Nick Triantos spent the past year trying to commercialize the technology being developed at SRI International, the Stanford research offshoot that helped invent the Internet and Siri, among other things. Now he has joined Ignition Partners at its new office in Los Altos, where he says he will be able to help create companies from a broader range of innovation. Triantos said his focus will be on business-focused startups, particularly ones working in artificial intelligence, cybersecurity and augmented and virtual reality. But he doesn't expect that will include startups in the currently hot space of chatbots, despite his background with voice recognition and machine learning at SRI. The following Q&A about these and other topics has been edited for length and clarity. What was your role at SRI and why are you leaving there? Unfortunately, not many people know about SRI.

Explicit high-order feature interactions efficiently capture essential structural knowledge about the data of interest and have been used for constructing generative models. We present a supervised discriminative High-Order Parametric Embedding (HOPE) approach to data visualization and compression. Compared to deep embedding models with complicated deep architectures, HOPE generates more effective high-order feature mapping through an embarrassingly simple shallow model. Furthermore, two approaches to generating a small number of exemplars conveying high-order interactions to represent large-scale data sets are proposed. These exemplars in combination with the feature mapping learned by HOPE effectively capture essential data variations. Moreover, through HOPE, these exemplars are employed to increase the computational efficiency of kNN classification for fast information retrieval by thousands of times. For classification in two-dimensional embedding space on MNIST and USPS datasets, our shallow method HOPE with simple Sigmoid transformations significantly outperforms state-of-the-art supervised deep embedding models based on deep neural networks, and even achieved historically low test error rate of 0.65% in two-dimensional space on MNIST, which demonstrates the representational efficiency and power of supervised shallow models with high-order feature interactions.

A novel efficient method for computing the Knowledge-Gradient policy for Continuous Parameters (KGCP) for deterministic optimization is derived. The differences with Expected Improvement (EI), a popular choice for Bayesian optimization of deterministic engineering simulations, are explored. Both policies and the Upper Confidence Bound (UCB) policy are compared on a number of benchmark functions including a problem from structural dynamics. It is empirically shown that KGCP has similar performance as the EI policy for many problems, but has better convergence properties for complex (multi-modal) optimization problems as it emphasizes more on exploration when the model is confident about the shape of optimal regions. In addition, the relationship between Maximum Likelihood Estimation (MLE) and slice sampling for estimation of the hyperparameters of the underlying models, and the complexity of the problem at hand, is studied.

In the probabilistic topic models, the quantity of interest---a low-rank matrix consisting of topic vectors---is hidden in the text corpus matrix, masked by noise, and the Singular Value Decomposition (SVD) is a potentially useful tool for learning such a low-rank matrix. However, the connection between this low-rank matrix and the singular vectors of the text corpus matrix are usually complicated and hard to spell out, so how to use SVD for learning topic models faces challenges. In this paper, we overcome the challenge by revealing a surprising insight: there is a low-dimensional simplex structure which can be viewed as a bridge between the low-rank matrix of interest and the SVD of the text corpus matrix, and allows us to conveniently reconstruct the former using the latter. Such an insight motivates a new SVD approach to learning topic models, which we analyze with delicate random matrix theory and derive the rate of convergence. We support our methods and theory numerically, using both simulated data and real data.

Most learning algorithms are not invariant to the scale of the function that is being approximated. We propose to adaptively normalize the targets used in learning. This is useful in value-based reinforcement learning, where the magnitude of appropriate value approximations can change over time when we update the policy of behavior. Our main motivation is prior work on learning to play Atari games, where the rewards were all clipped to a predetermined range. This clipping facilitates learning across many different games with a single learning algorithm, but a clipped reward function can result in qualitatively different behavior. Using the adaptive normalization we can remove this domain-specific heuristic without diminishing overall performance.

Machine Learning is one of the most exciting fields in the world. Every week we discover something new, something amazing, something revolutionary. It's incredible, but it can also be overwhelming. That's why we created This Week in Machine Learning! Each week we publish a curated list of Machine Learning stories as a resource to help you keep pace with all these exciting developments.

The People First Social Network Gab What is Gab? gab?ab/ informal verb talk, typically at length, about trivial matters. Gab is a people first social network-Users can post "Gabs," which have a 300 character limit-Users can follow other Gabbers and be followed back-Users can upvote or downvote Gabs-Top Gabs are ranked based on these votes-Gabs are also displayed in a chronological home feed, something that is no longer a defau... Rose Behar August 15, 2016 4:56pm In an expansive interview with the Washington Post, Apple CEO Tim Cook opened up about a number of personal and professional subjects, from the importance of his public coming out to why he believes analysts are wrong that Apple has nowhere left to grow. Perhaps most interesting for Apple product enthusiasts was his admissions about what he sees as core technologies of the future, namely AI and augmented reality. In explaining why he believes mobile "is the grea... An Elon Musk-backed artificial intelligence research group just got a brand new toy from chip maker Nvidia.

From the invention of the wheel and steam engine to fax machines and desktop computers, technology has always shaped the way we work – but in the last few decades, the pace of innovation has sped up exponentially, forcing employees and those who lead them to constantly blaze new ground and determine new paradigms for the way things are done. The biggest recent change in work and workplace culture is the introduction of robots. This change has caused a lot of panic and unsettlement among researchers, pundits and everyday workers. There are many conflicting stats and studies that spell out doomsday scenarios for human workers, both in-house and outsourced, in the age of robotics: Forrester estimates that 22.7 million jobs will be displaced by 2025; the World Economic Forum estimates that it's closer to 5 million jobs by 2020. But these doomsday scenarios, observe Professor Leslie Willcocks and Professor Mary Lacity in their book Service Automation: Robotics & the Future of Work, rest on a few crucial flaws – namely, attaching scary numbers to vague dates, highlighting job loss without highlighting job creation and forgetting that historically we've adapted to larger changes in the job market without disaster.

There is a surge of interest and research into Artificial Intelligence (AI). AI is seen as the new technological revolution in the work place with machines tipped to replace most human jobs. There is significant improvement in the field of AI currently, than in the history of mankind, AI has been branded a failure or a hype in the past, but that notion does not seem to be the case anymore with the emergence of more sophisticated computer algorithms that have helped machines to pass the Turin Test – A test which determines whether or not a machine computer is capable of thinking like a human. Critics have branded the Turin test an emotionally unsatisfying test for intelligence but agreed that a machine passing the Turin test is an important milestone for AI. It is important to highlight that to get more conclusive results from tests of AI, the Turin test has had twists and variations from the original test by Alan Turin.

We've been hearing a drumbeat recently of claims that a universal basic income--in effect, a monthly welfare check sent to everyone--is going to be necessary to save all the poor unfortunate souls put out of work by self-driving cars, artificial intelligence, robots, and other new forms of automation. We are told that the basic income will be "the only way to keep the country's economy afloat" in an age of automation, or that it will be necessary to absorb millions of truckers thrown out of middle-class jobs by the advent of autonomous vehicles. Of course, this being the field of high technology, there are always those who will say that it's not a bug but a feature. So we get Peter Diamandis reassuring us that "technological socialism" can "demonetize living." I have already thrown some skepticism at the idea that there is going to be a traumatic transition that will throw middle class people out on the streets without warning--rather than a long and gradual transition over decades, to which people can adapt.