Collaborating Authors

Social Media

[R] Generative Adversarial Transformers (2103.01209)


Abstract: We introduce the GANsformer, a novel and efficient type of transformer, and explore it for the task of visual generative modeling. The network employs a bipartite structure that enables long-range interactions across the image, while maintaining computation of linearly efficiency, that can readily scale to high-resolution synthesis. It iteratively propagates information from a set of latent variables to the evolving visual features and vice versa, to support the refinement of each in light of the other and encourage the emergence of compositional representations of objects and scenes. In contrast to the classic transformer architecture, it utilizes multiplicative integration that allows flexible region-based modulation, and can thus be seen as a generalization of the successful StyleGAN network. We demonstrate the model's strength and robustness through a careful evaluation over a range of datasets, from simulated multi-object environments to rich real-world indoor and outdoor scenes, showing it achieves state-of-the-art results in terms of image quality and diversity, while enjoying fast learning and better data-efficiency.

AI brings photos of Amelia Earhart, Albert Einstein, Marie Curie to life


Artificial intelligence is turning old pictures of people into short, animated clips that show them moving and blinking. The feature, called Deep Nostalgia, comes from genealogy company MyHeritage. It uses machine learning to create facial expressions and movements that look super realistic, Tom's Guide reported Tuesday. In a blog post, MyHeritage shared social media posts from users who were thrilled to see their loved ones who'd passed come to life, if only for a few moments. The clips show the people in black-and-white or faded photos tilting their heads and looking around.

[D] Simple Questions Thread December 20, 2020


Hi, I'm working in a museum, currently trying to optically characterize a big historic lens. Unfortunately, it is mounted in a device which can't really be taken apart (issues of conservation), so conventional methods are rather hard to do. I've been loosely following the advances in neural network based approaches ("Two minute papers" kinda stuff) and was wondering if anyone has already realized a solution to my problem using machine learning or similar techniques. That is: Print out a defined optical pattern (like a QR code), "wave" it on one side of the lens and record the image with a camera on the other to get a 3D model of the lens in the end. In my head, it should be possible to train a network using conventional light simulation of randomly generated glass bodies.

Servion Global Solutions on LinkedIn: Lyra: A New Very Low-Bitrate Codec for Speech Compression


The trade-off between voice quality and bandwidth remains critical for contact centers. Voice codecs continue to evolve, and with AI can now deliver good quality with bandwidth as low as 3k. This will help contact centers contain cost and deliver a great customer experience - there is nothing worse than bad audio, as we all know.

Valheim Is Changing How We Play Survival Games


With four million games sold on Steam Early Access in three weeks and overwhelmingly positive reviews, Valheim became a commercial and critical darling at an almost unprecedented speed. The viking survival game, developed by a small Swedish team at Iron Gate Studio, might appear to be an overnight success, but CEO Richard Svensson has been directly communicating with the gaming community about this project for years. In September of 2017, Svensson posted a video to his personal YouTube page that captures what seems to be the infancy stages of Valheim and demonstrates Svensson's philosophy of public communication concerning the game's ongoing development. When the game's working title was changed from Fejd (Swedish for "feud") to Valheim in 2018, Svensson noted the switch in the YouTube comments section. Video game studios can often be tight lipped during the development process, but Iron Gate Studio took the opposite approach, directly listened to what their players wanted, and built a vibrant community on Discord.

The New Morality of Debt – IMF F&D


Throughout history, society has debated the morality of debt. In ancient times, debt--borrowing from another on the promise of repayment--was viewed in many cultures as sinful, with lending at interest especially repugnant. The concern that borrowers would become overindebted and enslaved to lenders meant that debts were routinely forgiven. These concerns continue to influence perceptions of lending and the regulation of credit markets today. Consider the prohibition against charging interest in Islamic finance and interest rate caps on payday lenders--companies that offer high-cost, short-term loans.

Data Discovery Platforms and Their Open Source Solutions


In the past year or two, many companies have shared their data discovery platforms (the latest being Facebook's Nemo). Based on this list, we now know of more than 10 implementations. I haven't been paying much attention to these developments in data discovery and wanted to catch up. By the end of this, we'll learn about the key features that solve 80% of data discoverability problems. We'll also see how the platforms compare on these features, and take a closer look at open source solutions available.

ARGO Vision on LinkedIn: #AI #AI #artificialintelligence


MediaPipe #AI from Google updated for fitness/yoga practitioners MediaPipe Pose is a ML solution for high-fidelity body pose tracking, inferring 33 3D landmarks on the whole body (or 25 upper-body landmarks) from RGB video frames. Highlights: Trained specifically for fitness/yoga activities Updated models enable custom pose classification Real-time performance on mobile and browser The pipeline is implemented as a MediaPipe graph that uses a pose landmark subgraph from the pose landmark module and renders using a dedicated pose renderer subgraph. Link to the original news and to the new updated models in the first comment Want to know more about #AI and our projects? Follow ARGO Vision or ping Alessandro Ferrari We See The Future No Magic.

[D] support set concept in Machine learning and one shot learning


In many papers like " Matching Networks for One Shot Learning " support set of images play a big role. However, I have trouble understanding what support set is? My understanding is- If I have trained my model on classes-"cat" and "dog". So during training my support set will have images of cats and dogs. But when at time of testing I wanted to see if image of " Horse" is there in image of support set or not. So can my support set have new image class(horse here). During testing time can I use this kind of 1 shot network to find my test image in collection of support image given while inference?

Who Gets First Amendment Protections These Days, Anyway?


On a recent episode of Amicus, Dahlia Lithwick talked with Jameel Jaffer, executive director of the Knight First Amendment Institute at Columbia University, to unpack how the scope of the First Amendment continues to grow even as it fails in the face of so many of the free speech issues we face today. A portion of their conversation, which has been edited and condensed for clarity, has been transcribed below. Dahlia Lithwick: I think I've had a Post-it note pinned to my screen saying, "Do a First Amendment show" for three years. It sweeps in every news cycle. From the Facebook "Supreme Court," your own litigation around Trump's tweets, cancel culture, the speech defenses that came up at the impeachment trial--I think of the First Amendment as a framework that governs all of those things. As you suggested to me, when we were thinking about this show, the First Amendment is "everywhere but nowhere."