We study the problem of estimating the mean of a random vector $X$ given a sample of $N$ independent, identically distributed points. We introduce a new estimator that achieves a purely sub-Gaussian performance under the only condition that the second moment of $X$ exists. The estimator is based on a novel concept of a multivariate median.

Similarly, the observation is limited in angular resolution by the resolving power of the telescope (R λ/D) and, if taken from the ground, by the distortions caused by the moving atmosphere (the "seeing"). The total blurring introduced by the combination of the telescope and the atmosphere is described by the point spread function (PSF). An image taken by a telescope can therefore be thought of as a convolution of the true light distribution with this point spread function plus the addition of various sources of noise. The Shannon-Nyquist sampling theorem (Nyquist 1928; Shannon 1949) limits the ability of deconvolution techniques in removing the effect of the PSF, particularly in the presence of noise (Magain et al. 1998; Courbin 1999; Starck et al. 2002). Deconvolution has long been known as an "ill-posed" ABSTRACT Observations of astrophysical objects such as galaxies are limited by various sources of random and systematic noise from the sky background, the optical system of the telescope and the detector used to record the data. Conventional deconvolution techniques are limited in their ability to recover features in imaging data by the Shannon-Nyquist sampling theorem. Here we train a generative adversarial network (GAN) on a sample of 4, 550 images of nearby galaxies at 0.01 z 0.02 from the Sloan Digital Sky Survey and conduct 10 cross validation to evaluate the results. We present a method using a GAN trained on galaxy images that can recover features from artificially degraded images with worse seeing and higher noise than the original with a performance which far exceeds simple deconvolution. The ability to better recover detailed features such as galaxy morphology from low-signal-to-noise and low angular resolution imaging data significantly increases our ability to study existing data sets of astrophysical objects as well as future observations with observatories such as the Large Synoptic Sky Telescope (LSST) and the Hubble and James Webb space telescopes.

In this paper, we present a novel approach for initializing deep neural networks, i.e., by turning PCA into neural layers. Usually, the initialization of the weights of a deep neural network is done in one of the three following ways: 1) with random values, 2) layer-wise, usually as Deep Belief Network or as auto-encoder, and 3) re-use of layers from another network (transfer learning). Therefore, typically, many training epochs are needed before meaningful weights are learned, or a rather similar dataset is required for seeding a fine-tuning of transfer learning. In this paper, we describe how to turn a PCA into an auto-encoder, by generating an encoder layer of the PCA parameters and furthermore adding a decoding layer. We analyze the initialization technique on real documents. First, we show that a PCA-based initialization is quick and leads to a very stable initialization. Furthermore, for the task of layout analysis we investigate the effectiveness of PCA-based initialization and show that it outperforms state-of-the-art random weight initialization methods.

Free variables occur frequently in mathematics and computer science with ad hoc and altering semantics. We present the most recent version of our free-variable framework for two-valued logics with properly improved functionality, but only two kinds of free variables left (instead of three): implicitly universally and implicitly existentially quantified ones, now simply called "free atoms" and "free variables", respectively. The quantificational expressiveness and the problem-solving facilities of our framework exceed standard first-order and even higher-order modal logics, and directly support Fermat's descente infinie. With the improved version of our framework, we can now model also Henkin quantification, neither using quantifiers (binders) nor raising (Skolemization). We propose a new semantics for Hilbert's epsilon as a choice operator with the following features: We avoid overspecification (such as right-uniqueness), but admit indefinite choice, committed choice, and classical logics. Moreover, our semantics for the epsilon supports reductive proof search optimally.

The Rivers Casino in Pittsburgh may not seem a likely setting for a major scientific breakthrough. But on Tuesday, it was: Libratus, an AI system developed by Carnegie Mellon University, beat the world's top four human players in a 20-day tournament of Head's-Up No-Limit Texas Hold'em poker. Libratus, developed by Carnegie Mellon's Tuomas Sandholm, a professor of computer science, and Noam Brown, a Ph.D. student in computer science, competed against Dong Kim, Jimmy Chou, Daniel McAulay, and Jason Les in a competition called "Brains Vs. Artificial Intelligence: Upping the Ante"--during which 120,000 hands were played. "This is the last frontier," said Sandholm during a press conference on Tuesday.

It was no surprise that artificial intelligence (AI) and its impact on business was a key talking point at this year's World Economic Forum in Davos, Switzerland (Davos 2017), especially as concern of AI-powered machinery displacing human workers grows. Speaking at an AI panel at Davos 2017, Microsoft CEO Satya Nadella discussed how simple it was to eliminate human input altogether: "its augmentation or replacement – that's a design choice. You can say replacement [of humans] is the goal, or you can say augmentation is the goal." But while Microsoft are developing tech to aid with human interaction, others may be less willing to design AI machinery that interacts with humans, instead opting to replace them altogether. However, the way in which AI is currently being developed is to work alongside individuals, or, as IBM CEO Ginni Rometty at the panel puts it, "in service" of them.

We want to apply new data to our network and classify inputs If we overtrain / overfit our network to our training data then our accuracy will be deceiving. It might work very well for training data, but will not work on test data. In order to prevent overfitting we implement preprocessing techniques and tune our hyper parameters.

Lily Robotics, which promised a autonomous flying camera, is shutting down operations. Lily Robotics, a failed drone startup that closed last month amid a consumer-protection civil suit from the San Francisco District Attorney's office, may now be the subject of a criminal investigation. Earlier this month, law enforcement agents raided the company's San Francisco headquarters for a potential criminal investigation against the company, according to one source who asked to remain anonymous because they were not authorized to speak openly about the matter. When contacted by FORBES about the visit from law enforcement individuals, Henry Bradlow, Lily Robotic's cofounder and chief technology officer, said he could not comment "on rumors or speculation" and hung up the phone. A spokesperson for the San Francisco District Attorney's office said that he could not confirm or deny if a raid had occurred.

SYS-CON Events announced today that Conference Guru has been named "Media Sponsor" of SYS-CON's 20th International Cloud Expo, which will take place on June 6-8, 2017, at the Javits Center in New York City, NY. A valuable conference experience generates new contacts, sales leads, potential strategic partners and potential investors; helps gather competitive intelligence and even provides inspiration for new products and services. Conference Guru works with conference organizers to pass great deals to great conferences, helping you discover new conferences and increase your return on investment. All major researchers estimate there will be tens of billions devices - computers, smartphones, tablets, and sensors - connected to the Internet by 2020. This number will continue to grow at a rapid pace for the next several decades.

When I think about what makes a fun VR experience, I constantly come back to the work of Tyler Hurd: lead artist on a batch of experiences that can best be described as smile machines. His beat-heavy bits basically force participants to dance like they just don't care (a trick aided by the fact that eye-covering VR headsets allow you to pretend that you're dancing like nobody is watching). When I first came across his work at the Tribeca Film Festival, where his VR video the Future Islands song Old Friend made a well-received appearance, I spent a fair chunk of time simply watching people glow with joy as they danced with abandon. This January at the Sundance Film Festival, I witnessed a similar stream of smiles with his newest experience, Chocolate, which he produced with Viacom Next. Towards the end of Sundance, I caught up with Tyler to talk about his new experience, and what the secret is to making VR that people seem to really, really like.