical to measuring the reliability and robustness of deep learning systems. However, this often involves costly or inaccurate sampling methods and approximations. This paper presents a sample-free moment propagation technique that propagates mean vectors and covariance matrices across a network to accurately characterize the input-output distributions of neural networks. A key enabler of our technique is an analytic solution for the covariance of random variables passed through nonlinear activation functions, such as Heaviside, ReLU, and GELU. The wide applicability and merits of the proposed technique are shown in experiments analyzing the input-output distributions of trained neural networks and training Bayesian neural networks.

A massive blaze is seen over Moscow's Crocus City Hall on Friday after several gunmen burst into the music venue and fired automatic weapons at the crowd, killing at least 133 people.Sergei Vedyashkin/AP After gunmen on the outskirts of Moscow opened fire at a popular concert hall Friday night, in the deadliest attack that Russia's capital has seen in more than a decade, claims and counterclaims are now mounting about who is responsible for the violence, and a deepfake video is already adding to the swirl of disinformation. In the hours after the attack at Crocus City Hall, which killed at least 133 people, the Islamic State claimed responsibility. US security officials blamed the Islamic State in Khorasan, a branch that works in Pakistan, Afghanistan, and Iran. But some Russian lawmakers quickly pointed a finger at Ukraine, and Russia's NTV television channel soon aired a deepfake video that fueled those suspicions. The fake video appeared to show Ukraine's top security official, Oleksiy Danilov, speaking about the attack.

I hope you're having a good weekend so far. Unfortunately, our recording schedule meant I didn't get to shoehorn in the fact that the Department of Justice filed an antitrust lawsuit against Apple -- it'll pop up again and again for the next six months -- but we do have Apple striking a possible deal with Google to use its Gemini AI in future iPhones. Yes, I didn't see that coming, either. If you're one of our money-to-spend readers, prepare for Dyson's next-gen robot vacuum, which is finally debuting in the US. Apple wants to bring Google's Gemini AI to iPhones Just read as Engadget Editor (and Doctor Who critic) Daniel Cooper punches Disney in the solar plexus with its awful global release strategy for the next series featuring the timelord.

Elon Musk's Neuralink has been accused of allowing and enabling animal cruelty in its labs for years, but the company is not the first to sacrifice heaps of animals on the altar of brain-computer interface science. In fact, the US government may have a worse body count over many decades, even though Neuralink has reportedly killed more than 1,500 animals along the way, including monkeys, pigs, and sheep while developing its brain chip. In some of the worst cases, a cat was operated on repeatedly to turn it into a secret listening device, and a shark was subjected to open-brain surgery to implant electrodes in an effort to control its behavior - all while the animals were still alive. A pig in the Neuralink facility, shown with its handler. Musk revealed the first recieptiant of his Neuralink brain chip this week.

Iran has made it no secret that it plans to invest heavily in artificial intelligence (AI) to help better its military capabilities, but Iranian President Ebrahim Raisi is now turning to Iran's private sector in a move he thinks will boost his crippling economy. On Sunday, Raisi met with private sector companies to announce Tehran's intent to invest in digital businesses. Raisi claimed the move would not only help develop Iran's AI capabilities, but help achieve his goal to grow the economy by 8%, reported pro-government media outlet Tasnim News Agency. However, experts remain skeptical about whether the move will actually fix Iran's economic woes and said they are more concerned by the abilities AI would grant Tehran when it comes to the battlefield. An Iranian-made unmanned aerial vehicle, the Shahed-136, is being displayed at Azadi Square in western Tehran, Iran, on Feb. 11, 2024, during a rally to mark the 45th anniversary of the victory of Iran's 1979 Islamic Revolution.

