Democratizing Machine Learning has always been BigML's founding mission, so we are continually searching for new opportunities. As such, when a company is interested in our technology and is willing to help us further our cause of "Machine Learning for everyone", we feel the urge to collaborate. This is exactly what happened with our new education partner. Today we are happy to announce our educational collaboration with CICE, the Leading School in New Technologies Training in Madrid, Spain. CICE, the only Official Training Center in Spain for more than 20 multinational companies, is already a community of 70,000 students from 30 different countries.

Online learning is becoming the norm in universities across the globe, bringing sweeping changes to the way we learn. But earlier this year on online graduate class at Georgia Tech took things a stage further. "Our Teaching Assistants are getting bogged down answering routine questions," said Ashok Goel, who teaches a graduate science course. Students in the class typically post 10,000 messages a semester on the Piazza forum for the course, many of which are either variations on a theme or simple logistical questions. To address this problem, Ashok turned to IBM Watson to create a virtual TA called Jill Watson who was trained on 40,000 posts and released to the wild on the live forum in March as an addition to the other eight TAs.

Performance of machine learning algorithms depends critically on identifying a good set of hyperparameters. While current methods offer efficiencies by adaptively choosing new configurations to train, an alternative strategy is to adaptively allocate resources across the selected configurations. We formulate hyperparameter optimization as a pure-exploration non-stochastic infinitely many armed bandit problem where a predefined resource like iterations, data samples, or features is allocated to randomly sampled configurations. We introduce Hyperband for this framework and analyze its theoretical properties, providing several desirable guarantees. Furthermore, we compare Hyperband with state-of-the-art methods on a suite of hyperparameter optimization problems. We observe that Hyperband provides five times to thirty times speedup over state-of-the-art Bayesian optimization algorithms on a variety of deep-learning and kernel-based learning problems.

These days, it seems that nearly everything is a smart product. We have smart boards in K-12 classrooms, smart watches, and even smart refrigerators. You can immediately tell that manufacturers love to use this misguided descriptor whenever they integrate modern technology, like touch screens or internet connectivity, to a previously existing product. Do products like these deserve this term? What exactly makes them smart?

Amazon, Baidu, Facebook, Google and Microsoft, among other technology companies, have been investing heavily in artificial intelligence and related disciplines like machine learning because they see the technology enabling services that become a source of revenue. Consultancy Accenture earlier this week quantified this enthusiasm, predicting that AI "could double annual economic growth rates by 2035 by changing the nature of work and spawning a new relationship between man and machine" and by boosting labor productivity by 40 per cent. Certainly things could work out well for Accenture, which a day later announced a partnership with Google to help companies deploy Google technology like machine learning. It's as if the global services firm has a stake in the future it foresees. But the machine learning algorithms underpinning this harmonious union of people and circuits aren't secure. In a paper [PDF] presented in August at the 25th Annual Usenix Security Symposium, researchers at École Polytechnique Fédérale de Lausanne, Cornell University, and The University of North Carolina at Chapel Hill showed that machine learning models can be stolen and that basic security measures don't really mitigate attacks.

According to the Financial Times, Pablo Picasso once said, "Computers are useless. They can only give you answers." Unfortunately for us, computers may now be asking more questions than they answer. As a result, the possibilities are rather overwhelming, with answers more ambiguous and uncertain than straightforward. Similarly, we might ask ourselves where we draw the line when it comes to what we find ethically acceptable in terms of artificial intelligence (AI) as it relates to composition/creation in the worlds of art, writing, performing arts and music--as well as liberal arts education. Most of us are aware of music streaming services that select songs for us based on data about users' listening preferences.

Tom provides an overview of how to build a simple neural net in this introductory tutorial. This screencast shows how to build a Linear Classifier using Deeplearning4j.

On behalf of all at Intel who are focused everyday on supporting you to build, optimize and innovate on Intel architecture, I am proud to announce the launch of Intel Nervana AI Academy. We're at a great moment in the evolution of artificial intelligence (AI) that will open up incredible new experiences fueling the next wave of business opportunity, scientific discovery, and societal improvement. After decades of collectively advancing Intel architecture, we have reached the point that Moore's Law is creating new opportunities for the data science community. We're now supporting artificial intelligence as an enablement technology, through optimized machine and deep learning, providing developers new ways to add value to solutions and applications. Intel is investing heavily in AI and you.

John Charles Thomas was born in Akron, Ohio and attended Ellet High School. He graduated from Case Western Reserve University majoring in psychology and minoring in mathematics and drama. He received a Ph.D. in experimental psychology from the University of Michigan. His dissertation compared human performance in a problem solving task to that of an early AI system called "The General Problem Solver." After graduate school, Dr. Thomas managed a research project on the psychology of aging at Harvard Medical School.

We consider the problem of estimating the expected value of information (the knowledge gradient) for Bayesian learning problems where the belief model is nonlinear in the parameters. Our goal is to maximize some metric, while simultaneously learning the unknown parameters of the nonlinear belief model, by guiding a sequential experimentation process which is expensive. We overcome the problem of computing the expected value of an experiment, which is computationally intractable, by using a sampled approximation, which helps to guide experiments but does not provide an accurate estimate of the unknown parameters. We then introduce a resampling process which allows the sampled model to adapt to new information, exploiting past experiments. We show theoretically that the method converges asymptotically to the true parameters, while simultaneously maximizing our metric. We show empirically that the process exhibits rapid convergence, yielding good results with a very small number of experiments.