Ensemble approaches for uncertainty estimation have recently been applied to the tasks of misclassification detection, out-of-distribution input detection and adversarial attack detection. Prior Networks have been proposed as an approach to efficiently emulating an ensemble of models by parameterising a Dirichlet prior distribution over output distributions. These models have been shown to outperform ensemble approaches, such as Monte-Carlo Dropout, on the task of out-of-distribution input detection. However, scaling Prior Networks to complex datasets with many classes is difficult using the training criteria originally proposed. This paper makes two contributions. Firstly, we show that the appropriate training criterion for Prior Networks is the reverse KL-divergence between Dirichlet distributions. Using this loss we successfully train Prior Networks on image classification datasets with up to 200 classes and improve out-of-distribution detection performance. Secondly, taking advantage of the new training criterion, this paper investigates using Prior Networks to detect adversarial attacks. It is shown that the construction of successful adaptive whitebox attacks, which affect the prediction and evade detection, against Prior Networks trained on CIFAR-10 and CIFAR-100 takes a greater amount of computational effort than against standard neural networks, adversarially trained neural networks and dropout-defended networks.

Effective complements to human judgment, artificial intelligence techniques have started to aid human decisions in complicated social problems across the world. In the context of United States for instance, automated ML/DL classification models offer complements to human decisions in determining Medicaid eligibility. However, given the limitations in ML/DL model design, these algorithms may fail to leverage various factors for decision making, resulting in improper decisions that allocate resources to individuals who may not be in the most need. In view of such an issue, we propose in this paper the method of \textit{fairgroup construction}, based on the legal doctrine of \textit{disparate impact}, to improve the fairness of regressive classifiers. Experiments on American Community Survey dataset demonstrate that our method could be easily adapted to a variety of regressive classification models to boost their fairness in deciding Medicaid Eligibility, while maintaining high levels of classification accuracy.

Neural networks have successfully been applied to solving reasoning tasks, ranging from learning simple concepts like "close to", to intricate questions whose reasoning procedures resemble algorithms. Empirically, not all network structures work equally well for reasoning. For example, Graph Neural Networks have achieved impressive empirical results, while the less structured neural networks may fail to learn to reason. Theoretically, there is currently limited understanding of the interplay between reasoning tasks and network learning. In this paper, we develop a framework to characterize which tasks a neural network can learn well, by studying how well its structure aligns with the algorithmic structure of the relevant reasoning procedure. This suggests that Graph Neural Networks can learn dynamic programming, a powerful algorithmic strategy that solves a broad class of reasoning problems, such as relational question answering, sorting, intuitive physics, and shortest paths. Our perspective also implies strategies to design neural architectures for complex reasoning. On several abstract reasoning tasks, we see empirically that our theory aligns well with practice.

World leaders in technology are uniting to establish a common set of guidelines on the use of artificial intelligence and reel in the potential for misuse. The Global AI Council, which was created as part of a summit by the World Economic Forum in San Francisco, will focus not just on establishing standards for how AI should and shouldn't be applied across fields, but in making those standards mesh among world powers, particularly the U.S. and China. The goal of connecting disparate governments is arguably best exemplified through the council's leaders -- Microsoft President Brad Smith and Chinese AI expert Kai-Fu Lee. According to a statement from the World Economic Forum, specifically, the council hopes to establish channels of communication between partners of the council on best practices and case studies as well as addressing what it calls'governance gaps' -- presumably areas where regulation has yet to keep up with potentially harmful technology. As noted by MIT Technology Review, one particular area that will likely be a flashpoint for regulatory and ethical guidelines surrounding AI is surveillance.

Need to extract content from a document quickly and automatically? Amazon today announced the general availability of Textract, a cloud-hosted and fully managed service that uses machine learning to parse data tables, forms, and whole pages for text and data. Virginia), US West (Oregon), and EU (Ireland) regions and will expand to additional regions in the coming year. Textract is more capable than your average optical character recognition system. From files stored in an Amazon S3 bucket, it's able to suss out the contents of fields and tables and the context in which this information is presented, like names and social security numbers in tax forms or totals from photographed receipts.

By Pavel Sharikov, for the EastWest Institute Artificial intelligence (AI) -- defined by John McCarthy, one of the doyens of AI, as "the science and engineering of making intelligent machines" -- is slowly gaining relevance in the military domain. While commercial use of AI is widening, there are only three countries that are reported to be developing serious military AI technologies: the United States, China and Russia. AI promises a significant military advantage to a nation's offensive and defensive military capabilities.

Digital technologies are becoming ubiquitous, effective, and cost efficient, but are underutilized in medicine. These technologies -- like wearable health monitors, sensors, and even ingestible devices that can measure everything from how many steps you take, to blood pressure, and how a drug interacts with your body once ingested -- have the potential to disrupt every aspect of health care, including high-stakes, high-cost drug development. Specifically, these devices can revolutionize the antiquated process of developing new drug therapies and can vastly improve how we collect, measure, and assess health data so that we can offer new treatments to patients without wasting valuable time and limited resources. Clinical trials are designed to evaluate whether a new drug is safe and effective while protecting volunteer patients participating in the trials from risk. All good intentions, but some of today's research processes date back to 1946. New and proven digital technologies can make drug development smarter, better, and faster.

On Tuesday, news broke that Microsoft refused to sell its facial recognition software to law enforcement in California and an unnamed country. The move led to some praise for the company for being consistent with its policy to oppose questionable human rights applications, but a broader examination of Microsoft's actions in the past year indicates that the company has been saying one thing and doing another. Last week, the Financial Times reported that Microsoft Research Asia worked with a university associated with the Chinese military on facial recognition tech that is being used to monitor the nation's population of Uighur Muslims. Up to 500,000 members of the group, primarily in western China, were monitored over the course of a month, according to a New York Times report. Microsoft defended the work as helpful to advance the technology, but U.S. Senator Marco Rubio called the company complicit in human rights abuses.

WHEN I was growing up, nobody promised me a flying car. But I was promised an AI apocalypse. Those shiny machines were going to crush our skulls underfoot, and we were all going to welcome our new robot overlords. Many people still seem to think it is likely to happen. But we might still get a deadly AI nightmare.

Deepfakes are video manipulations that can make people say seemingly strange things. Barack Obama and Nicolas Cage have been featured in these videos. SAN FRANCISCO – When a doctored video of House Speaker Nancy Pelosi – one altered to show the Democratic leader slurring her words – began making the rounds on Facebook last week, the social network didn't take it down. Instead, it "downranked" the video, a behind-the-scenes move intended to limit its spread. That outraged some people who believe Facebook should do more to clamp down on misinformation.