A team of researchers working at Seoul National University has developed a soft robot chameleon that can change its colors in real time to match its background. The group describes their multi-layer skin design and possible uses for it in their paper published in the journal Nature Communications.
Read more at techxplore.com
A pair of Facebook engineers have announced the development of Time Cards. Time Cards are PCIe cards that can be used in x86 architecture machines to serve as a timekeeping device. When the Time Card is plugged into a server, it becomes, in Facebook terminology, a time appliance. The cards work by listening in on radio waves sent by GNSS navigation satellites that are coupled with oscillators.
Read more at engineering.fb.com
Artificial intelligence company OpenAI has announced a system that translates natural language to programming code called Codex. The system is being released as a free API, at least for the time being. It builds on Copilot, a tool for use with Microsoft's GitHub code repository. As a general-purpose programming tool, Codex can be used for virtually any programming task.
Read more at openai.com
Researchers in SFU's Computational Photography Lab hope to give computers a visual advantage that we humans take for granted. The researchers are now making it possible to create detailed 3D renderings that look realistic to a human eye. They say their work could be used by artists and content creators.
Read more at techxplore.com
Unidentified Video Objects (UVO) is a benchmark to facilitate research on open-world segmentation. It aims to detect, segment, and track all objects exhaustively in a video. While machines typically must learn specific object concepts in order to recognize them, UVO can help them mimic humans' ability to detect unfamiliar visual objects.
Read more at ai.facebook.com
SoundStream is the first neural audio codec to work on speech and music, while being able to run in real-time on a smartphone CPU. It is able to deliver state-of-the-art quality over a broad range of bitrates with a single trained model. The main technical ingredient of SoundStream is a neural network, consisting of an encoder, decoder and quantizer.
Read more at ai.googleblog.com
Facebook has won first place at the International Conference on Spoken Language Translation. We used our latest speech-to-text translation techniques, including transfer learning across modalities, tasks, and languages. Our system was also used to power subtitles from Portuguese to English for Facebook's Connectivity Research Workshop.
Read more at ai.facebook.com
Facebook AI has built a family of visual recognition models that can train entirely from scratch on a video recognition data set. When presented with an image or video, MViT models identify the objects present in the image or the actions being performed in the video. In the future, the application of these models to videos and images in the wild may help machines better understand the real world.
Read more at ai.facebook.com
Grammatical error correction attempts to model grammar and other types of writing errors. There is very limited training data available for GEC, even for high-resource languages like English. Researchers used tagged corruption models to generate a new 200M sentence dataset in order to provide researchers with realistic pre-training data.
Read more at ai.googleblog.com
Current generation quantum processors still have high operational error rates of 10-3 per operation. Google has developed a new technique for reducing the amount of correlated errors in our physical qubits. Quantum computers of the future will have to use quantum error correction (QEC).
Read more at ai.googleblog.com
Researchers and even hobbyists can build more intelligent real-world robots with droidlet. Droidlet is a modular, heterogeneous embodied agent architecture that sits at the intersection of natural language processing, computer vision, and robotics. It enables researchers to easily build agents that can accomplish complex tasks either in the real world or in simulated environments like Minecraft or Habitat.
Read more at ai.facebook.com
RaggedTensors TF-Ranking, an open-source library for developing scalable neural learning-to-rank (LTR) models, is now working with Keras. The latest version of TF-ranking contains a long list of advances, including a flexible ModelBuilder, DatasetBuilder, and Pipeline to train the model with the provided dataset. We encourage everyone to try out the latest version and follow this introductory example for a hands-on experience.
Read more at ai.googleblog.com
An accurate record of building footprints is important for a range of applications, from population estimation and urban planning to humanitarian response and environmental science. The Open Buildings dataset contains the locations and footprints of 516 million buildings with coverage across most of the African continent. It will support several practical, scientific and humanitarian applications, ranging from disaster response or population mapping to planning services such as new medical facilities.
Read more at ai.googleblog.com
AttentiveNAS and AlphaNet significantly improve the accuracy of so-called supernets. These approaches outperform existing supernet training methods. They deliver state-of-the-art results on the ImageNet data set and achieve greater than 80 percent accuracy with only 444 MFLOPS.
Read more at ai.facebook.com
A recent paper demonstrated the potential of a new algorithm that could produce agents that could play Atari 2600 games very effectively. To train Rainbow on the full suite of 57 games required around 34,200 GPU hours (or 1425 days) in order to provide convincing empirical performance statistics.
Read more at ai.googleblog.com
Semi-supervised learning requires accurately labeled data, the collection of which is often labor intensive. This shortcoming motivated the development of Noisy Student Training, a new approach that worked well in the low-data regime but not in the high-data one.
Read more at ai.googleblog.com
BRAX is a new physics simulation engine that matches the performance of a large compute cluster with just a single TPU or GPU. The engine is designed to efficiently run thousands of parallel physics simulations alongside a machine learning algorithm on a single accelerator and scale millions of simulations seamlessly.
Read more at ai.googleblog.com
Researchers at the National Institute of Advanced Industrial Science and Technology introduce a new lithium battery cell design. The anode-free design could help to increase the energy density and safety of lithium-metal batteries. Researchers hope to use Li2O as a sacrificial agent.
Read more at techxplore.com
A research team at the University of Cordoba has developed and evaluated models for the prediction of solar radiation in southern Spain and North Carolina (USA) The work has been featured in the journal Applied Energy, which was published on Monday. Researchers say their findings could lead to a new way of measuring solar radiation.
Read more at techxplore.com
Robotic systems typically monitor their own movements and surroundings in different ways. Researchers at Purdue University and University of Virginia have developed a system that can detect frame-related inconsistencies. The system, PHYSFRAME, could improve the effectiveness and reliability of frame translation practices in robotics.
Read more at techxplore.com
