The famous Policy Iteration algorithm alternates between policy improvement and policy evaluation. Implementations of this algorithm with several variants of the latter evaluation stage, e.g, $n$-step and trace-based returns, have been analyzed in previous works. However, the case of multiple-step lookahead policy improvement, despite the recent increase in empirical evidence of its strength, has to our knowledge not been carefully analyzed yet. In this work, we introduce the first such analysis. Namely, we formulate variants of multiple-step policy improvement, derive new algorithms using these definitions and prove their convergence. Moreover, we show that recent prominent Reinforcement Learning algorithms are, in fact, instances of our framework. We thus shed light on their empirical success and give a recipe for deriving new algorithms for future study.

We study the rates of convergence from empirical surrogate risk minimizers to the Bayes optimal classifier. Specifically, we introduce the notion of \emph{consistency intensity} to characterize a surrogate loss function and exploit this notion to obtain the rate of convergence from an empirical surrogate risk minimizer to the Bayes optimal classifier, enabling fair comparisons of the excess risks of different surrogate risk minimizers. The main result of the paper has practical implications including (1) showing that hinge loss is superior to logistic and exponential loss in the sense that its empirical minimizer converges faster to the Bayes optimal classifier and (2) guiding to modify surrogate loss functions to accelerate the convergence to the Bayes optimal classifier.

Inference amortization methods allow the sharing of statistical strength across related observations when learning to perform posterior inference. Generally this requires the inversion of the dependency structure in the generative model, as the modeller must design and learn a distribution to approximate the posterior. Previous methods invert the dependency structure in a heuristic way and fail to capture the dependencies in the model, therefore limiting the performance of the eventual inference algorithm. We introduce an algorithm for faithfully and minimally inverting the graphical model structure of any generative model. Such an inversion has two crucial properties: a) it does not encode any independence assertions absent from the model, and b) for a given inversion, it encodes as many true independence assertions as possible. Our algorithm works by simulating variable elimination on the generative model to reparametrize the distribution. We show with experiments how such minimal inversions can assist in performing better inference.

A startup using artificial intelligence (AI) to tackle the proliferation of fake news and extremist content online will expand after closing a $1m (£720,000) seed funding round that attracted major US backers. London-based Factmata's funding round was led by tech entrepreneur Mark Cuban, but also attracted high-profile investors such as Twitter co-founder Biz Stone, internet entrepreneur Sunil Paul and, more recently, Craigslist founder Craig Newmark. The company will put the cash towards research and development, product development and expanding its team beyond machine-learning specialists. The round was initially closed last August but soon reopened to close formally in the last couple of weeks. Read more: Twitter fails to give MPs "straight answers" on Russian influence on Brexit "Every day there's an article or there's someone talking about fake news at the World Economic Forum, so I wanted to keep the round open so people could come in quite late in the process," founder and chief executive Dhruv Ghulati told City A.M. Factmata is developing both business and consumer-facing products, such as an anti-fake news platform for journalists, researchers and the public to use, as well as collaborating globally with advertisers and businesses that can use its algorithms to sift through and identify fake news, spoof sites and hate content. Ghulati, who was recently named one of Forbes' 30 Under 30 On the B2B side, this would help advertisers find junk content on potential sites they would be advertising on.

We communicate a lot with our eyes; a glance can mean one thing, a stare something completely different. That's the promise of a new, peripheral-free eye tracking software called IrisGo that was introduced at CES 2018, the consumer technology show that ran Jan. 9-12, 2018, in Las Vegas. According to its developer, San Sebastián, Spain-based Irisbond, the software measures and responds to the motion of the human eye and allows users to navigate devices and control screens with a gaze. Because it's not tethered to additional hardware, it also enables users to communicate on any device with an embedded camera. IrisGo has been used to control collaborative robots at ABB Spain; improve the fidelity of neuromarketing data collected by Lumen Research; and further the public conversation by being tested in smartphones by Twitter.

The World Health Organization took an unprecedented step in January when it decided to include "gaming disorder" in its 11th International Classification of Diseases (IDC). Though doctors and researchers have examined the effects of heavy internet usage since the days when access arrived on AOL CDs, this marks the first time that the organization has listed this disorder as a mental health condition. Doing so could have far-reaching, and potentially negative, implications for how the disorder is diagnosed and treated. But video games aren't the only aspect of internet society that has people concerned. A 2016 study by Common Sense Media, a nonprofit that helps teens and their parents navigate modern media, found that nearly half of the teens surveyed described themselves as "addicted" to their phones.

Google's quest is to show results in the best interest of the searcher – Faster, better, relevant, and useful. And that should be your quest too! However, Google wouldn't just find out you're creating amazing content which is the best fit for their searchers. She won't just find you. You have to let her know. That's what SEO is all about. One of the best ways to predict and understand how Google works is through experiments and experts. Over the past few weeks, We had the opportunity to interview SEO experts and get their best tips and advice for the changes in the SEO landscape for 2018.

The global tech startup scene is a noisy, crowded space. And then there are the stories of those entrepreneurs whose ideas have endured into something truly transformational; Amazon, Apple, Google, you name it--many of the biggest and most influential companies in the world today were born of this heritage. "One thing we can all agree on: The key attribute of a startup is its ability to grow," wrote Forbes' Natalie Robehmed. And as my former CEO, Mark Jones, used to say, "All big companies were once small companies too. The only difference is that they grew up."

Basic human error or a lack of understanding of how disclosure works will always remain potential stumbling blocks even if AI was made available to help alleviate the increased volume of data now being gathered to ensure a successful prosecution. A senior UK police chief revealed in a speech in London on Wednesday, February 7, forces should examine using artificial intelligence to help cope with the scale of information involved in investigations and avoid the kind of mistakes that resulted in a string of collapsed rape trials. Sara Thornton, the chair of the National Police Chiefs' Council, said the volume of date held by individuals had massively increased the number of potential lines of enquiry that officers must pursue to understand a case. Her views have, however, been questioned by a leading expert in counter-terrorism and organised crime who insists the police have already had the ability to comb through large volumes of digital data under the Criminal Procedure and Investigations Act 1996. In an exclusive interview, David Vidicette, now the author of several crime thriller books including The Theseus Paradox, said this has always been standard procedure and normal police and detective work.

The Carlsberg Research Laboratory in Denmark creates 1,000 different beer samples, daily. With such a strong focus on research, it's no surprise that Carlsberg is looking towards the future, and the opportunities that technology can provide. The company's new multi-million research study, enticingly named The Beer Fingerprinting Project, looks set to change the way how new beers are created and enjoyed – and it's all thanks to Artificial Intelligence (AI). Intelligent beer The brainchild of Jochen Förster, Director and Professor Yeast Fermentation, Carlsberg Group, the aim of the pioneering project is to use a series of high-tech sensors which can accurately guage the delicate nuances and aromas in the beer, mapping out a'flavour fingerprint' for each individual sample. Information gained from this system can then be used to explore new brewing organisms, ultimately leading to the creation of new beers.