In an earlier post I explored the value of using scalable machine learning to extract value from huge amounts of data. In this post, I will dive down into the technical side of things, particularly the challenges and benefits that come with making algorithms scalable on large clusters of computers. Machine learning algorithms are written to run on single-node systems, or on specialized supercomputer hardware, which I'll refer to as HPC boxes. They grew up in a world where they didn't have to scale across multiple nodes. It's relatively easy to get high performance when running algorithms on a single computer.

Probabilistic programming languages are in the spotlight. This is due to the announcement of a new DARPA program to support their fundamental research. But what is probabilistic programming? What can we expect from this research? Will this effort pay off?

DeepMind, Google's artificial intelligence (AI) division, certainly thinks there's a risk. They've teamed up with Oxford University to develop a "red button" that would interrupt an AI machine's actions. Their paper "explores a way to make sure a learning agent will not learn to prevent (or seek!) being interrupted by the environment or a human operator."

NATIONAL HARBOR, MARYLAND – A new maker of self-driving vehicles burst onto the scene Thursday in partnership with IBM's supercomputer platform Watson, and it is ready to roll right now. Arizona-based startup Local Motors is offering Olli, an electric minibus capable of carrying 12 people. It says the vehicle can be produced to order by 3-D printing. Olli was designed as an on-demand transportation solution that passengers can summon with a mobile app, like Uber rides. And it can be "printed" to specification in "micro factories" in a matter of hours. Olli will be demonstrated in National Harbor, Maryland, over the next few months with additional trials expected in Las Vegas and Miami.

Machine Learning has definitely gone mainstream with almost all the major vendors announcing support for Machine Learning platforms, frameworks, libraries and how they have started to use it in their applications. This blog post highlights the key Machine Learning process and the spectrum of Machine Learning platforms that are available today for developers to get started. The first thing that developers need to understand that Machine Learning techniques are very different from the traditional programming constructs that they use, while developing their applications. It is often remarked in a lighter vein that some vendors pass of multiple if, then, else statements are machine learning in their applications. The diagram shown below depicts a typical Machine Learning process. The key point to take away from the above process is that the individual iterations around Data Preparation, Model building will keep on happening and we are never finished.

GraphX is a powerful graph processing API for the Apache Spark analytics engine that lets you draw insights from large datasets. GraphX gives you unprecedented speed and capacity for running massively parallel and machine learning algorithms.

The defining advance of the next decade, if you listen to the prophets of Silicon Valley, will be the seismic and unavoidable ascent of artificial intelligence. It might be hard to take the thought seriously when a satnav sends you down a dead-end country road, or your phone's autocorrect feature turns a carefully-constructed text message to gibberish, but the milestones reached in the past year alone have been exceptional. DeepMind, the British AI company owned by Google, has defeated the world champion at Go, the ancient game that requires a finely tuned sense of intuition to master. Driverless cars now seem like an inevitability rather than a curiosity. Error rates on image recognition technology have dropped from 25pc in 2011 to less than 4pc.

That's the advice you might give a student struggling for a better grip on the ins and outs of language. On 18 February, Facebook released several data sets that it uses to train its home-grown neural networks. One is packed with the text of classic novels, such as The Jungle Book, Peter Pan, Little Women, A Christmas Carol and Alice in Wonderland. There are more than a hundred stories in all, taken from the free online library Project Gutenberg. It's a reading list for fledgling AI (see box below).

Google is expected to dive deeper into virtual reality and artificial intelligence Wednesday during an annual conference that serves as a launching pad for its latest products and innovations. Although Google keeps its plans under wraps until the big event, the conference agenda makes it clear that virtual reality and artificial intelligence, or "machine learning," will be among the focal points. That has spurred speculation that Google is getting ready to release a virtual-reality device to compete with Facebook's new Oculus Rift headset, as well as the Samsung's Gear VR and the Vive from HTC and Valve. Reporters and bloggers from around the world will attend, ensuring that whatever the company unveils will also be featured in stories, pictures and video delivered to a vast audience of consumers. The three-day showcase also attracts thousands of computer programmers, giving Google an opportunity to convince them why they should design applications and other services that work with its gadgets and an array of software that includes the Chrome Web browser and Android operating system for mobile devices.

Researchers from the Kuhl Lab at the University of Oregon explored how faces could be decoded from neural activity in the study Reconstructing Perceived and Retrieved Faces from Activity Patterns in Lateral Parietal Cortex, published in the Journal of Neuroscience. Hongmi Lee and Brice A. Kuhl tested whether faces could be reconstructed from the'angular gyrus' (ANG) located in the upper back area of the brain through functional magnetic resonance imaging (fMRI) activity patterns. They conducted the experiment by making facial reconstructions based on brainwave patterns from participants, initially during their perception of faces and later just from memory. Participants were shown more than 1,000 color photos of different faces, one after another, while an fMRI scan recorded their neural responses. The researchers then applied principal component analysis (PCA) to generate 300 'eigenfaces' - a set of vectors used in human face recognition.