AI is a technology on the cusp. While nothing actually new in itself, a recent convergence of increased computational power with a mushrooming of large datasets and the refinement of existing understanding of the techniques involved has seen it become an important differentiator in the industry. Following a breakthrough year in 2015 when its use, particularly by Google for image processing and translation, jumped markedly, 2017 has seen it rapidly spread into many different niches. Of all the zeitgeist-friendly products on the showfloor at IBC2017, TVU Networks would have won a buzzword bingo competition with its Smart Captions subtitling software. Since renamed TVU Transcriber, it marries AI routines with voice recognition -- another fecund key technology segment in Amsterdam -- to provide an audio to text transcribing service.

Every useful machine – from advanced AI through to the humble toaster – has to possess'knowledge' about the world in which it operates. For example, within the design of a household toaster, much knowledge about the world must have already been implemented. The toaster'knows' what electrical voltage is available through the design of its electric components. Through its physical dimensions it knows the size of a slice of bread (with some allowance for variation). Within the toaster's regulation mechanisms lies knowledge on the duration of toasting typically needed, and in its switch and knob design it even knows the typical size of people's fingers with which these buttons are being pushed and turned.

The GraphGrail Ai project has begun conducting operations directed at establishing working relations with leading Russian universities. After a series of successful meetings, GraphGrail Ai is entering negotiations with MEPhI (Moscow Engineering Physics Institute), RUSNANO, ROSATOM, SFEDU and other universities. The higher educational institutions are interested in working with the GraphGrail Ai database and platform, as well as offering the scientific work and efforts of their staff to the development of the project. In addition, work on the joint creation of a new platform is being discussed, which will be used to monitor the completion of tasks by students, thus identifying potential future employees for various companies in the scientific field. This endeavor will allow GraphGrail Ai to be incorporated into the social programs of universities and be present in the public domain, providing workplaces for future specialists.

Artificial intelligence technology that can detect Isis videos and prevent them from being uploaded is being released to stop the spread of the "poisonous" material. Developers funded by the Home Office are sharing their software for free with any website or app in the world in the hope it will make the terrorist group's propaganda harder to access and share. Tests suggest it can detect 94 per cent of Isis videos and makes so few mistakes that a single person could moderate borderline cases for the whole of YouTube. Dr Marc Warner, chief executive of ASI Data Science, told The Independent the technology's success depends on how many companies build it into their systems but "we hope that this can play its part in removing extremist content from the web". "Lone-wolf attacks are hard to spot with conventional surveillance – it is a difficult problem if someone is radicalised in their bedroom," he added.

The UK Government Digital Service (GDS) will work with government departments over the next 12 months to support the use of biometrics and artificial intelligence for services, according to a blog post by GDS head Kevin Cunnington. He suggests in the post that government services such as GOV.UK can make use of "innovative technologies like machine learning and voice control," and says GDS will work with government departments and public bodies to identify areas where digital technologies can solve challenges. GDS will also act as a "front door" for tech firms, to connect providers of such technologies with government entities that can benefit from them. "Once a finalised product is created, the public sector body can then choose to buy it from the tech company," Cunnington writes in the post. GDS is also considering developing courses in AI, geospatial data, and distributed ledgers for government workers from its GDS Academy, according to the post.

Training Gaussian process-based models typically involves an $ O(N^3)$ computational bottleneck. Popular methods for overcoming this matrix inversion problem cannot adequately model all types of latent functions, and are often not parallelizable. We present an embarrassingly parallel method that takes advantage of inverting block diagonal approximations, while maintaining much of the expressivity of a full covariance matrix. By using importance sampling to average over different realizations of low-rank GP approximations, we ensure our algorithm is both asymptotically unbiased and embarrassingly parallel. We show comparable or improved performance over competing methods, on a range of synthetic and real datasets.

Modern scientific instruments produce vast amounts of data, which can overwhelm the processing ability of computer systems. Lossy compression of data is an intriguing solution but comes with its own dangers, such as potential signal loss, and the need for careful parameter optimization. In this work, we focus on a setting where this problem is especially acute compressive sensing frameworks for radio astronomy and ask: Can the precision of the data representation be lowered for all input data, with recovery guarantees and good practical performance? Our first contribution is a theoretical analysis of the Iterative Hard Thresholding (IHT) algorithm when all input data, that is, the measurement matrix and the observation, are quantized aggressively, to as little as 2 bits per value. Under reasonable constraints, we show that there exists a variant of low precision IHT which can still provide recovery guarantees. The second contribution is a tailored analysis of our general quantized framework to radio astronomy, showing that its conditions are satisfied in this case. We evaluate our approach using an FPGA implementation, and show that it can achieve up to 9.19x speed up with negligible loss of recovery quality, on real telescope data

We provide a comprehensive overview and tooling for GP modeling with non-Gaussian likelihoods using state space methods. The state space formulation allows for solving one-dimensional GP models in $\mathcal{O}(n)$ time and memory complexity. While existing literature has focused on the connection between GP regression and state space methods, the computational primitives allowing for inference using general likelihoods in combination with the Laplace approximation (LA), variational Bayes (VB), and assumed density filtering (ADF) / expectation propagation (EP) schemes has been largely overlooked. We present means of combining the efficient $\mathcal{O}(n)$ state space methodology with existing inference methods. We also further extend existing methods, and provide unifying code implementing all approaches.

In recent years, the importance of deep learning has significantly increased in pattern recognition, computer vision, and artificial intelligence research, as well as in industry. However, despite the existence of multiple deep learning frameworks, there is a lack of comprehensible and easy-to-use high-level tools for the design, training, and testing of deep neural networks (DNNs). In this paper, we introduce Barista, an open-source graphical high-level interface for the Caffe deep learning framework. While Caffe is one of the most popular frameworks for training DNNs, editing prototext files in order to specify the net architecture and hyper parameters can become a cumbersome and error-prone task. Instead, Barista offers a fully graphical user interface with a graph-based net topology editor and provides an end-to-end training facility for DNNs, which allows researchers to focus on solving their problems without having to write code, edit text files, or manually parse logged data.

GANs excel at learning high dimensional distributions, but they can update generator parameters in directions that do not correspond to the steepest descent direction of the objective. Prominent examples of problematic update directions include those used in both Goodfellow's original GAN and the WGAN-GP. To formally describe an optimal update direction, we introduce a theoretical framework which allows the derivation of requirements on both the divergence and corresponding method for determining an update direction. These requirements guarantee unbiased mini-batch updates in the direction of steepest descent. We propose a novel divergence which approximates the Wasserstein distance while regularizing the critic's first order information. Together with an accompanying update direction, this divergence fulfills the requirements for unbiased steepest descent updates. We verify our method, the First Order GAN, with CelebA image generation and set a new state of the art on the One Billion Word language generation task. Code to reproduce experiments is available https://github.com/