This report considers the problem of Byzantine fault-tolerance in homogeneous multi-agent distributed learning. In this problem, each agent samples i.i.d. data points, and the goal for the agents is to compute a mathematical model that optimally fits, in expectation, the data points sampled by all the agents. We consider the case when a certain number of agents may be Byzantine faulty. Such faulty agents may not follow a prescribed learning algorithm. Faulty agents may share arbitrary incorrect information regarding their data points to prevent the non-faulty agents from learning a correct model. We propose a fault-tolerance mechanism for the distributed stochastic gradient descent (D-SGD) method -- a standard distributed supervised learning algorithm. Our fault-tolerance mechanism relies on a norm based gradient-filter, named comparative gradient elimination (CGE), that aims to mitigate the detrimental impact of malicious incorrect stochastic gradients shared by the faulty agents by limiting their Euclidean norms. We show that the CGE gradient-filter guarantees fault-tolerance against a bounded number of Byzantine faulty agents if the stochastic gradients computed by the non-faulty agents satisfy the standard assumption of bounded variance. We demonstrate the applicability of the CGE gradient-filer for distributed supervised learning of artificial neural networks. We show that the fault-tolerance by the CGE gradient-filter is comparable to that by other state-of-the-art gradient-filters, namely the multi-KRUM, geometric median of means, and coordinate-wise trimmed mean. Lastly, we propose a gradient averaging scheme that aims to reduce the sensitivity of a supervised learning process to individual agents' data batch-sizes. We show that gradient averaging improves the fault-tolerance property of a gradient-filter, including, but not limited to, the CGE gradient-filter.

Federated learning (FL) is an emerging paradigm for distributed training of large-scale deep neural networks in which participants' data remains on their own devices with only model updates being shared with a central server. However, the distributed nature of FL gives rise to new threats caused by potentially malicious participants. In this paper, we study targeted data poisoning attacks against FL systems in which a malicious subset of the participants aim to poison the global model by sending model updates derived from mislabeled data. We first demonstrate that such data poisoning attacks can cause substantial drops in classification accuracy and recall, even with a small percentage of malicious participants. We additionally show that the attacks can be targeted, i.e., they have a large negative impact only on classes that are under attack. We also study attack longevity in early/late round training, the impact of malicious participant availability, and the relationships between the two. Finally, we propose a defense strategy that can help identify malicious participants in FL to circumvent poisoning attacks, and demonstrate its effectiveness.

The White House and Defense Department on Monday announced a plan to accelerate the process by making a crucial new chunk of spectrum available to the wireless industry. The spectrum, which telecom companies will share with the Pentagon, aims to help wireless carriers offer 5G more broadly across the US. It also should generate billions of dollars for the US Treasury when auctioned off. The frequency is currently being used for high-power defense radar, but the DoD has determined that it can be freed up without affecting military systems. "It's a big deal," for the wireless industry, says Jason Leigh, an analyst at IDC who focuses on 5G.

A newly published study by University of Michigan researchers shows facial recognition technology in schools presents multiple problems and has limited efficacy. Led by Shobita Parthasarathy, director of the university's Science, Technology, and Public Policy (STPP) program, the research say the technology isn't suited to security purposes and can actively promote racial discrimination, normalize surveillance, and erode privacy while institutionalizing inaccuracy and marginalizing non-conforming students. The study follows the New York legislature's passage of a moratorium on the use of facial recognition and other forms of biometric identification in schools until 2022. The bill, which came in response to the launch of facial recognition by the Lockport City School District, was among the first in the nation to explicitly regulate or ban use of the technology in schools. That development came after companies including Amazon, IBM, and Microsoft halted or ended the sale of facial recognition products in response to the first wave of Black Lives Matter protests in the U.S. The Michigan University study -- a part of STPP's Technology Assessment Project -- employs an analogical case comparison method to look at previous uses of security technology like CCTV cameras and metal detectors as well as biometric technologies and anticipate the implications of facial recognition.

Members of Congress are pushing the Pentagon to invest more in artificial intelligence for warfighting and improving business operations. The Defense Department has identified microelectronics, 5G communications and hypersonics as its top three research-and-development priorities. But House Armed Services Committee Chairman Rep. Adam Smith, D-Wash., said AI should be at the top of the list. "The most important technological advance is AI," he told reporters. "How we develop our AI technology and how we use it is going to be, I think, the No.1 priority" for those pulling the purse strings.

Managing Partner and Co-Founder of Scale-Up VC, a Silicon Valley venture capital firm based in Palo Alto, California. Experts have warned against its potential misuse. It's now affecting aspects of our lives that many of us never anticipated: healthcare, education, employment and even national security. What could I be talking about? Artificial intelligence, or the "big AI," as I call it.

New York – The robots are no longer coming; they are here. The COVID-19 pandemic is hastening the spread of artificial intelligence, but few have fully considered the short- and long-run consequences. In thinking about AI, it is natural to start from the perspective of welfare economics -- productivity and distribution. What are the economic effects of robots that can replicate human labor? Such concerns are not new.

Privacy campaigners have expressed alarm after the government revealed it had hired an artificial intelligence firm to collect and analyse the tweets of UK citizens as part of a coronavirus-related contract. Faculty, which was hired by Dominic Cummings to work for the Vote Leave campaign and counts two current and former Conservative ministers among its shareholders, was paid £400,000 by the Ministry of Housing, Communities and Local Government for the work, according to a copy of the contract published online. In June the Guardian reported Faculty had been awarded the contract, but that key passages in the published version of the document describing the work that the company would carry out had been redacted. In response to questions about the contract in the House of Lords, the government published an unredacted version of the contract, which describes the company's work as "topic analysis of social media to understand public perception and emerging issues of concern to HMG arising from the Covid-19 crisis". A further paragraph describes how machine learning will be applied to social media data.

In the near future, autonomous vehicles and artificial intelligence (AI) will play a larger role in how your food is grown. Farming is as old as civilization itself, but with industrialization, modern agriculture grew in scale and sophistication to degrees never seen before in history, especially during the Green Revolution of the 1950s-60s. The sector may be poised to go through another comparable evolutionary step with machines doing important jobs in the fields. The United Nations projects the global population will increase to 9.73 billion people by 2050. While 60 percent of the global population lived in rural areas 35 years ago, about 54 percent now live in urban ones.

An upcoming event to display and test AI-powered jet fighters will now be held virtually due to COVID-19. "We are still excited to see how the AI algorithms perform against each other as well as a Weapons School-trained human and hope that fighter pilots from across the Air Force, Navy, and Marine Corps, as well as military leaders and members of the AI tech community will register and watch online," said Col. Dan Javorsek, program manager in DARPA's Strategic Technology Office. "It's been amazing to see how far the teams have advanced AI for autonomous dogfighting in less than a year." DARPA (Defense Advanced Research Projects Agency) is using the AlphaDogfight Trial event to recruit more AI developers for its Air Combat Evolution (ACE) program. The upcoming event is the final in a series of three and will finish with a bang as the AI-powered F-16 fighter planes virtually take on a human pilot.