Summary: In this last article in our series on recommenders we look to the future to see how the rapidly emerging capabilities of Deep Learning can be used to enhance recommender performance. In our first article, "Understanding and Selecting Recommenders" we talked about the broader business considerations and issues for recommenders as a group. In our second article, "5 Types of Recommenders" we attempted to detail the most dominant styles of Recommenders. Our third article, "Recommenders: Packaged Solutions or Home Grown" focused on how to acquire different types of recommenders and how those sources differ. In this last article in our series on recommenders we look to the future to see how the rapidly emerging capabilities of Deep Learning can be used to enhance performance.

Every so often a new technology buzzword appears, to be picked up and repeated ad infinitum in presentations, pitches and articles just like this. From big data to blockchain, they are convenient selling tools, a necessary shorthand; but all too often our knowledge of what they actually refer to is only skin-deep. And there are few terms more mysterious to the uninitiated as deep learning. The problem is that to use these technologies effectively, or even develop a strategy around them, we need to fully understand their nature and their capabilities before we start. The deep learning market is predicted to grow rapidly in the next few years to reach $1.7 billion by 2022, fuelled by growing usage across a wide range of industries.

Inside a brightly decorated lab at the University of California, Berkeley, an ordinary-looking robot has developed an exceptional knack for picking up awkward and unusual objects. What's stunning, though, is that the robot got so good at grasping by working with virtual objects. The robot learned what kind of grip should work for different items by studying a vast data set of 3-D shapes and suitable grasps. The UC Berkeley researchers fed images to a large deep-learning neural network connected to an off-the-shelf 3-D sensor and a standard robot arm. When a new object is placed in front of it, the robot's deep-learning system quickly figures out what grasp the arm should use.

Boston-based Partners HealthCare on Wednesday said it plans to integrate deep learning technology from GE Healthcare across its network. The 10-year collaboration will involve Massachusetts General Hospital and Brigham and Women's Hospital Center for Clinical Data Science. The initiative will feature co-located, multidisciplinary teams with broad access to data, computational infrastructure and clinical expertise. The initial focus will be on the development of applications aimed at improving clinician productivity and patient outcomes in diagnostic imaging. Eventually, the groups will create new business models for applying AI to healthcare, and they will develop products for additional medical specialties, such as molecular pathology, genomics and population health.

Late last year, famed physicist Stephen Hawking issued a warning that the continued advancement of artificial intelligence will either be "the best, or the worst thing, ever to happen to humanity". We've all seen the Terminator movies, and the apocalyptic nightmare that the self-aware AI system, Skynet, wrought upon humanity, and now results from recent behaviour tests of Google's new DeepMind AI system are making it clear just how careful we need to be when building the robots of the future. In tests late last year, Google's DeepMind AI system demonstrated an ability to learn independently from its own memory, and beat the world's best Go players at their own game. It's since been figuring out how to seamlessly mimic a human voice. Now, researchers have been testing its willingness to cooperate with others, and have revealed that when DeepMind feels like it's about to lose, it opts for "highly aggressive" strategies to ensure that it comes out on top. The Google team ran 40 million turns of a simple'fruit gathering' computer game that asks two DeepMind'agents' to compete against each other to gather as many virtual apples as they could.

"An important shift from a mobile first world to an AI first world," declared Google CEO Sundar Pichai, summarizing the Google I/O 2017 keynote yesterday. His description of the changes underway at his company apply to nearly every business today. Almost all of Google's announcements touched on AI in one way or another. From introducing a second generation of TPU chips to accelerate deep learning for such applications as cancer research and DNA sequencing, to a broad effort to get Google Home on as many screens and devices as possible. The company also shared that it's speech recognition technology was now better than 95 percent accurate.

We empirically characterize the performance of discriminative and generative LSTM models for text classification. We find that although RNN-based generative models are more powerful than their bag-of-words ancestors (e.g., they account for conditional dependencies across words in a document), they have higher asymptotic error rates than discriminatively trained RNN models. However we also find that generative models approach their asymptotic error rate more rapidly than their discriminative counterparts---the same pattern that Ng & Jordan (2001) proved holds for linear classification models that make more naive conditional independence assumptions. Building on this finding, we hypothesize that RNN-based generative classification models will be more robust to shifts in the data distribution. This hypothesis is confirmed in a series of experiments in zero-shot and continual learning settings that show that generative models substantially outperform discriminative models.

Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function. Recent work has largely focused on minimizing the mean squared reconstruction error. The resulting estimates have high peak signal-to-noise ratios, but they are often lacking high-frequency details and are perceptually unsatisfying in the sense that they fail to match the fidelity expected at the higher resolution. In this paper, we present SRGAN, a generative adversarial network (GAN) for image super-resolution (SR). To our knowledge, it is the first framework capable of inferring photo-realistic natural images for 4x upscaling factors. To achieve this, we propose a perceptual loss function which consists of an adversarial loss and a content loss. The adversarial loss pushes our solution to the natural image manifold using a discriminator network that is trained to differentiate between the super-resolved images and original photo-realistic images. In addition, we use a content loss motivated by perceptual similarity instead of similarity in pixel space. Our deep residual network is able to recover photo-realistic textures from heavily downsampled images on public benchmarks. An extensive mean-opinion-score (MOS) test shows hugely significant gains in perceptual quality using SRGAN. The MOS scores obtained with SRGAN are closer to those of the original high-resolution images than to those obtained with any state-of-the-art method.

Despite billions being spent on research, even our best deep learning neural networks look pitiful when compared to the intricate design of the brain of a bumble bee or even an ant, Peter Bentley tells John Thornhill.

We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 different classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0%, respectively, which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implementation of the convolution operation. To reduce overfitting in the fully connected layers we employed a recently developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry. Four years ago, a paper by Yann LeCun and his collaborators was rejected by the leading computer vision conference on the grounds that it used neural networks and therefore provided no insight into how to design a vision system. At the time, most computer vision researchers believed that a vision system needed to be carefully hand-designed using a detailed understanding of the nature of the task. They assumed that the task of classifying objects in natural images would never be solved by simply presenting examples of images and the names of the objects they contained to a neural network that acquired all of its knowledge from this training data. What many in the vision research community failed to appreciate was that methods that require careful hand-engineering by a programmer who understands the domain do not scale as well as methods that replace the programmer with a powerful general-purpose learning procedure.