Bear Naked is tapping into IBM Watson's cognitive capabilities While artificial intelligence has the potential to enhance the smartphone's role as a marketing platform through highly relevant one-to-one services, it could also eliminate the need for a phone to enable engagements as the technology evolves.

Watson, IBM's computer brain, has a lot of talents. It mastered "Jeopardy!," it cooks, and even tries to cure cancer. On Tuesday, IBM Security announced a new cloud-based version of the cognitive technology, dubbed "Watson for Cybersecurity." In the fall, IBM will be partnering with eight universities to help get Watson up to speed by flooding it with security reports and data. The plan as of now is for Watson to process up to 15,000 documents about digital security a month -- including everything from blog posts to videos -- so that it can get a feel for the sometimes esoteric terminology of the cybersecurity world.

It's no secret that much of the wisdom of the world lies in unstructured data, or the kind that's not necessarily quantifiable and tidy. So it is in cybersecurity, and now IBM is putting Watson to work to make that knowledge more accessible. Towards that end, IBM Security on Tuesday announced a new year-long research project through which it will collaborate with eight universities to help train its Watson artificial-intelligence system to tackle cybercrime. Knowledge about threats is often hidden in unstructured sources such as blogs, research reports and documentation, said Kevin Skapinetz, director of strategy for IBM Security. "Let's say tomorrow there's an article about a new type of malware, then a bunch of follow-up blogs," Skapinetz explained.

Over the last few blog posts, I've discussed some of the basics of what machine learning is and why it's important: Throughout those posts, I've been using the following definition of machine learning: creating computational systems that learn from data in order to make predictions and inferences. However, machine learning isn't the only subject in which we use data for prediction and inference. Anyone who's taken an introductory statistics class has heard a similar definition about statistics itself. And if you talk to someone who works in data-mining, you'll hear the same thing: data mining is about using data to make predictions and draw conclusions from data. This raises the question: what is the difference between machine learning, statistics, and data mining? The long answer has a bit of nuance (which we'll discuss soon), but the short answer answer is very simple: machine learning, statistical learning, and data mining are almost exactly the same.

As far as intelligent personal assistants go, the ecosystem has been clearly divided between the top three AI based assistants - Siri, Google Now and Cortana. Of course, there's also Amazon's Alexa, which drives the company's Echo smarthome hub, but it still hasn't reached larger audiences because of its non-smartphone existence. Now, Siri co-founder, Dag Kittlaus has showcased an impressive new AI based digital assistant called Viv, and it may just kick the top three personal assistants out of business, unless of course they up their game. Yesterday, Kittlaus introduced Viv at Tech Crunch Disrupt in New York, where it managed to wow audiences with its ability to handle extremely complicated commands. "Will it be warmer than 70 degrees near the Golden Gate Bridge after 5PM the day after tomorrow?"

There's a new high-tech gym in the works, but it has nothing to do with toning muscles. Elon Musk, the tech-loving chief executive of both Tesla and SpaceX, is building an open-source "gym" for computer programmers to train their robots, Sharon Gaudin reported for ComputerWorld. Developers can share their findings on artificial intelligence (AI) systems as they work out. "Nothing beats a competitive environment to motivate developers," said Patrick Moorhead, an analyst at Moor Insights & Strategy, to ComputerWorld. Championing AI marks part of a calculated strategy by Mr. Musk, who has said AI could represent humanity's "biggest existential threat" because it could prove destructive in the wrong hands or simply go astray on its own.

The graphics processors, or GPUs, that make possible the eye-poppingly realistic graphics of games like "Quantum Break" are also really well-suited to powering artificial intelligence and other high-intensity tasks. It turns out that as video game graphics have gotten better, the hardware used to produce them is increasingly well-suited to powering the AI future envisioned by companies like Google and Facebook. "[After] 2007, all the big advances in FLOPS came from gaming video cards designed for high speed real time 3D rendering, and as an incredibly beneficial side effect, they also turn out to be crazily fast at machine learning tasks," wrote Stack Overflow founder Jeff Atwood in a March 2016 blog entry. In fact, when the Google DeepMind AI won its history-making Go series against Lee Sedol, it was sporting 1,202 CPUs, or traditional processors, and 176 Nvidia GPUs under the hood. Nvidia and Google are actually partners on artificial intelligence, dating back to the Google Brain image recognition system, as detailed in an Nvidia blog entry.

With the growing importance of large network models and enormous training datasets, GPUs have become increasingly necessary to train neural networks. This is largely because conventional optimization algorithms rely on stochastic gradient methods that don't scale well to large numbers of cores in a cluster setting. Furthermore, the convergence of all gradient methods, including batch methods, suffers from common problems like saturation effects, poor conditioning, and saddle points. This paper explores an unconventional training method that uses alternating direction methods and Bregman iteration to train networks without gradient descent steps. The proposed method reduces the network training problem to a sequence of minimization sub-steps that can each be solved globally in closed form. The proposed method is advantageous because it avoids many of the caveats that make gradient methods slow on highly non-convex problems.

As part of my quest to learn about AI, I set myself the goal of building a simple neural network in Python. To ensure I truly understand it, I had to build it from scratch without using a neural network library. Thanks to an excellent blog post by Andrew Trask I achieved my goal. In this blog post, I'll explain how I did it, so you can build your own. I'll also provide a longer, but more beautiful version of the source code.