So there's been a lot of coverage by various websites, data science gurus, and AI experts about what 2019 holds in store for us. Everywhere you look, we have new fads and concepts for the new year. This article is going to be rather different. We are going to highlight the dark horses – the trends that no one has thought about but will completely disrupt the working IT environment (for both good and bad – depends upon which side of the disruption you are on), in a significant manner. So, in order to give you a taste of what's coming up, let's go through the top four (plus 1 (bonus) five) top trends of 2019 for data science: This single innovation is going to change the way machine learning works in the real world.

DAVOS - Singapore has released a framework on how artificial intelligence (AI) can be ethically and responsibly used, which businesses in the Republic and elsewhere can adopt as they grapple with issues that have emerged with new technology. This model framework for AI governance is a "living document" intended to evolve along with the fast-paced changes in a digital economy. It takes in feedback from the industry, and will be tweaked when it gets more views. It was released by Mr S. Iswaran, Minister for Communications and Information, at the World Economic Forum (WEF) meeting which he is attending. The framework is the first in Asia to provide detailed and readily implementable guidance to private sector organisations using AI, said the Infocomm Media Development Authority (IMDA).

This week, Jordan Erica Webber chats to a computer scientist who programs robots to help them become more likeable, making it easier for humans to welcome them at work or at home. Heather Knight directs the Charisma* Robotics Lab, whose goal is to borrow methods from the performing arts to produce more charismatic robots. She also created Marilyn Monrobot, a robot theatre company with comedy performances, and the annual Robot film festival. She believes that in order to know what we need from our technologies, we first need to program them to be more delightful and less frustrating. A special shout out to engineer Eric Gleske, Larry Prybil and Steve Lunderberg at Oregon State University for helping us set up this interview.

For the first time this year, the United States has sent two warships through the strategic Taiwan Strait, according to the Taiwanese government. The move risks further heightening tensions with China, which considers Taiwan a breakaway province and has not ruled out the use of force to bring the self-ruled island under its control. It is also likely to be viewed in Taiwan as a sign of support from US President Donald Trump's administration amid growing friction between Taipei and Beijing. Taiwan's defence ministry said in a statement late on Thursday the US ships were moving in a northerly direction and that their voyage was in accordance with regulations. It added that Taiwan closely monitored the operation to "ensure the security of the seas and regional stability".

The first time Azim Shariff met Iyad Rahwan--the first real time, after communicating with him by phone and e-mail--was in a driverless car. It was November, 2012, and Rahwan, a thirty-four-year-old professor of computing and information science, was researching artificial intelligence at the Masdar Institute of Science and Technology, a university in Abu Dhabi. He was eager to explore how concepts within psychology--including social networks and collective reasoning--might inform machine learning, but there were few psychologists working in the U.A.E. Shariff, a thirty-one-year-old with wild hair and expressive eyebrows, was teaching psychology at New York University's campus in Abu Dhabi; he guesses that he was one of four research psychologists in the region at the time, an estimate that Rahwan told me "doesn't sound like an exaggeration." Rahwan cold-e-mailed Shariff and invited him to visit his research group.

Language is a tool, and just like any tool, it has equal capacity to inflict both good and bad on the world. Language is a beautiful, human thing; the connective tissue that transfers culture, knowledge and critical information across borders and generations. It's that second function -- the divisive one -- that inspired developer Rami Ismail and voice actor Sarah Elmaleh to produce a conference for game creators that removes language as a barrier to entry. Gamedev.world is billed as the first truly global online games conference, with plans to host 48 hours of expert panels and live Q&A sessions on Twitch, YouTube and Mixer, translated in real-time into English, Japanese, Spanish, Portuguese, Russian, Arabic and Simplified Chinese. It's all scheduled to take place later this year. "If games can be played by anyone, and made by anyone, we want to make sure everyone feels like they truly belong here," Ismail told Engadget.

Which countries are the top data producers? After all, with data-fueled applications of artificial intelligence projected, by McKinsey, to generate $13 trillion in new global economic activity by 2030, this could determine the next world order, much like the role that oil production has played in creating economic power players in the preceding century. While China and the U.S. could emerge as two AI superpowers, data sources can't be limited to concentrations in a few places as we have with an oil-driven economy -- it needs to be drawn from many, diverse sources and future AI applications will emerge from new and unexpected players. The new world order taking shape is likely to be more complex than a simple bi-polar structure, especially since data is being produced at a pace that boggles the mind. Building on our past work mapping the digital evolution and digital competitiveness of different countries around the world, we wanted to try to locate the deepest and widest pools of useful data. This is essential to run the myriad machine learning models critical to AI.

Adaptive methods such as Adam and RMSProp are widely used in deep learning but are not well understood. In this paper, we seek a crisp, clean and precise characterization of their behavior in nonconvex settings. To this end, we first provide a novel view of adaptive methods as preconditioned SGD, where the preconditioner is estimated in an online manner. By studying the preconditioner on its own, we elucidate its purpose: it rescales the stochastic gradient noise to be isotropic near stationary points, which helps escape saddle points. Furthermore, we show that adaptive methods can efficiently estimate the aforementioned preconditioner. By gluing together these two components, we provide the first (to our knowledge) second-order convergence result for any adaptive method. The key insight from our analysis is that, compared to SGD, adaptive methods escape saddle points faster, and can converge faster overall to second-order stationary points.

How humans can distinguish between general categories of objects? Are the subcategories of living things visually distinctive? In a number of semantic-category deficits, patients are good at making broad categorization but are unable to remember fine and specific details. It has been well accepted that general information about concepts are more robust to damages related to semantic memory. Results from patients with semantic memory disorders demonstrate the loss of ability in subcategory recognition. While bottom-up feature construction has been studied in detail, little attention has been served to top-down approach and the type of features that could account for general categorization. In this paper, we show that broad categories of animal and plant are visually distinguishable without processing textural information. To this aim, we utilize shape descriptors with an additional phase of feature learning. The results are evaluated with both supervised and unsupervised learning mechanisms. The obtained results demonstrate that global encoding of visual appearance of objects accounts for high discrimination between animal and plant object categories.

Many privacy mechanisms reveal high-level information about a data distribution through noisy measurements. It is common to use this information to estimate the answers to new queries. In this work, we provide an approach to solve this estimation problem efficiently using graphical models, which is particularly effective when the distribution is high-dimensional but the measurements are over low-dimensional marginals. We show that our approach is far more efficient than existing estimation techniques from the privacy literature and that it can improve the accuracy and scalability of many state-of-the-art mechanisms.