Artificial intelligence (AI) has increasingly become one of the hottest topics in both business and science. More leading tech companies are showing their interest in AI investment, from Google's $400 million acquisition of DeepMind and Faraday Future's unveiling of self-driving supercars at CES 2017. These are just a few examples of the commitment companies have towards this cutting-edge technology, but one of the most promising areas for AI is in mobile. The idea of having a personal assistant to help tackle everyday tasks is becoming more appealing to users everywhere. However, intelligent apps are not just limited to digital assistants but for a variety of purposes from security to e-commerce.

An ongoing challenge in neuromorphic computing is to devise general and computationally efficient models of inference and learning which are compatible with the spatial and temporal constraints of the brain. One increasingly popular and successful approach is to take inspiration from inference and learning algorithms used in deep neural networks. However, the workhorse of deep learning, the gradient descent Back Propagation (BP) rule, often relies on the immediate availability of network-wide information stored with high-precision memory, and precise operations that are difficult to realize in neuromorphic hardware. Remarkably, recent work showed that exact backpropagated weights are not essential for learning deep representations. Random BP replaces feedback weights with random ones and encourages the network to adjust its feed-forward weights to learn pseudo-inverses of the (random) feedback weights. Building on these results, we demonstrate an event-driven random BP (eRBP) rule that uses an error-modulated synaptic plasticity for learning deep representations in neuromorphic computing hardware. The rule requires only one addition and two comparisons for each synaptic weight using a two-compartment leaky Integrate & Fire (I&F) neuron, making it very suitable for implementation in digital or mixed-signal neuromorphic hardware. Our results show that using eRBP, deep representations are rapidly learned, achieving nearly identical classification accuracies compared to artificial neural network simulations on GPUs, while being robust to neural and synaptic state quantizations during learning.

In many applications of black-box optimization, one can evaluate multiple points simultaneously, e.g. when evaluating the performances of several different neural network architectures in a parallel computing environment. In this paper, we develop a novel batch Bayesian optimization algorithm --- the parallel knowledge gradient method. By construction, this method provides the one-step Bayes optimal batch of points to sample. We provide an efficient strategy for computing this Bayes-optimal batch of points, and we demonstrate that the parallel knowledge gradient method finds global optima significantly faster than previous batch Bayesian optimization algorithms on both synthetic test functions and when tuning hyperparameters of practical machine learning algorithms, especially when function evaluations are noisy.

Mountain View, California-based Drive.ai is a startup created by former lab mates from Stanford University's Artificial Intelligence Lab. Originally founded in 2015 by Carol Reiley and Fred Rosenzweig, Drive.ai raised $12 million in Series A funding earlier this year to develop deep learning algorithms to control the operation of autonomous vehicles. Building on experience gained from the DARPA Grand Challenge, Google and other self-driving pioneers programmed the first self-driving car to rely primarily on light detection and ranging (LIDAR), which is a remote sensing method that uses pulses of laser light to measure distances, and detailed mapping. Although this has worked pretty well, the current technology is expensive. Making autonomous vehicles easier to manufacture with less expensive parts will make them more affordable.

We are excited to welcome back Creative AI: Applications of AI in Art, Music, Film, Design Meetup, who are running a session looking at artists and machine learning. This meetup will focus on the intersection between machine learning and artistic practice. Recent developments in deep learning have provided artists with new possibilities of using image, sound and text data for creative expression. The speakers will cover recent art projects in the field, discuss the potential of machine learning in art and suggest online resources for getting started yourself. Please join the Creative AI Meetup group to register for this event.

The first FDA approval for a machine learning application to be used in a clinical setting is a big step forward for AI and machine learning in healthcare and industry as a whole. Arterys's medical imaging platform has been approved to be put into use to help doctors diagnose heart problems. It uses a self-teaching artificial neural network which has learned from 1,000 cases so far, and will continue to improve its knowledge and understanding of how the heart works with each new case it examines. In order to be approved by the US Food and Drug Administration (FDA), it had to pass tests to show it can produce results at least as accurately as humans are currently able to. The key difference though is that Arterys takes an average of 15 seconds to produce a result for one case, which a professional human analyst would expect to spend between 30 minutes to an hour working on. Arterys was founded by Fabien Beckers, John Axerio-Cilies, Albert Hsiao and Shreyas Vasanawala when they met at Stanford University with a shared passion for the transformative potential of machine learning.

Let me also mention some of the advances in my main area of expertise: Recommender Systems. Of course Deep Learning has also impacted this area. While I would still not recommend DL as the default approach to recommender systems, it is interesting to see how it is already being used in practice, and in large scale, by products like Youtube. That said, there has been interesting research in the area that is not related to Deep Learning. The best paper award in this year's ACM Recsys went to "Local Item-Item Models For Top-N Recommendation", an interesting extension to Sparse Linear Methods (i.e.

In recent months, machine learning has become somewhat of a buzzword, with major players such as Google highlighting the positive impact it can have for businesses. Defined as a subset of artificial intelligence, machine learning focuses on the development of computer programmes that can teach themselves, and grow and change in response to new data. With 90% of the world's data having been created in the last two years alone, the ability to develop automated processes to efficiently adapt to new information is invaluable. For mobile operators, machine learning has the potential to drive huge benefits, in particular when it comes to tackling fraud. In 2016 alone, telcos are expected to face global losses of $294 billion, making it vital that they look to utilise all tools at their disposal to combat such a pressing issue, says Raul Gomes Azevedo, director Product Development, WeDo Technologies. For starters, fraud management involves identifying specific profiles and behaviour, and checking whether everything's running as expected, or if there are any anomalies.

Does your head sometimes spin when a bunch of technical terms are tossed around that you can't interpret using simple, everyday common sense? The terms surrounding the notion of artificial intelligence are certainly one thorny example. People talk about deep learning, machine learning, neural networks and natural language processing. We want to give you a clear sense of the meanings of these different terms, without getting too scientific on you. All technologies involved in generating thinking that only humans could previously achieve, are included under the umbrella of AI. These days AI is simply an overall term we can use when we don't want to go into too much detail.