Multiple Pretext-T ask for Self-Supervised Learning via Mixing Multiple Image Transformations Shin'ya Y amaguchi, Sekitoshi Kanai, Tetsuya Shioda, Shoichiro Takeda NTT Tokyo, Japan {shinya.yamaguchi.mw,sekitoshi.kanai.fu,tetsuya.shioda.yf,shoichiro.takeda.us}@hco.ntt.co.jp Abstract Self-supervised learning is one of the most promising approaches to learn representations capturing semantic features in images without any manual annotation cost. T o learn useful representations, a self-supervised model solves a pretext-task, which is defined by data itself. Among a number of pretext-tasks, the rotation prediction task (Rotation) achieves better representations for solving various target tasks despite its simplicity of the implementation. However, we found that Rotation can fail to capture semantic features related to image textures and colors. T o tackle this problem, we introduce a learning technique called multiple pretext-task for self-supervised learning (MP-SSL), which solves multiple pretext-task in addition to Rotation simultaneously. In order to capture features of textures and colors, we employ the transformations of image enhancements (e.g., sharpening and solarizing) as the additional pretext-tasks. MP-SSL efficiently trains a model by leveraging a Frank-W olfe based multi-task training algorithm. Our experimental results show MP-SSL models outperform Rotation on multiple standard benchmarks and achieve state-of- the-art performance on Places-205. 1. Introduction Convolutional neural networks (CNNs) [27, 16, 44] are widely adopted to solve many target tasks in applications of computer vision such as object recognition [30], semantic segmentation [4], and object detection [42]. However, these successes depend on supervised training of CNNs with the vast amount of labeled data [43], which is expensive and impractical because of the manual annotation cost. Since the cost of labeled data limits the practical applications of CNNs, a number of researches focus on the training techniques to alleviate the requirement of many labeled data; the techniques include transfer learning, semi-supervised learning, and self-supervised learning . A demonstration describing our motivation to modify self-supervised learning by predicting rotations of images (Rotation).

Star Wars-style lasers could help future space missions to accurately pinpoint the location of space junk in Earth's orbit and avoid deadly collisions. Millions of pieces of space junk are currently whizzing around the planet at around 20,000 miles per hour -- providing a threat to future rockets and space craft. However a new system based on neural networks has provided a better way to keep track of these hazardous objects with telescopes. The approach can detect debris as small as 3 feet wide and allow courses to be plotted to avoid them. Star Wars-style lasers could help future space missions to accurately pinpoint the location of space junk in Earth's orbit and avoid deadly collisions.

'Hidden numbers' have a profound effect on our lives - from online dating and falling in love to manipulating our decisions through targeted advertising and fake news. In this year's Royal Institution Christmas Lectures, Mathematician and author Hannah Fry will explore how maths, numbers and data patterns can help explain make sense of life's human experiences. She will also be asking the big ethical questions about the data (and trust) that we gift companies such as Netflix and Amazon and how they use this information to develop new unprecedented levels of control. 'I'd argue that maths is the most important idea that humans have ever had. It's the foundation of science,' said Dr Fry.

KUALA LUMPUR – Disney has cut a lesbian kiss from the latest "Star Wars" movie, Singapore's media regulator said on Tuesday, so that more children can watch it. The two minor female characters embrace but do not kiss in the version of "The Rise of Skywalker" shown in Singapore, local media said, as the ninth film in the celebrated science-fiction series rakes in millions from loyal fans. "The applicant has omitted a brief scene which under the Film Classification Guidelines would require a higher rating," a spokeswoman from Singapore's Infocomm Media Development Authority said. Disney, which owns the "Star Wars" production company Lucasfilm, did not respond to a request for comment on its decision to cut the scene from the last installment of the second highest-grossing movie franchise of all time. It concludes a story that began in 1977, when filmmaker George Lucas introduced a young hero named Luke Skywalker and delighted audiences with a galaxy of robots, furry warriors known as Wookiees and a host of other eclectic characters.

Teradyne, Inc. (NasdaqGS:TER) and AI firm Enflame Tech have announced that the latter's Deep Thinking Unit (DTU), which is "designed to accelerate deep learning deployment for data center training", is being supported by the former's UltraFLEX test system. "The market for AI chips is very competitive and time sensitive. The design complexity and device power requirements make device bring-up and volume production a significant challenge. For AI startup companies like Enflame, achieving the fastest time to market with best quality is crucial and with the UltraFLEX our customers can reliably meet device capability and schedule requirements. Teradyne's partnership with Enflame is another example of Teradyne's success in providing the technical capability and support required to help ensure emerging AI IC developers bring a quality device to market quickly."

Let's say we have a simple "autoencoding transformer" architecture: Now we ask about the properties of Z - the latent representation of the data, after the model is trained. Will Z differ between the two objectives? Will it capture different information? Which loss will preserve more information in Z? Does this have an obvious interpretation?

We discuss why James wrote this book 3 years ago now, how far away he really thinks we are from artificial human intelligence, the warning bells recently being sounded about artificial intelligence, and why he thinks there will not be another AI winter. Our guest today is James Barrat author of the book "Our final Invention" Artificial Intelligence and the end of the Human Era". It's good to be here. Kathleen Walch: [00:00:39] Great, I'd like to get started by having you introduce yourself to our listeners and to tell us a little bit about your book and also what additional things that you're doing in the field of AI and let's go from there. I got into artificial intelligence, or the study of artificial intelligence, and the critique of AI because I made a film about 17 years ago now about artificial intelligence. I interviewed Ray Kurzweil and Rodney Brooks and Arthur C. Clarke among others … and Ray Kurzweil of course who is now chief engineer at Google and the Google brain project.