Developer Community Tech Week is back with us again, so check out this week’s important news for us developers.

In one minute!

  • Google Live captioning is now available in Google Chrome
  • Artificial intelligence can debate humans: good opening remarks, but losing to professional debaters
  • Taichi (Taichi) 0.7.15 release
  • Jingdong Technology “Zhizhen chain” to get through the provincial cold chain market, cloud tracing to promote industry reform
  • Cloud Network Industry Promotion Phalanx releases the industry’s first White Paper on Cloud Network Industry Development
  • China’s Ict to launch ‘Trusted Digital Evaluation System’
  • CVPR 2021 | Single-sheet depth image super-resolution Method based on Cross-task scene structure knowledge transfer
  • CVPR 2021 | AdCo Based Contrastive Learning

News of the industry

1. Live captioning is now available in Google Chrome

Live captioning will launch in Chrome 89 on March 18. The feature, which automatically captions audio and video using machine learning technology, is designed to make the web more accessible to people who are deaf or hard of hearing.

Previously, at Google I/O 2019, Google first showed off its amazing real-time captioning accessibility. The feature was introduced first on Pixel phones with an updated Android 10, and later on many non-Pixel phones — including the Galaxy S20 series, the OnePlus 8 Series, the OnePlus Nord, and more. To be sure, the live captioning feature now appears in Settings > Advanced > Accessibility. If you don’t see it on Chrome 89 you may need to restart Chrome.

2. Artificial intelligence can debate humans: Great opening remarks, but losing to professional debaters

A study published in the academic journal Nature describes an autonomous agent that can engage in competitive debate with humans. The human debater was judged the winner, but the author believes that ARTIFICIAL intelligence may be able to participate in complex human activities. In the study, IBM engineer Noam Slonim and colleagues describe an autonomous system called Project Debater (” Debater “) that can engage in meaningful debate with humans. The system can organize its own opening lines and rebuttals by scanning an archive of 400 million news stories and Wikipedia pages. IBM believes ai debaters can help people reason, build sound arguments and make better decisions. Potential applications for Project Debater include financial advisers, lawyers, public policy makers, student assistants and corporate decision makers.

3.Taichi (Taichi) 0.7.15 release

Taichi is a programming language designed for high performance computer graphics. It is deeply embedded in Python, and its just-in-time compiler shifts computationally intensive tasks to multi-core cpus and massively parallel Gpus. This 0.7.15 release has the following major updates:

  • [refactor] Shift TypedConstants to Taichi/IR /type;
  • [refactor] moved ASTBuilder and FrontenContext to frontend_ir;
  • [ir] [Transforms] Added assertions that indexes don’t overflow in debug mode.

For more updates please visit:

Github.com/taichi-dev/…

4. Jingdong Technology “Zhizhen chain” opens up the provincial cold chain market and pushes forward the industry reform by tracing to the source on the cloud

With the globalization of cold-chain commodity circulation, hundreds of millions of vegetables, fruits, eggs, milk and seafood are circulating around the world every day. With the normalization of the epidemic, tracing the origin of cold-chain commodities becomes particularly important. Recently, with jingdong cloud as the base, jingdong technology based on the “Zhizhen chain” anti-counterfeiting traceability platform, joint jingdong logistics, jingdong fresh to create the “jingdong cold chain traceability platform” officially launched. At present, the platform has covered all categories of fresh fruits and vegetables, meat and poultry, seafood and other fresh products, serving more than 4 million fresh goods from nearly 300 merchants. Through the platform, not only can realize the goods from the origin to the consumer can be traced, and the future can realize the domestic provinces and cities back to each other, code to the end.

5. Cloud Network Industry Promotion Phalanx publishes the industry’s first White Paper on Cloud Network Industry Development

On March 18, the “2021 Cloud Tube and Cloud Network Conference” hosted by China Academy of Information and Communications Technology (CAICT) and China Communications Standardization Association was successfully held in Beijing. The conference released the industry’s first White Paper on cloud Network Industry Development (hereinafter referred to as “White Paper”), which was led by THE China Academy of Information and Communication Technology and jointly compiled by a number of enterprises in the industry. The white paper describes in detail the status quo, technical characteristics, application scenarios and practices of China’s cloud network industry, as well as the future development trend. It answers many focal issues of cloud network development that the industry is concerned about, and points out that cloud network will become the key development direction of cloud computing in the future.

6. The Chinese Academy of Information and Communication (CAIT) will soon release a “Trusted digital Evaluation system”

The digital wave is coming. This year’s government work report clearly points out that we should “accelerate digital development, create new advantages of the digital economy, jointly promote digital industrialization and industrial digital transformation, and accelerate the pace of building a digital society.” At the recently concluded National Two sessions, “digital transformation” has also become a hot word of concern among deputies and committee members.

As more companies and industries accelerate their digital transformation, pain points are starting to emerge. Where should enterprises start with digital transformation? Are there “holes” in the digitalization programs that have been implemented? How to choose a truly credible digital solution? In the upcoming “2021 Digital Transformation Development Summit Forum” on March 31, China Academy of Information and Communication Technology (HEREINAFTER referred to as “China Academy of Information and Communication”) will officially release the “Trusted digital Evaluation System” to help enterprises find the answer to the above questions.

The academic frontier

1.CVPR 2021 | Single-sheet depth image super-resolution method based on Cross-task scene structure knowledge transfer

Sensing system for the study on depth scene depth problem such as low resolution images and details lost, break through the existing limitations of scene depth recovery method based on the color guide, which at the same time in the training and testing phase need high-resolution color images and drop quality, and the depth image as the input to estimate depth image quality (in the actual test environment, High-resolution color auxiliary information from the same perspective is not readily available). A single scene depth image super-resolution method based on cross-task scene structure knowledge transfer is proposed for the first time. In the training stage, scene structure information is distilled from color images to assist in improving depth restoration performance, while in the testing stage, only a single degraded depth image is provided as input to achieve depth image reconstruction.

* Paper links:

Faculty.dlut.edu.cn/yexinchen/z…

2.CVPR 2021 | AdCo Contrastive learning based on antagonism

In the field of self-supervised learning, contrastive learning (CONTRastive learning) has achieved obvious advantages in downstream classification detection and tasks. Among them, how to make full use of negative samples to improve learning efficiency and learning effect has always been a direction worth exploring. In this paper, for the first time, a new approach of end-to-end learning negative samples was proposed to achieve SOTA in ImageNet and downstream tasks.

AdCo was able to achieve the same accuracy with just 8,196 negative samples (one-eighth of the negative sample size of MoCo V2). At the same time, these directly trainable negative samples can still achieve similar results with the same number of Prediction MLP parameters as in BYOL. This shows that in the era of self-supervised learning, contrast learning still has a series of advantages such as high learning efficiency, stable training and high accuracy through learnability of negative samples.

* Links to papers: arxiv.org/abs/2011.08…

The above information is from the network, edited by the “JINGdong Technology Developer” public account, does not represent the position of Jingdong Technology

Recommended reading

  • The world’s first artificial intelligence earthquake monitoring system was launched; AAAI RMB 2021 | using depth study on city sales forecast
  • Jingdong cloud new generation from the cloud server online in April; COLING 2020 | Bidirectional cognitive thinking Network for machine reading Comprehension
  • Blockbuster! Jingdong cloud since the fourth generation cloud host release; Foreign physicists have developed assembly language for quantum computers

Welcome to [JINGdong Technology] to learn about the developer community

More wonderful technical practice and exclusive dry goods analysis

Welcome to “JINGdong Technology Developer” public account