We've seen RoboBees that can fly, stick to walls, and dive into water. Now, get ready for a hybrid RoboBee that can fly, dive into water, swim, propel itself back out of water, and safely land. New floating devices allow this multipurpose air-water microrobot to stabilize on the water's surface before an internal combustion system ignites to propel it back into the air. This latest-generation RoboBee, which is 1,000 times lighter than any previous aerial-to-aquatic robot, could be used for numerous applications, from search-and-rescue operations to environmental monitoring and biological studies. The research is described in Science Robotics.

We consider the problem of finding the minimizer of a function $f: \mathbb{R}^d \rightarrow \mathbb{R}$ of the form $\min f(w) = \frac{1}{n}\sum_{i}f_i({w})$. This problem has been studied intensively in recent years in machine learning research field. One typical but promising approach for large-scale data is stochastic optimization algorithm. SGDLibrary is a flexible, extensible and efficient pure-Matlab library of a collection of stochastic optimization algorithms. The purpose of the library is to provide researchers and implementers a comprehensive evaluation environment of those algorithms on various machine learning problems.

In reinforcement learning an agent interacts with the environment by taking actions and observing the next state and reward. When sampled probabilistically, these state transitions, rewards, and actions can all induce randomness in the observed long-term return. Traditionally, reinforcement learning algorithms average over this randomness to estimate the value function. In this paper, we build on recent work advocating a distributional approach to reinforcement learning in which the distribution over returns is modeled explicitly instead of only estimating the mean. That is, we examine methods of learning the value distribution instead of the value function. We give results that close a number of gaps between the theoretical and algorithmic results given by Bellemare, Dabney, and Munos (2017). First, we extend existing results to the approximate distribution setting. Second, we present a novel distributional reinforcement learning algorithm consistent with our theoretical formulation. Finally, we evaluate this new algorithm on the Atari 2600 games, observing that it significantly outperforms many of the recent improvements on DQN, including the related distributional algorithm C51.

Ensuring stable, secure and reliable operations of the power grid is a primary concern for system operators [1]. Security assessment and control actions heavily rely on the accuracy of the assumed power system model and its parameters and of the estimated state [2]. Thus, inaccuracies in state estimation data or in the networked dynamic model can impact the assessment of the system stability and the efficacy of the corresponding control measures. In this paper, we explore the possibility to leverage the proliferation of Phasor Measurement Units (PMUs) that collect time synchronous data in a distributed way, for validating the assumed power system model and the current system state. In particular, our goal is to develop a data-efficient learning framework for performing an online reconstruction of the dynamic model using the minimal number of assumptions and exclusively relying on the PMU measurements. A number of recent works showed promising results in attacking this problem [3], [4], [5], [6], [7], [8], [9]. Here, we propose to extend the scope of existing works to the problem of extracting the dynamic state matrix from PMU measurements in a purely data-driven way, without assuming any knowledge of model parameters. We take advantage of the separation of scales that exists in the regime of ambient fluctuations around the steady state leading to power system dynamics excited by stochastic load variations.

The goal of regression and classification methods in supervised learning is to minimize the empirical risk, that is, the expectation of some loss function quantifying the prediction error under the empirical distribution. When facing scarce training data, overfitting is typically mitigated by adding regularization terms to the objective that penalize hypothesis complexity. In this paper we introduce new regularization techniques using ideas from distributionally robust optimization, and we give new probabilistic interpretations to existing techniques. Specifically, we propose to minimize the worst-case expected loss, where the worst case is taken over the ball of all (continuous or discrete) distributions that have a bounded transportation distance from the (discrete) empirical distribution. By choosing the radius of this ball judiciously, we can guarantee that the worst-case expected loss provides an upper confidence bound on the loss on test data, thus offering new generalization bounds. We prove that the resulting regularized learning problems are tractable and can be tractably kernelized for many popular loss functions. We validate our theoretical out-of-sample guarantees through simulated and empirical experiments.

Josiah Renaudin: Welcome back to another TechWell interview. I'm joined by Jason Arbon, the CEO of Appdiff and a speaker at this year's STAR WEST. First, could you tell us a bit about where you worked at before you started Appdiff? Jason Arbon: Hi, Josiah, nice to chat with you again. After college, I started my career at Microsoft doing testing and automation for products like Windows and Bing.

The Artificial Intelligence and Global Security Summit will bring together technology leaders and top policymakers to explore the state of artificial intelligence and discuss the implications of the AI revolution on global security. Past industrial revolutions led to changes in the balance of power between nations and even the fundamental building blocks of power, with coal- and steel-producing nations benefitting and oil becoming a global strategic resource. The AI revolution has similar transformative potential to alter power dynamics, the character of conflict, and strategic stability among nations and private actors. The United States must anticipate these changes and capitalize on opportunities to stay ahead of competitors. To anticipate these challenges, CNAS' all-day summit will explore technology trends, uncertainties, and possible trajectories for how AI may affect global security.

Since its founding by Elon Musk and others nearly two years ago, nonprofit research lab OpenAI has published dozens of research papers. One posted online Thursday is different: Its lead author is still in high school. The wunderkind is Kevin Frans, a senior currently working on his college applications. He trained his first neural net--the kind of system that tech giants use to recognize your voice or face--two years ago, at the age of 15. Inspired by reports of software mastering Atari games and the board game Go, he has since been reading research papers and building pieces of what they described.

Some people are afraid that heavily armed artificially intelligent robots might take over the world, enslaving humanity -- or perhaps exterminating us. These people, including tech-industry billionaire Elon Musk and eminent physicist Stephen Hawking, say artificial intelligence technology needs to be regulated to manage the risks. But Microsoft founder Bill Gates and Facebook's Mark Zuckerberg disagree, saying the technology is not nearly advanced enough for those worries to be realistic. As someone who researches how AI works in robotic decision-making, drones and self-driving vehicles, I've seen how beneficial it can be. I've developed AI software that lets robots working in teams make individual decisions, as part of collective efforts to explore and solve problems.

As a new sub-discipline of Data Science, I notice that SYSTEMS Analytics is starting to get some traction! There are a couple of Analytics graduate level programs with *Systems* in its title (Stevens Institute of Technology and University of North Carolina are the only ones I know). Web search brings up NO books on *Systems* Analytics. With the publication of my book with *Systems* in the title, that gap has been filled now! "SYSTEMS Analytics: Adaptive Machine Learning workbook". My last Analytics startup launched in 2013 explicitly used SYSTEMS Analytics in our Retail Recommendation and Uplift SaaS product; my initial bias for the Systems approach was confirmed by the success of our product.