My name is Kris Coratti. Thank you for joining us on this very rainy morning. I'm glad you all made it out. We are going to have a fascinating series of discussion this morning on artificial intelligence. This is the latest in our ongoing event series that we call "Transformers." And our speakers this morning are going to explore the regulatory questions around this technology. They going to look at how AI is reshaping the way we live and work. And they're going to discuss how to make sure this technology is used responsibly in the future. Before we begin, I just want to quickly thank our presenting sponsor for this even, Software.org, And so now I'd like to go ahead and welcome to the stage The Washington Post's Tony Romm and Senators Maria Cantwell and Todd Young. And for those who don't know, Senator Cantwell is a Democrat from Washington State. Both are members of a Senate commerce committee which touches on artificial intelligence and many tech issues that we'll talk about today.

In recent years, with the popularization of deep learning frameworks and large datasets, researchers have started parallelizing their models in order to train faster. This is crucially important, because they typically explore many hyperparameters in order to find the best ones for their applications. This process is time consuming and, consequently, speeding up training improves productivity. One approach to parallelize deep learning models followed by many researchers is based on weak scaling. The minibatches increase in size as new GPUs are added to the system. In addition, new learning rates schedules have been proposed to fix optimization issues that occur with large minibatch sizes. In this paper, however, we show that the recommendations provided by recent work do not apply to models that lack large datasets. In fact, we argument in favor of using strong scaling for achieving reliable performance in such cases. We evaluated our approach with up to 32 GPUs and show that weak scaling not only does not have the same accuracy as the sequential model, it also fails to converge most of time. Meanwhile, strong scaling has good scalability while having exactly the same accuracy of a sequential implementation.

In this project, we developed a deep learning system applied to human retina images for medical diagnostic decision support. The retina images were provided by EyePACS (Eyepacs, LLC). These images were used in the framework of a Kaggle contest (Kaggle INC, 2017), whose purpose to identify diabetic retinopathy signs through an automatic detection system. Using as inspiration one of the solutions proposed in the contest, we implemented a model that successfully detects diabetic retinopathy from retina images. After a carefully designed preprocessing, the images were used as input to a deep convolutional neural network (CNN). The CNN performed a feature extraction process followed by a classification stage, which allowed the system to differentiate between healthy and ill patients using five categories. Our model was able to identify diabetic retinopathy in the patients with an agreement rate of 76.73% with respect to the medical expert's labels for the test data.

The problem with AI is that while it's relatively easy to define the "A", the "I" remains elusive. We don't know what our own intelligence is, nor how we generate our familiar conscious experience, so it's tricky to know how we might create an artificial consciousness, or indeed recognise it if we did. Algorithms can knit together plausible conversation by sampling enormous numbers of exchanges between humans, but they have no greater understanding of those exchanges than would an enormous set of punch cards speaking through a bellows and a brass trumpet. The old Turing test now looks sadly inadequate. A machine-learning program might well counterfeit human speech and yet fail to recognise a snow leopard standing on green grass because the image contains no actual snow, and therefore the cat does not meet the definition.

The problem with AI is that while it's relatively easy to define the "A", the "I" remains elusive. We don't know what our own intelligence is, nor how we generate our familiar conscious experience, so it's tricky to know how we might create an artificial consciousness, or indeed recognise it if we did. Algorithms can knit together plausible conversation by sampling enormous numbers of exchanges between humans, but they have no greater understanding of those exchanges than would an enormous set of punch cards speaking through a bellows and a brass trumpet. The old Turing test now looks sadly inadequate. A machine-learning program might well counterfeit human speech and yet fail to recognise a snow leopard standing on green grass because the image contains no actual snow, and therefore the cat does not meet the definition.

The World Cup 2018 is all over. Germany was kicked out in the group stages, Brazil was beaten by Belgium, football didn't come home to England, Croatia with its population of four million people reached the final for the first time ever, only to lose to France in the end. Beyond being glued to our screens to watch the action on pitch, we've been looking at what newsrooms are doing off-pitch to cover the competition… with automation and artificial intelligence. Fox Sports (US) teamed up with IBM Watson to make AI-powered highlight videos, French publication Le Figaro created automated visual summaries, and The Times (UK) launched its very own World Cup Alexa Skill. The US didn't qualify for the World Cup this year, but that didn't stop Fox Sports from airing all 64 matches and teaming up with IBM Watson to create the World Cup highlight machine.

NSFW Artificially intelligent software is used more and more to automatically detect and ban nude images on social networks and similar sites. However, today's algorithms and models aren't perfect at clocking racy snaps, and a lot of content moderation still falls to humans. Enter an alternative solution: use AI to magically draw bikinis on photos to, er, cover up a woman's naughty bits. A group of researchers from the Pontifical Catholic University of Rio Grande do Sul, Brazil, have trained generative adversarial networks to perform this very act, and automatically censor nudity. In a paper for the IEEE International Joint Conference on Neural Networks (IJCNN) in Rio de Janeiro earlier this month, the eggheads presented some of their results.

Increased information sharing through short and long-range skip connections between layers in fully convolutional networks have demonstrated significant improvement in performance for semantic segmentation. In this paper, we propose Competitive Dense Fully Convolutional Networks (CDFNet) by introducing competitive maxout activations in place of naive feature concatenation for inducing competition amongst layers. Within CDFNet, we propose two architectural contributions, namely competitive dense block (CDB) and competitive unpooling block (CUB) to induce competition at local and global scales for short and long-range skip connections respectively. This extension is demonstrated to boost learning of specialized sub-networks targeted at segmenting specific anatomies, which in turn eases the training of complex tasks. We present the proof-of-concept on the challenging task of whole body segmentation in the publicly available VISCERAL benchmark and demonstrate improved performance over multiple learning and registration based state-of-the-art methods.

When Meghan Cruz says "Hey, Alexa," her Amazon smart speaker bursts to life, offering the kind of helpful response she now expects from her automated assistant. With a few words in her breezy West Coast accent, the lab technician in Vancouver gets Alexa to tell her the weather in Berlin (70 degrees), the world's most poisonous animal (a geography cone snail) and the square root of 128, which it offers to the ninth decimal place. But when Andrea Moncada, a college student and fellow Vancouver resident who was raised in Colombia, says the same in her light Spanish accent, Alexa offers only a virtual shrug. She asks it to add a few numbers, and Alexa says sorry. She tells Alexa to turn the music off; instead, the volume turns up.

Chinese ride-sharing giant Didi Chuxing will explore ways to export its artificial intelligence and strength in big data to become a global enterprise, President Jean Liu told the Nikkei Asian Review in an interview. The company, known for its ride-hailing platform, on Thursday announced a plan for its joint venture with Japanese tech giant SoftBank to bring taxi hailing services to Japan staring this autumn. That same time day, however, SoftBank Group founder Masayoshi Son referred to the Japanese government as "stupid" for restricting ride-sharing and other services. Still, Didi and SoftBank hope their partnership with Japanese taxi companies using Didi algorithms to match drivers with customers will improve vehicle occupancy rates. Drivers will not have to pay fees for the service.