On Thursday in Montreal, Canadian prime minister Justin Trudeau boasted about his country's leading position in artificial intelligence and openness to international collaboration. A few miles away, the world's largest AI conference proceeded without scores of researchers denied visas by Trudeau's government. All week, Montreal has played host to 8,000 people attending the NeurIPS conference, which ends Saturday. But well over 100 researchers with tickets to attend the event or its associated workshops, including many who planned to present work, are absent due to visa denials or delays. AI researchers say the visa problems undermine efforts to make their field more inclusive, and less likely to produce technology that discriminates or disadvantages people who aren't white or Western.

George Orwell's nightmare vision of the future in his novel 1984 could become reality unless governments are curbed, claims Microsoft's president. Brad Smith has warned that facial recognition technology will lead to dystopian state powers unless new measures are adopted. Microsoft say it has developed a new set of principles for deployment of the technology, called for new laws and industry rivals to follow suit. Mr Smith made the announcement at a Brookings Institution speech and an accompanying blog post, saying it was urgent to begin placing limits on facial recognition to avoid a surveillance state. 'We must ensure that the year 2024 doesn't look like a page from the novel '1984,'' Mr Smith said.

On Monday, The White House announced plans to co-host four upcoming public workshops on various AI topics to "spur public dialogue on artificial intelligence and machine learning and identify challenges and opportunities related to this emerging technology." Spearheaded by the Office of Science and Technology Policy, the workshops will be rolled out over the next few months (May to July) and will cover topics including implications in law and government, as well as the social and economic impacts. Workshop co-hosts include academic and non-profit institutions, as well as the National Economic Council. In addition, a new National Science and Technology Council (NSTC) subcommittee on machine learning and artificial intelligence will meet for the first time next week. The NSTC is currently working to leverage AI and machine learning technology in a variety of government services.

Your face is now your passport. American Airlines is joining a growing list of companies that are beginning to offer facial recognition as a means of identification. The world's largest airline will now let some passengers simply scan their face to board their flight at the Los Angeles International Airport. American Airlines is joining a growing list of companies that are beginning to offer facial recognition as a means of identification. American Airlines is rolling out facial recognition cameras as part of a 90-day test to identify people before they get on board their flights from LAX's Terminal 4. The pilot program, which launched on Wednesday, came about as a result of partnership with digital security company Gemalto.

Artificial intelligence systems and creators now need direct intervention by governments and human rights organizations, according to a recent report from researchers at Google, Microsoft and others at AI Now. In the report (PDF) published this week, the New York University-based organization (with Microsoft Research and Google-associated members) indicates that AI-based tools are being used with little regard for possible ill effects it can have on the general public. It would've been a different story if it was happening in controlled, regulated environment here and there; instead these untested, undocumented AI systems are being put to work in places where they can severely affect millions of people. These systems are causing real damage, and not only are there no systems in place to stop them, but few to even track and quantify that harm. Think border patrol, entire school districts and police departments, and so on.

In July, Microsoft President Brad Smith published a blog post discussing some of the challenges and risks surrounding facial recognition technology and how his company plans to address them. He said Microsoft was working on developing a set of guiding principles to direct the development of its own technology and this week, he shared what those principles are. In a new blog post, Smith described six principles that he says will govern Microsoft's work while it continues to advocate for legislation that will more concretely guide this technology as a whole. Smith said the company sought input from employees, customers, public officials, academics and other groups as it designed these principles. Going forward, the company will develop and deploy facial recognition technology in a way that aims "to treat all people fairly" and it will "document and clearly communicate the capabilities and limitations of facial recognition technology" as it does.

NASA's InSight lander arrived on Mars last week and now it's prepping for the next stage of its mission. The lander is equipped with a nearly six-foot-long arm, which will be used to lift sensitive scientific instruments off of the lander and place them on the Martian surface -- the first time a robot has been used to do so on another planet. Thus, the arm plays a really key role in this next step of the InSight mission, as placement will be crucial for the mission's overall success. But new images from InSight show the lander is up for the challenge, as the arm is working and has already started to collect images of its surroundings, information that scientists will use to determine where InSight's instruments should be placed. "Today we can see the first glimpses of our workspace," Bruce Banerdt, the mission's principal investigator, said in a statement.

The AI revolution is coming, and both Canada and France want to make sure we're approaching it responsibly. Today, the countries announced plans for the International Panel on Artificial Intelligence (IPAI), a platform to discuss "responsible adoption of AI that is human-centric and grounded in human rights, inclusion, diversity, innovation and economic growth," according to a mandate from the office of Canadian Prime Minister Justin Trudeau. It's still unclear which other countries will be participating, but Mounir Mahjoubi, France's secretary of state for digital affairs, says it'll include both G7 and EU countries, Technology Review reports. It won't just be politicians joining the conversation. France and Canada plan to get the scientific community involved, as well as industry and civil society experts.

Classifiers fail to classify correctly input images that have been purposefully and imperceptibly perturbed to cause misclassification. This susceptability has been shown to be consistent across classifiers, regardless of their type, architecture or parameters. Common defenses against adversarial attacks modify the classifer boundary by training on additional adversarial examples created in various ways. In this paper, we introduce AutoGAN, which counters adversarial attacks by enhancing the lower-dimensional manifold defined by the training data and by projecting perturbed data points onto it. AutoGAN mitigates the need for knowing the attack type and magnitude as well as the need for having adversarial samples of the attack. Our approach uses a Generative Adversarial Network (GAN) with an autoencoder generator and a discriminator that also serves as a classifier. We test AutoGAN against adversarial samples generated with state-of-the-art Fast Gradient Sign Method (FGSM) as well as samples generated with random Gaussian noise, both using the MNIST dataset. For different magnitudes of perturbation in training and testing, AutoGAN can surpass the accuracy of FGSM method by up to 25\% points on samples perturbed using FGSM. Without an augmented training dataset, AutoGAN achieves an accuracy of 89\% compared to 1\% achieved by FGSM method on FGSM testing adversarial samples.

We survey distributed deep learning models for training or inference without accessing raw data from clients. These methods aim to protect confidential patterns in data while still allowing servers to train models. The distributed deep learning methods of federated learning, split learning and large batch stochastic gradient descent are compared in addition to private and secure approaches of differential privacy, homomorphic encryption, oblivious transfer and garbled circuits in the context of neural networks. We study their benefits, limitations and trade-offs with regards to computational resources, data leakage and communication efficiency and also share our anticipated future trends.