Spy agencies are to use computer algorithms to keep tabs on 20,000 former terror suspects because they don't have the manpower to physically watch all of them at once. MI5 and MI6 will use the complex system to sift through massive amounts of data to alert intelligence agents to worrying behaviour, under potential plans. A review is expected to recommend security agencies widen their searches when examining the online footprint and movement of these former jihadis. Britain's counter-terrorism officials currently actively monitor around 3,000 people. But following a wave of terror attacks in the UK, the Government admitted there were as many as 20,000 former subjects of interest no longer classed as posing serious danger.

The most dangerous part of any car, say the experts, 'is the nut behind the steering wheel'. Human error is to blame for most accidents, so remove that'nut' and let the car drive itself and many lives will be saved, runs the argument now pushed by ministers, manufacturers and supporters of what is known as'autonomous driving'. And it certainly seems as if it's full speed ahead for the driverless car. The Prime Minister Theresa May and Chancellor Philip Hammond yesterday confirmed plans -- widely trailed ahead of the Budget tomorrow -- to invest £900 million to deliver'fully driverless cars' by 2021. But is the Government right to be putting its foot on the accelerator?

Detecting art forgeries is hard and expensive. Art historians might bring a suspect work into a lab for infrared spectroscopy, radiometric dating, gas chromatography, or a combination of such tests. AI, it turns out, doesn't need all that: it can spot a fake just by looking at the strokes used to compose a piece. In a new paper, researchers from Rutgers University and the Atelier for Restoration & Research of Paintings in the Netherlands document how their system broke down almost 300 line drawings by Picasso, Matisse, Modigliani, and other famous artists into 80,000 individual strokes. Then a deep recurrent neural network (RNN) learned what features in the strokes were important to identify the artist.

As part of my experience, I have had the luxury of observing the way possibilities regarding AI have drastically arisen during the last decade. What many considered unachievable in the past is now certainly achievable and may be out in full flow during the coming future. While there is a lot to talk about, we can all expect the future to be really exciting. As part of the limitless possibilities of Artificial Intelligence, which is popularly known as AI, I had the experience of attending Adobe's Symposium in Amsterdam. While the symposium itself was a success and helped highlight the role of AI in the industry, I was pleased to have an opportunity to host VP of Adobe John Mellor right after the ceremony ended.

There is an extensive literature on the complexity of planning, but explicit bounds on time and space complexity are very rare. On the other hand, problems like the constraint satisfaction problem (CSP) have been thoroughly analysed in this respect. We provide a number of upper- and lower-bound results (the latter based on various complexity-theoretic assumptions such as the Exponential Time Hypothesis) for both satisficing and optimal planning. We show that many classes of planning instances exhibit a dichotomy: either they can be solved in polynomial time or they cannot be solved in subexponential time. In many cases, we can even prove closely matching upper and lower bounds. Our results also indicate, analogously to CSPs, the existence of sharp phase transitions. We finally study and discuss the trade-off between time and space. In particular, we show that depth-first search may sometimes be a viable option for planning under severe space constraints.

A major challenge in computational chemistry is the generation of novel molecular structures with desirable pharmacological and physiochemical properties. In this work, we investigate the potential use of autoencoder, a deep learning methodology, for de novo molecular design. Various generative autoencoders were used to map molecule structures into a continuous latent space and vice versa and their performance as structure generator was assessed. Our results show that the latent space preserves chemical similarity principle and thus can be used for the generation of analogue structures. Furthermore, the latent space created by autoencoders were searched systematically to generate novel compounds with predicted activity against dopamine receptor type 2 and compounds similar to known active compounds not included in the trainings set were identified.

This paper demonstrates the use of genetic algorithms for evolving: 1) a grandmaster-level evaluation function, and 2) a search mechanism for a chess program, the parameter values of which are initialized randomly. The evaluation function of the program is evolved by learning from databases of (human) grandmaster games. At first, the organisms are evolved to mimic the behavior of human grandmasters, and then these organisms are further improved upon by means of coevolution. The search mechanism is evolved by learning from tactical test suites. Our results show that the evolved program outperforms a two-time world computer chess champion and is at par with the other leading computer chess programs.

We propose a novel approach for the generation of polyphonic music based on LSTMs. We generate music in two steps. First, a chord LSTM predicts a chord progression based on a chord embedding. A second LSTM then generates polyphonic music from the predicted chord progression. The generated music sounds pleasing and harmonic, with only few dissonant notes. It has clear long-term structure that is similar to what a musician would play during a jam session. We show that our approach is sensible from a music theory perspective by evaluating the learned chord embeddings. Surprisingly, our simple model managed to extract the circle of fifths, an important tool in music theory, from the dataset.

Both scientists and children make important structural discoveries, yet their computational underpinnings are not well understood. Structure discovery has previously been formalized as probabilistic inference about the right structural form --- where form could be a tree, ring, chain, grid, etc. [Kemp & Tenenbaum (2008). The discovery of structural form. PNAS, 105(3), 10687-10692]. While this approach can learn intuitive organizations, including a tree for animals and a ring for the color circle, it assumes a strong inductive bias that considers only these particular forms, and each form is explicitly provided as initial knowledge. Here we introduce a new computational model of how organizing structure can be discovered, utilizing a broad hypothesis space with a preference for sparse connectivity. Given that the inductive bias is more general, the model's initial knowledge shows little qualitative resemblance to some of the discoveries it supports. As a consequence, the model can also learn complex structures for domains that lack intuitive description, as well as predict human property induction judgments without explicit structural forms. By allowing form to emerge from sparsity, our approach clarifies how both the richness and flexibility of human conceptual organization can coexist.

Natural gradient descent is an optimization method traditionally motivated from the perspective of information geometry, and works well for many applications as an alternative to stochastic gradient descent. In this paper we critically analyze this method and its properties, and show how it can be viewed as a type of approximate 2nd-order optimization method, where the Fisher information matrix can be viewed as an approximation of the Hessian. This perspective turns out to have significant implications for how to design a practical and robust version of the method. Additionally, we make the following contributions to the understanding of natural gradient and 2nd-order methods: a thorough analysis of the convergence speed of stochastic natural gradient descent (and more general stochastic 2nd-order methods) as applied to convex quadratics, a critical examination of the oft-used "empirical" approximation of the Fisher matrix, and an analysis of the (approximate) parameterization invariance property possessed by natural gradient methods, which we show still holds for certain choices of the curvature matrix other than the Fisher, but notably not the Hessian.