Back-Propagation is the most common algorithm for training neural networks. Here's how to implement it in C#. Back-propagation is the most common algorithm used to train neural networks. There are many ways that back-propagation can be implemented. This article presents a code implementation, using C#, which closely mirrors the terminology and explanation of back-propagation given in the Wikipedia entry on the topic.

Machine Learning (ML) services are quickly becoming a taken-for-granted part of the software developer's toolbox, in any domain. These days, databases or networking are a standard component of almost any non-trivial application, so easily integrated that almost no special expertise is required. We expect to see Machine Learning becoming, in the very near future, a similar layer in the software stack. This commoditization of ML services has been driven so far by Service-oriented platforms such as BigML, which have provided a key ingredient of the process: abstraction. Simple and easy to use REST APIs hide away not only the details of the sophisticated algorithms underlying the services at hand, but also the complexities of scaling those computations both over CPU cycles and input data volumes.

Google has developed its own accelerator chips for artificial intelligence it calls tensor processing units (TPUs) after the open source TensorFlow algorithms it released last year. The news was the big surprise saved for the end of a two-hour keynote at the search giant s annual Google IO event in the heart of Silicon Valley. We have started building tensor processing units TPUs are an order of magnitude higher performance per Watt than commercial FPGAs and GPUs, they powered the AlphaGo system, said Sundar Pichai, Google s chief executive, citing the Google computer that beat a human Go champion. The accelerators have been running in Google s data centers for more than a year, according to a blog by Norm Jouppi, a distinguished hardware engineer at Google. TPUs already power many applications at Google, including RankBrain, used to improve the relevancy of search results and Street View, to improve the accuracy and quality of our maps and navigation, he said.

When he introduced the TPU at the I/O conference, Google CEO Sundar Pichai said it provides an order of magnitude better performance per watt than existing chips for machine learning tasks. While it's unlikely to usurp CPUs and GPUs already in use in the machine learning world, the TPU could potentially speed the machine learning process without using much more energy. Google has been carefully guarding the details of the TPU project, but it's been generally understood that the project was in progress. Based on the company's job postings, it had become evident over the past year that Google was working on a chip of some kind. Now, Google confirms the chip has been under development for about two years.

Then there are'quantum machine learning algorithms,' developed over the last decade following a breakthrough by Harrow, Hassidim, and Lloyd, which do address problems like clustering, classification, support-vector machines, etc. But these algorithms typically require a bunch of conditions to work: for example, that the data are well-conditioned; that they can be accessed in quantum superposition (for example, using a "quantum RAM") or else computed on the fly; and that the properties of the data one cares about can actually be estimated by measuring the resulting quantum states. And we don't yet know how often those conditions will hold in practical applications---and equally important, in the cases where they do hold, we don't have strong evidence that there couldn't be classical random sampling algorithms with similar performance to the quantum algorithms.

When all you have is a hammer, everything looks like a nail. When neural networks are hot in the field of machine learning, everything looks like a pattern-matching problem. When you leave the milk out, a robot will tell you. Watch-Bot is basically your mom, if your mom was a Kinect sensor programmed with unsupervised learning algorithms. It's trained on data from videos of regular people doing regular people things.

Life would be pretty dull without imagination. In fact, maybe the biggest problem for computers is that they don't have any. That's the belief motivating the founders of Vicarious, an enigmatic AI company backed by some of the most famous and successful names in Silicon Valley. Vicarious is developing a new way of processing data, inspired by the way information seems to flow through the brain. The company's leaders say this gives computers something akin to imagination, which they hope will help make the machines a lot smarter.

In a recent Scientific American article entitled "Springtime for AI: The Rise of Deep Learning," computer scientist Yoshua Bengio explains why complex neural networks are the key to true artificial intelligence as people have long envisioned it. It seems logical that the way to make computers as smart as humans is to program them to behave like human brains. However, given how little we know of how the brain functions, this task seems more than a little daunting. So how does deep learning work? This visualization by Jen Christiansen explains the basic structure and function of neural networks.

Salesforce spent roughly 75 million on the three startups it acquired last quarter, according to a regulatory filing submitted Friday. Salesforce spent 32.8 million on MetaMind, the natural language processing and image recognition startup it bought in April. It spent an additional 41.6 million on two other companies it didn't name, but they are likely PredictionIO and Implisit Insights. This is the first time Salesforce publicly disclosed the financial details of these deals. One interesting nugget from the MetaMind acquisition is that Salesforce recorded 31.2 million in goodwill, which represents the amount paid for the company beyond what's valued on the balance sheet.

ICRA is almost over, and we hope you've been enjoying our coverage, which so far has featured robot moths, zipper actuators, machine learning, and duckies. We'll have lots more from the converence over the next few weeks, but for you impatient types, we're cramming Video Friday this week with a painstakingly curated selection of ICRA videos--emphasis on pain: there were nearly 500 videos! We tried to include videos from many different areas of robotics: control, sensing, humanoids, actuators, exoskeletons, manipulators, prosthetics, aerial vehicles, grasping, AI, VR, haptics, vision, and microrobots. We're posting the abstracts along with the videos, but if you have any questions about these projects, let us know and we'll get more details from the authors. Have a great weekend everyone! We present an adaptive filter model of cerebellar function applied to the calibration of a tactile sensory map to improve the accuracy of directed movements of a robotic manipulator.