Olena Rozumovska is at the end of her rope. Her two-bedroom apartment in an Soviet-era concrete building has no electricity or water supply, and the central heating is off after Russian drones and missiles struck Kharkiv, Ukraine's second largest city, on Friday. I want to howl with despair," the 33-year-old, whose husband, Mykhailo, is fighting against Russian forces in southeastern Ukraine, told Al Jazeera over the phone. The outdoor temperatures in Kharkiv barely rose above freezing on Friday, a cold drizzle was falling, and her apartment building "is losing warmth", she said. Early in the morning, she jumped out of bed on hearing the thud of a powerful explosion. More than a dozen heavy, blood-curdling blasts followed as she hid in the frigid basement with her two children, Bohdan, who is seven, and four-year-old Roxana. The children were "hysterical" because they had to leave their Siamese cat behind. Their pet, named Monya, wouldn't come out from under the sofa. What roiled her and millions of Ukrainians was the scope of the bombardment, which became the largest strike on their nation's energy infrastructure since the war began in 2022. "The aim is not just to destroy but to try yet again, like last year, to cause a massive disruption of the energy infrastructure," Energy Minister Herman Halushchenko wrote on Facebook. In the winter of 2022-2023, Moscow switched to massive shelling that targeted energy infrastructure and civilian sites after realising that its blitzkrieg to take over all of Ukraine had failed. Friday's attacks with about 60 drones and 90 missiles killed at least two people, wounded scores, struck Ukraine's largest dam and severed the power supply to the Russia-occupied Zaporizhzhia nuclear plant, officials said. Ukrainian President Volodymyr Zelenskyy rebuked the West for months-long delays in military aid. "Russian missiles have no delays, unlike aid packages for Ukraine.

Learning and Artificial Intelligence (ML/AI) techniques have become increasingly prevalent in high performance computing (HPC). However, these methods depend on vast volumes of floating point data for training and validation which need methods to share the data on a wide area network (WAN) or to transfer it from edge devices to data centers. Data compression can be a solution to these problems, but an in-depth understanding of how lossy compression affects model quality is needed. Prior work largely considers a single application or compression method. We designed a systematic methodology for evaluating data reduction techniques for ML/AI, and we use it to perform a very comprehensive evaluation with 17 data reduction methods on 7 ML/AI applications to show modern lossy compression methods can achieve a 50-100x compression ratio improvement for a 1% or less loss in quality. We identify critical insights that guide the future use and design of lossy compressors for ML/AI.

This paper addresses the challenges of distributed formation control in multiple mobile robots, introducing a novel approach that enhances real-world practicability. We first introduce a distributed estimator using a variable structure and cascaded design technique, eliminating the need for derivative information to improve the real time performance. Then, a kinematic tracking control method is developed utilizing a bioinspired neural dynamic-based approach aimed at providing smooth control inputs and effectively resolving the speed jump issue. Furthermore, to address the challenges for robots operating with completely unknown dynamics and disturbances, a learning-based robust dynamic controller is developed. This controller provides real time parameter estimates while maintaining its robustness against disturbances. The overall stability of the proposed method is proved with rigorous mathematical analysis. At last, multiple comprehensive simulation studies have shown the advantages and effectiveness of the proposed method.

Users are dissatisfied with services. Since the service is not tailor-made for a user, it is natural for dissatisfaction to arise. The problem is, that even if users are dissatisfied, they often do not have the means to resolve their dissatisfaction. The user cannot alter the source code of the service, nor can they force the service provider to change. The user has no choice but to remain dissatisfied or quit the service. User-side realization offers proactive solutions to this problem by providing general algorithms to deal with common problems on the user's side. These algorithms run on the user's side and solve the problems without having the service provider change the service itself.

Optical Character Recognition (OCR) is an established task with the objective of identifying the text present in an image. While many off-the-shelf OCR models exist, they are often trained for either scientific (e.g., formulae) or generic printed English text. Extracting text from chemistry publications requires an OCR model that is capable in both realms. Nougat, a recent tool, exhibits strong ability to parse academic documents, but is unable to parse tables in PubMed articles, which comprises a significant part of the academic community and is the focus of this work. To mitigate this gap, we present the Printed English and Chemical Equations (PEaCE) dataset, containing both synthetic and real-world records, and evaluate the efficacy of transformer-based OCR models when trained on this resource. Given that real-world records contain artifacts not present in synthetic records, we propose transformations that mimic such qualities. We perform a suite of experiments to explore the impact of patch size, multi-domain training, and our proposed transformations, ultimately finding that models with a small patch size trained on multiple domains using the proposed transformations yield the best performance. Our dataset and code is available at https://github.com/ZN1010/PEaCE.