Researchers at UC Berkeley have developed a robot that can pick up awkward and unusually shaped objects. The robot learned how to grasp different objects by studying a virtual library of 10,000 3D objects and suitable grasps. When a new object is placed in front of the bot, its deep-learning system quickly figures out what grasp the arm should use. When the robot was unsure of how to grasp an object, it poked it to figure out how to better grasp it. Deep-learning software tries to mimic the activity in layers of neurons in the neocortex, which makes up 80 percent of the brain and is where thinking occurs.

The loss surface of deep neural networks has recently attracted interest in the optimization and machine learning communities as a prime example of high-dimensional non-convex problem. Some insights were recently gained using spin glass models and mean-field approximations, but at the expense of strongly simplifying the nonlinear nature of the model. In this work, we do not make any such assumption and study conditions on the data distribution and model architecture that prevent the existence of bad local minima. Our theoretical work quantifies and formalizes two important \emph{folklore} facts: (i) the landscape of deep linear networks has a radically different topology from that of deep half-rectified ones, and (ii) that the energy landscape in the non-linear case is fundamentally controlled by the interplay between the smoothness of the data distribution and model over-parametrization. Our main theoretical contribution is to prove that half-rectified single layer networks are asymptotically connected, and we provide explicit bounds that reveal the aforementioned interplay. The conditioning of gradient descent is the next challenge we address. We study this question through the geometry of the level sets, and we introduce an algorithm to efficiently estimate the regularity of such sets on large-scale networks. Our empirical results show that these level sets remain connected throughout all the learning phase, suggesting a near convex behavior, but they become exponentially more curvy as the energy level decays, in accordance to what is observed in practice with very low curvature attractors.

End-to-end training of automated speech recognition (ASR) systems requires massive data and compute resources. We explore transfer learning based on model adaptation as an approach for training ASR models under constrained GPU memory, throughput and training data. We conduct several systematic experiments adapting a Wav2Letter convolutional neural network originally trained for English ASR to the German language. We show that this technique allows faster training on consumer-grade resources while requiring less training data in order to achieve the same accuracy, thereby lowering the cost of training ASR models in other languages. Model introspection revealed that small adaptations to the network's weights were sufficient for good performance, especially for inner layers.

Summary: Quantum computing is now a commercial reality. Here's the story of the companies that are currently using it in operations and how this will soon disrupt artificial intelligence and deep learning. Like a magician distracting us with one hand while pulling a fast one with the other Quantum computing has crossed over from research to commercialization almost without us noticing. Has the dream of Quantum computing actually stepped out of the lab into the world of actual application. Well, Lockheed Martin has been using it for seven years.

Mix, Mingle and Learn from IBM featured speakers on such topics as IBM's Data Science Experience, Scalable TensorFlow Deep Learning as a Service with Docker and OpenPOWER with GPUs, and more. I'll demonstrate how to train a TensorFlow model and how IBM Power Systems with OpenPOWER architecture make TensorFlow models even more powerful. Improving Data Scientist Productivity with Data Science Experience - Patrick Pitre Data Science is often hampered by the inability of data scientists to collaborate on a shared code base. In this demonstration, I will discuss the use of composable data services and a collaborative development space to increase the speed to market of analytics using IBM's Data Science Experience and IBM Bluemix.

Nobody likes it, but we all have to deal with it. As the world's cities grow more densely populated, scientists and entrepreneurs are looking for solutions to gridlock, pollution and the other byproducts of a world filled with cars. Two sessions at the GPU Technology Conference earlier this month spoke to the role that data, deep learning and intelligent video analytics can play in easing traffic and improving quality of life for city dwellers the world over. Kurtis McBride, CEO of Miovision Technologies, an IVA startup based in Ontario, Canada, spoke to a room full of developers about his company's efforts -- and their 40 percent year-over-year growth -- to make traffic flow a little easier. Miovision's Open City platform gets data from existing city infrastructure and the company's own video cameras, and applies AI to create insights from it.

Analyzing images and videos, and using them in various applications such as self driven cars, drones etc. with underlying deep learning techniques has been the new research frontier. The recent research papers such as "A Neural Algorithm of Artistic Style", show how a styles can be transferred from an artist and applied to an image, to create a new image. Other papers such as "Generative Adversarial Networks" (GAN) and "Wasserstein GAN" have paved the path to develop models that can learn to create data that is similar to data that we give them. While these research areas are still on the generic images, our goal is to use these research into medical images to help healthcare. We need to start with some basics.

Financial services jobs go in and out of fashion. In 2001 equity research for internet companies was all the rage. In 2006, structuring collateralised debt obligations (CDOs) was the thing. In 2010, credit traders were popular. In 2014, compliance professionals were it.

Microchip manufacturer Intel has invested heavily in the driverless car race with the latest US$15 billion (A$19.5bn) Mobileye develops sensors and intelligence technology behind automated driver-assistance systems and many self-driving cars. Its tech enables a car to "see" and understand the world. Other recent purchases include the deep learning tech company Nervana, microchip maker Movidius and automotive tech company Delphi. Intel is also working with the automotive companies BMW and Volkswagen to begin trials later this year. Intel is strategically putting together all the critical capabilities required to develop self-driving cars that can "see" and intelligently understand the world around us.

Of course, the limitations of this is that it is a best efforts prediction by expert practitioners. As the case of solar energy predictions shows, insiders are among the most conservative in their own field. Google DeepMind has bested human players this year. It did so using a tensor processing unit, which uses one tenth of the power as the GPU set up last year. So undeniably progress is fast. AlphaGo also paired up with human Go players in an interesting example of intelligence augmentation.