Actor Kristen Stewart, known for her portrayal of Bella in the "Twilight" movie franchise and director of "Come Swim" at the Sundance Film Festival, now has another line on her résumé: co-author of a computer science paper. The paper, published online in the preprint journal ArXiv, is called "Bringing Impressionism to Life with Neural Style Transfer in Come Swim." The authors describe a set of programming shortcuts that can make movie shots look as though they were painted or drawn in a certain style, such as impressionism or pointillism. The process relies on machine learning, a type of artificial intelligence, and gave certain shots in the film short, which uses allusive images to follow a man through his day, the look of an impressionistic painting. The shot described in the paper is about 15 seconds long, and the painting is by Stewart herself.

WorkFusion, the leading provider of intelligent automation, has closed a $35 million Series D financing round led by Georgian Partners with participation from existing investors Mohr Davidow Ventures, iNovia, Nokia Growth Partners (NGP), Greycroft and RTP Ventures, bringing total funds raised to $71 million. The funding will accelerate customer adoption of AI-powered automation as they realize the immense productivity gains from digitizing business processes. "We invested in WorkFusion not only because it aligns with our thesis that AI will transform business, but most importantly because the company has made machine learning practical and powerful for enterprise operations," said Justin LaFayette, Managing Partner at Georgian Partners. "WorkFusion has category leading products and a strong leadership aligned by a powerful vision of helping businesses drive rapid productivity improvements with AI." WorkFusion transforms operations through self-learning, process-level automation that eliminates up to 90% of manual back-office work and AI-powered chat bots that increase front office service center capacity by 5x. "IDC estimates the AI and cognitive systems market will grow from $8 billion in 2016 to over $47 billion in 2020 and businesses are looking to invest in the most innovative and comprehensive automation technologies on the market," said Max Yankelevich, WorkFusion CEO.

Publishing data on Kaggle is a way organizations can reach a diverse audience of data scientists with an enthusiasm for learning, knowledge, and collaboration. For Dr. Erin Miller of START, the National Consortium for the Study of Terrorism and Responses to Terrorism, making her organization's Global Terrorism Database available for analysis by Kaggle users has brought new awareness to their cause. In this Open Data Spotlight, Erin discusses how setting aside agendas and focusing on understanding this unparalleled dataset of over 150,000 attack events allows users to undertake constructive analyses that may defy common conceptions about terrorism. Read on to learn more about the Global Terrorism Database project and the ways users of open data can make valuable contributions to the organizations that make them possible. My role started out (more than 12 years ago) as a graduate assistant cleaning raw data, and now I manage the project team, workflow, resources, and interaction with end users and related research projects.

A cure for cancer and other critical medical conditions could already be in existence as a recent study by Professors of Pharmaceutical Chemistry, Steven Atschuler and his wife, Lani Wu has developed a new method that uses artificial intelligence for faster and systematic drug discovery. The new technique reduces the time and cost previously used to search for possible new drugs to treat illnesses and diseases. The husband and wife research team at University Of California-San Francisco (UCSF) designed a new method to make drug discovery faster and at a cheaper cost than that of the traditional method. Atschuler and his wife have worked together since they met as students almost 30 years ago. Their study is informed not only by their extant collaboration but by other previously shared careers in other fields, according to Phys.

Which authors of this paper are endorsers? Disable MathJax (What is MathJax?)

With every passing day, we're reminded that the future is here. Yeah, that's sort of a redundant thing to say. What I really mean is that new innovations and disruptions are popping up every day, and they're materializing at a rate never seen before. I mean, think about the fact that the modern computer, which was created in either 1942 or 1946, depending on who you ask, used to cost a fortune and fill up an entire room.[1] It was almost 50 years before that computer would be reduced to the size of an affordable desktop in 1995.

Scalable and effective exploration remains a key challenge in reinforcement learning (RL). While there are methods with optimality guarantees in the setting of discrete state and action spaces, these methods cannot be applied in high-dimensional deep RL scenarios. As such, most contemporary RL relies on simple heuristics such as epsilon-greedy exploration or adding Gaussian noise to the controls. This paper introduces Variational Information Maximizing Exploration (VIME), an exploration strategy based on maximization of information gain about the agent's belief of environment dynamics. We propose a practical implementation, using variational inference in Bayesian neural networks which efficiently handles continuous state and action spaces. VIME modifies the MDP reward function, and can be applied with several different underlying RL algorithms. We demonstrate that VIME achieves significantly better performance compared to heuristic exploration methods across a variety of continuous control tasks and algorithms, including tasks with very sparse rewards.

How to self-localize large teams of underwater nodes using only noisy range measurements? How to do it in a distributed way, and incorporating dynamics into the problem? How to reject outliers and produce trustworthy position estimates? The stringent acoustic communication channel and the accuracy needs of our geophysical survey application demand faster and more accurate localization methods. We approach dynamic localization as a MAP estimation problem where the prior encodes dynamics, and we devise a convex relaxation method that takes advantage of previous estimates at each measurement acquisition step; The algorithm converges at an optimal rate for first order methods. LocDyn is distributed: there is no fusion center responsible for processing acquired data and the same simple computations are performed for each node. LocDyn is accurate: experiments attest to a smaller positioning error than a comparable Kalman filter. LocDyn is robust: it rejects outlier noise, while the comparing methods succumb in terms of positioning error.

In coming years residential consumers will face real-time electricity tariffs with energy prices varying day to day, and effective energy saving will require automation - a recommender system, which learns consumer's preferences from her actions. A consumer chooses a scenario of home appliance use to balance her comfort level and the energy bill. We propose a Bayesian learning algorithm to estimate the comfort level function from the history of appliance use. In numeric experiments with datasets generated from a simulation model of a consumer interacting with small home appliances the algorithm outperforms popular regression analysis tools. Our approach can be extended to control an air heating and conditioning system, which is responsible for up to half of a household's energy bill.

We introduce the implicitly constrained least squares (ICLS) classifier, a novel semi-supervised version of the least squares classifier. This classifier minimizes the squared loss on the labeled data among the set of parameters implied by all possible labelings of the unlabeled data. Unlike other discriminative semi-supervised methods, this approach does not introduce explicit additional assumptions into the objective function, but leverages implicit assumptions already present in the choice of the supervised least squares classifier. This method can be formulated as a quadratic programming problem and its solution can be found using a simple gradient descent procedure. We prove that, in a limited 1-dimensional setting, this approach never leads to performance worse than the supervised classifier. Experimental results show that also in the general multidimensional case performance improvements can be expected, both in terms of the squared loss that is intrinsic to the classifier, as well as in terms of the expected classification error.