The US military apparently never tires of thinking up capability gaps, and that means we may soon see fleets of small drones dropping out of bombers, then later being yanked out of the sky by cargo planes. Cartoonish as it may sound--as is the case with so many deadly-serious but still far-out military concepts--it makes a lot of sense. And Darpa, the Pentagon's weapon of choice for making crazy things happen, just chose four companies to push the idea forward. Called Gremlins (because you weren't already freaked out) the project calls for a new type of reusable unmanned aerial vehicle that can be air-launched on intelligence-gathering missions from cargo airplanes, bombers, or other military aircraft over "denied" (i.e., hostile) airspace. Once their missions are complete, up to three hours later, the drones will fly back to retrieval area where a C-130 cargo airplane will collect them.

For both humans and machines, understanding the visual world requires relating new percepts with past experience. We argue that a good visual representation for an image should encode what makes it similar to other images, enabling the recall of associated experiences. Current machine implementations of visual representations can capture some aspects of similarity, but fall far short of human ability overall. Even if one explicitly labels objects in millions of images to tell the computer what should be considered similar--a very expensive procedure--the labels still do not capture everything that might be relevant. This thesis shows that one can often train a representation which captures similarity beyond what is labeled in a given dataset.

Twenty years have passed since the IBM computer Deep Blue defeated world chess champion Garry Kasparov, and we all know computers have improved since then. But Deep Blue won through sheer computing power, using its ability to calculate the outcomes of more moves to a deeper level than even a world champion can. Go is played on a far larger board (19 by 19 squares, compared to 8x8 for chess) and has more possible moves than there are atoms in the universe, so raw computing power was unlikely to beat a human with a strong intuitive sense of the best moves. Instead, AlphaGo was designed to win by playing a huge number of games against other programs and adopting the strategies that proved successful. You could say that AlphaGo evolved to be the best Go player in the world, achieving in only two years what natural selection took millions of years to accomplish.

GE's electronic work instructions with the Google Glass by Novotek: Novotek combined GE's electronic work instructions with the Google Glass wearable computing device, and was demonstrated to Summit attendees in the Technology Fair. Novotek, is the largest European distributor for GE's Intelligent Platforms business and it has been awarded the Scanautomatic Prize for Innovation at Scanautomatic 2014 in Gothenburg, Sweden in October. Novotek has since 1986 worked with integration of IT and automation systems in the process and production industry. Novotek mainly supplies world-leading products and solutions from GE in the Nordics and Benelux. A work process management solution, GE's Proficy Workflow software provides users with interactive, step-by-step task instructions and captures process, traceability and quality data across systems to reduce errors, waste and delays.

Businesses and developers will be able to make their services available inside Messenger by way of Chat SDK. Powered by artificial intelligence, chatbots are computer software programs that mimic human conversations. Facebook says three times as many messages are sent on its platforms than SMS, with 60 billion messages a day sent and received on Facebook Messenger and WhatsApp. The news comes just weeks after Microsoft devoted a large chunk of its Build developer conference keynote to what its executives called "conversations as a platform". AI is already used in Messenger, and can do things like recognise faces in pictures to suggest potential recipients.

Receiver algorithms which combine belief propagation (BP) with the mean field (MF) approximation are well-suited for inference of both continuous and discrete random variables. In wireless scenarios involving detection of multiple signals, the standard construction of the combined BP-MF framework includes the equalization or multi-user detection functions within the MF subgraph. In this paper, we show that the MF approximation is not particularly effective for multi-signal detection. We develop a new factor graph construction for application of the BP-MF framework to problems involving the detection of multiple signals. We then develop a low-complexity variant to the proposed construction in which Gaussian BP is applied to the equalization factors. In this case, the factor graph of the joint probability distribution is divided into three subgraphs: (i) a MF subgraph comprised of the observation factors and channel estimation, (ii) a Gaussian BP subgraph which is applied to multi-signal detection, and (iii) a discrete BP subgraph which is applied to demodulation and decoding. Expectation propagation is used to approximate discrete distributions with a Gaussian distribution and links the discrete BP and Gaussian BP subgraphs. The result is a probabilistic receiver architecture with strong theoretical justification which can be applied to multi-signal detection.

In magnetoencephalography (MEG) the conventional approach to source reconstruction is to solve the underdetermined inverse problem independently over time and space. Here we present how the conventional approach can be extended by regularizing the solution in space and time by a Gaussian process (Gaussian random field) model. Assuming a separable covariance function in space and time, the computational complexity of the proposed model becomes (without any further assumptions or restrictions) $\mathcal{O}(t^3 + n^3 + m^2n)$, where $t$ is the number of time steps, $m$ is the number of sources, and $n$ is the number of sensors. We apply the method to both simulated and empirical data, and demonstrate the efficiency and generality of our Bayesian source reconstruction approach which subsumes various classical approaches in the literature.

The variational autoencoder (VAE) is a generative model with continuous latent variables where a pair of probabilistic encoder (bottom-up) and decoder (top-down) is jointly learned by stochastic gradient variational Bayes. We first elaborate Gaussian VAE, approximating the local covariance matrix of the decoder as an outer product of the principal direction at a position determined by a sample drawn from Gaussian distribution. We show that this model, referred to as VAE-ROC, better captures the data manifold, compared to the standard Gaussian VAE where independent multivariate Gaussian was used to model the decoder. Then we extend the VAE-ROC to handle mixed categorical and continuous data. To this end, we employ Gaussian copula to model the local dependency in mixed categorical and continuous data, leading to {\em Gaussian copula variational autoencoder} (GCVAE). As in VAE-ROC, we use the rank-one approximation for the covariance in the Gaussian copula, to capture the local dependency structure in the mixed data. Experiments on various datasets demonstrate the useful behaviour of VAE-ROC and GCVAE, compared to the standard VAE.

Factor analysis aims to determine latent factors, or traits, which summarize a given data set. Inter-battery factor analysis extends this notion to multiple views of the data. In this paper we show how a nonlinear, nonparametric version of these models can be recovered through the Gaussian process latent variable model. This gives us a flexible formalism for multi-view learning where the latent variables can be used both for exploratory purposes and for learning representations that enable efficient inference for ambiguous estimation tasks. Learning is performed in a Bayesian manner through the formulation of a variational compression scheme which gives a rigorous lower bound on the log likelihood. Our Bayesian framework provides strong regularization during training, allowing the structure of the latent space to be determined efficiently and automatically. We demonstrate this by producing the first (to our knowledge) published results of learning from dozens of views, even when data is scarce.

Variational inference is a powerful tool for approximate inference, and it has been recently applied for representation learning with deep generative models. We develop the variational Gaussian process (VGP), a Bayesian nonparametric variational family, which adapts its shape to match complex posterior distributions. The VGP generates approximate posterior samples by generating latent inputs and warping them through random non-linear mappings; the distribution over random mappings is learned during inference, enabling the transformed outputs to adapt to varying complexity. We prove a universal approximation theorem for the VGP, demonstrating its representative power for learning any model. For inference we present a variational objective inspired by auto-encoders and perform black box inference over a wide class of models. The VGP achieves new state-of-the-art results for unsupervised learning, inferring models such as the deep latent Gaussian model and the recently proposed DRAW.