Given a state-of-the-art deep neural network classifier, we show the existence of a universal (image-agnostic) and very small perturbation vector that causes natural images to be misclassified with high probability. We propose a systematic algorithm for computing universal perturbations, and show that state-of-the-art deep neural networks are highly vulnerable to such perturbations, albeit being quasi-imperceptible to the human eye. We further empirically analyze these universal perturbations and show, in particular, that they generalize very well across neural networks. The surprising existence of universal perturbations reveals important geometric correlations among the high-dimensional decision boundary of classifiers. It further outlines potential security breaches with the existence of single directions in the input space that adversaries can possibly exploit to break a classifier on most natural images.

We describe a strategy for constructing a neural network jet substructure tagger which powerfully discriminates boosted decay signals while remaining largely uncorrelated with the jet mass. This reduces the impact of systematic uncertainties in background modeling while enhancing signal purity, resulting in improved discovery significance relative to existing taggers. The network is trained using an adversarial strategy, resulting in a tagger that learns to balance classification accuracy with decorrelation. As a benchmark scenario, we consider the case where large-radius jets originating from a boosted resonance decay are discriminated from a background of nonresonant quark and gluon jets. We show that in the presence of systematic uncertainties on the background rate, our adversarially-trained, decorrelated tagger considerably outperforms a conventionally trained neural network, despite having a slightly worse signal-background separation power. We generalize the adversarial training technique to include a parametric dependence on the signal hypothesis, training a single network that provides optimized, interpolatable decorrelated jet tagging across a continuous range of hypothetical resonance masses, after training on discrete choices of the signal mass.

We introduce Joint Causal Inference (JCI), a powerful formulation of causal discovery from multiple datasets that allows to jointly learn both the causal structure and targets of interventions from statistical independences in pooled data. Compared with existing constraint-based approaches for causal discovery from multiple data sets, JCI offers several advantages: it allows for several different types of interventions in a unified fashion, it can learn intervention targets, it systematically pools data across different datasets which improves the statistical power of independence tests, and most importantly, it improves on the accuracy and identifiability of the predicted causal relations. A technical complication that arises in JCI is the occurrence of faithfulness violations due to deterministic relations. We propose a simple but effective strategy for dealing with this type of faithfulness violations. We implement it in ACID, a determinism-tolerant extension of Ancestral Causal Inference (ACI) (Magliacane et al., 2016), a recently proposed logic-based causal discovery method that improves reliability of the output by exploiting redundant information in the data. We illustrate the benefits of JCI with ACID with an evaluation on a simulated dataset.

The voice-activated Echo lets users control other smart home devices, play games and order products on Amazon's site. Amazon's mounting First Amendment battle has reached an anticlimactic end. The company agreed to hand over user data of an Amazon Echo speaker for a murder trial in Arkansas, after it spent months pushing back against a warrant for the information. Amazon changed its position after the user, defendant James Andrew Bates, consented to the disclosure, according to a court filing that was made public Monday. Before Bates consented, Amazon just last month offered a strong defense against releasing the user information, with the company saying Bates' audio recordings with the Echo were protected under the First Amendment.

Technology must be controlled in order to safeguard the future of humanity, Stephen Hawking has warned. The physicist, who has spoken out about the dangers of artificial intelligence in the past, says a'world government' could be our only hope. He says our'logic and reason' could be the only way to defeat the growing threat of nuclear or biological war. We are living through the most dangerous time in the history of the human race, according to Professor Stephen Hawking. 'Since civilisation began, aggression has been useful inasmuch as it has definite survival advantages,' he told The Times. 'It is hard-wired into our genes by Darwinian evolution. 'Now, however, technology has advanced at such a pace that this aggression may destroy us all by nuclear or biological war.

Researchers compared a group of gamers' emotional responses to those of others who didn't play frequently and found that their ability to empathize wasn't hindered by violent video games. Researchers at the Hannover Medical School in Germany studied a group of males who played first-person shooter video games for at least two hours daily for the past year. Those subjects were compared to another group of gamers who had not played violent video games or any video games on a regular basis. Each participant underwent an MRI scan while being presented with images designed to produce an emotional response. They were also asked questions on various subjects to determine how empathetic their responses were.

The Amazon Echo, a voice-controlled virtual assistant, is seen at its product launch for Britain and Germany in London, in 2016. After several months of pushback, Amazon has agreed to release user data from an Amazon Echo device involved in a high-profile Arkansas murder trial. The device, a popular, hands-free artificial intelligence assistant named "Alexa" that responds to human directives, contains audio recordings that prosecutors say could could provide information in the murder of Victor Collins, 47, who was found dead in his hot tub on Nov. 22, 2015, in Bentonville, Arkansas. James Bates, 31, was charged with first-degree murder and tampering with evidence in the case. Benton County Prosecuting Attorney Nathan Smith wrote in an email that prosecutors were "pleased" with Amazon's decision.

Artificial intelligence (AI) is not just transforming ecommerce. Ecommerce came about because the internet made it easy to find what you were looking for. The digital age made the physical world appear closer, so you could buy things from distant countries. Early adopters assumed ecommerce would be fueled by economics of scale and lower costs, but those weren't the main drivers: It was convenience. Websites made information freely available.

Imagine that, in 20 or 30 years, a company creates the first artificially intelligent humanoid robot. She looks like a person, talks like a person, interacts like a person. If you were to meet Ava, you could relate to her even though you know she's a robot. Ava is a fully conscious, fully self-aware being: She communicates; she wants things; she improves herself. She is also, importantly, far more intelligent than her human creators.

Machine learning is complex, but nonetheless has pushed its way to professional and maker communities alike. Nvidia has lead much of this with their TK1 and TX1 modules; now, with the new release of the Jetson TX2, the AI capabilities we have access to have just doubled. The new hardware, announced last night at a press event in San Francisco, retains the same form factor as the TX1 -- roughly the size of a credit card, it's meant as a drop-in replacement. It replaces the TX1's Mawell GPU with a Pascal unit, doubles the TX1's storage and memory, and increases its video encoding and decoding specs. With it, the company states that it can get either twice the performance of the TX1 (handling object detection and tracking from two 4K cameras simultaneously), or get double the efficiency running the same configuration as a TX1.