We replace the output layer of deep neural nets, typically the softmax function, by a novel interpolating function. And we propose end-to-end training and testing algorithms for this new architecture. Compared to classical neural nets with softmax function as output activation, the surrogate with interpolating function as output activation combines advantages of both deep and manifold learning. The new framework demonstrates the following major advantages: First, it is better applicable to the case with insufficient training data. Second, it significantly improves the generalization accuracy on a wide variety of networks. The algorithm is implemented in PyTorch, and the code is available at https://github.com/

Deep neural networks (DNNs) are computationally/memory-intensive and vulnerable to adversarial attacks, making them prohibitive in some real-world applications. By converting dense models into sparse ones, pruning appears to be a promising solution to reducing the computation/memory cost. This paper studies classification models, especially DNN-based ones, to demonstrate that there exists intrinsic relationships between their sparsity and adversarial robustness. Our analyses reveal, both theoretically and empirically, that nonlinear DNN-based classifiers behave differently under $l_2$ attacks from some linear ones. We further demonstrate that an appropriately higher model sparsity implies better robustness of nonlinear DNNs, whereas over-sparsified models can be more difficult to resist adversarial examples.

Asynchronous Gibbs sampling has been recently shown to be fast-mixing and an accurate method for estimating probabilities of events on a small number of variables of a graphical model satisfying Dobrushin's condition [DSOR16]. We investigate whether it can be used to accurately estimate expectations of functions of all the variables of the model. Under the same condition, we show that the synchronous (sequential) and asynchronous Gibbs samplers can be coupled so that the expected Hamming distance between their (multivariate) samples remains bounded by O(τ log n), where n is the number of variables in the graphical model, and τ is a measure of the asynchronicity. A similar bound holds for any constant power of the Hamming distance. Hence, the expectation of any function that is Lipschitz with respect to a power of the Hamming distance, can be estimated with a bias that grows logarithmically in n. Going beyond Lipschitz functions, we consider the bias arising from asynchronicity in estimating the expectation of polynomial functions of all variables in the model. Using recent concentration of measure results [DDK17, GLP17, GSS18], we show that the bias introduced by the asynchronicity is of smaller order than the standard deviation of the function value already present in the true model. We perform experiments on a multiprocessor machineto empirically illustrate our theoretical findings.

Machine learning and AI are the latest tools being utilised by businesses to help prevent cybercrime and stop cyberattacks. However, what if these same tools could be used to help stop insider threats such as employees leaking or misusing company data? TechRadar Pro spoke with Jazz Networks' CEO Hani Mustafa who explained how businesses have begun to employ user behaviour analytics (UBA) to help prevent negligence, fraud, data misuse and even sabotage. Yes, however employees that maliciously or unintentionally misuse data are a problem for everyone. E-commerce platforms can have a more seemingly immediate impact (if customer and payment data is exfiltrated), but the damages to any company can be detrimental.

The head of MI6, Alex Younger, says there is a need for "fourth-generation espionage" as adversaries take advantage of "blurred lines" between the cyber and physical worlds. Here are some of the ways Britain's security services are adopting, and responding to, modern technology. These are potentially valuable tools for agencies seeking to spot leads within vast volumes of data. The Government's latest counter-terrorism strategy states that developments in artificial intelligence will "allow us to filter and identify crucial information faster than ever", adding that the power of quantum computing can be combined with AI to improve the speed at which large datasets can be sorted and mined for key information. Data analysis capabilities are seen as particularly important for monitoring more than 20,000 former MI5 "subjects of interest" who are not under active investigation.

Pittsburgh, USA – Bosch in North America and Astrobotic Technology Inc. today announced a research partnership to send experimental sensor technology to the International Space Station (ISS) as early as May 2019. Bosch's SoundSee technology is a deep audio analytics capability that uses a custom array of microphones and machine learning to analyze information contained in emitted noises. SoundSee's analytics will investigate whether audio data from equipment could be learned and understood using advanced software, such that it could be used to improve the operations of the ISS. "Machines, such as motors and pumps, emit noise signatures while they operate," said Dr. Samarjit Das, principal researcher and SoundSee project lead at Bosch's Research and Technology Center in Pittsburgh. "Our SoundSee AI (artificial intelligence) algorithm uses machine learning to analyze these subtle acoustic clues and determine whether a machine, or even a single component of a machine, needs to be repaired or replaced."

U.S. Ambassador to the United Nations Nikki Haley said in a December interview with a small group of columnists in New York that one of the largest threats facing America was cyber warfare. A handful of House members suggested in a September letter to Director of National Intelligence Dan Coats that "deepfakes" -- videos that use artificial intelligence to develop realistic images of individuals -- were one of America's biggest emerging vulnerabilities. And over the last year or so, a group of the world's preeminent scientists and researchers, 26 in total, banded together to write a report warning of the time when artificial intelligence would surpass human performance and basically, take over much of what humanity currently does -- driving, writing, warring, even creating. That was their way of saying that one of America's most serious concerns was the potential for total technological dominance. All these sources suggest having some level of inside knowledge of the biggest tech-related security threats facing America.

It's the boring stocks that let you sleep well at night, and that's because there is a lot of money being made in boring industries. Take the cleaning industry for example. Nobody wants to work in an unclean office environment, and there are rules and regulations that compel employers to make sure everything is up to snuff. All that means loads of cleaning needs to be done every night. According to the Bureau of Labor Statistics, there are 2,384,600 building janitors and cleaners representing an annual spend of nearly $60 billion.

We may remember 2018 as a year in which great power rivalry materialized at the forefront of American strategy -- with emerging technologies as a critical dimension of this competition. At a time when divisive politics and intense partisanship have undermined solutions to even the most urgent policy dilemmas, there are reasons nonetheless for cautious optimism about the potential for progress on issues of technology, innovation and competition. The current advances in emerging technologies possess strategic significance in their own right, yet take on greater urgency because of this rivalry. Such new frontiers as biotechnology, artificial intelligence (AI), fifth-generation mobile communications (5G), and quantum computing are integral to economic competitiveness and are also the "very technologies that ensure we will be able to fight and win the wars of the future," according to the U.S. National Defense Strategy. Today, traditional American leadership is contested, as China emerges as a powerhouse in science and technology, with aspirations to become a global leader in innovation.

Knowledge bases (KB), both automatically and manually constructed, are often incomplete --- many valid facts can be inferred from the KB by synthesizing existing information. A popular approach to KB completion is to infer new relations by combinatory reasoning over the information found along other paths connecting a pair of entities. Given the enormous size of KBs and the exponential number of paths, previous path-based models have considered only the problem of predicting a missing relation given two entities or evaluating the truth of a proposed triple. Additionally, these methods have traditionally used random paths between fixed entity pairs or more recently learned to pick paths between them. We propose a new algorithm MINERVA, which addresses the much more difficult and practical task of answering questions where the relation is known, but only one entity. Since random walks are impractical in a setting with combinatorially many destinations from a start node, we present a neural reinforcement learning approach which learns how to navigate the graph conditioned on the input query to find predictive paths. Empirically, this approach obtains state-of-the-art results on several datasets, significantly outperforming prior methods.