In 2021, the trend of intelligent data collection and annotation industry of four major industries is predicted; Tsinghua University proposes depth aligned clustering for new intention map discovery

Time:2021-4-21

Developer community technology weeklyI’m meeting you again. Let’s take a look at the important news that we developers should pay attention to this week.

  • Digital currency 50000 200 red packets, Jingdong appointment lottery!
  • Cloud survey data: four trends forecast of AI data acquisition and labeling industry in 2021
  • Google cloud hard disk will integrate enterprise and personal services, and merge desktop client software
  • Microsoft released win10 cloud configuration: flexible and secure deployment of Win32 applications, etc
  • Faster and clearer! NVIDIA realizes SDF real-time rendering for the first time, and the speed is improved by 3 orders of magnitude
  • In depth learning performance measurement of NVIDIA A100: the training speed can reach 3.5 times of V100
  • AAAI 2021 Tsinghua proposes depth aligned clustering for new idea map discovery
  • ACM mm: 3D dense connection network based on time frequency space characteristics of emotional EEG signals

Industry news     News

1. Digital goods 50000 200 red packets, Jingdong appointment lottery!

In 2021, the trend of intelligent data collection and annotation industry of four major industries is predicted; Tsinghua University proposes depth aligned clustering for new intention map discovery

Digital RMB is coming to Beijing! Open the “Jingdong” app,Search for “digital RMB”,From February 7 to February 8, 200 yuan digital currency red envelope can be extracted from the registration number (subject to GPS positioning) in Beijing,The limit is 50000. The lottery results were announced on February 10,Red packets can be used from February 10 to February 17, and can be consumed online in Beijing and Tokyo. If the lottery wins, Jingdong willPush the winning information through SMS.

Supporting banks: ICBC, ABC, BOC, CCB, BOCOM, postal savings.

2. Forecast of four trends of AI data acquisition and labeling industry in 2021

In 2021, the trend of intelligent data collection and annotation industry of four major industries is predicted; Tsinghua University proposes depth aligned clustering for new intention map discovery

The cloud measurement data of artificial intelligence data acquisition enterprises was released recentlyFour trends forecast of artificial intelligence data acquisition and labeling industry in 2021. According to cloud measurement data, in the future, high-precision data will pursue hot spots in the intelligent training stage of adult workers, with the characteristics of stable main demand side, and the stock market will grow steadily; With the further expansion of the data demand of AI for long tail scenarios, 3 – Within five years, scenario data will have a broader incremental space and become a new engine for accelerating the development of the industry; At the same time, “underlying technology” + “Service capability” will become more and more important until it becomes the core competition point. Artificial intelligence needs service providers who can provide integrated data solutions.

3. Google cloud hard disk will integrate enterprise and personal services, and merge desktop client software

According to foreign media, Neowin,Google plans to integrate its previous cloud hard disk services for enterprises with personal cloud hard disk services,In order to solve the problem of user confusion. At present, Google drive has two desktop clients, which are used for enterprise users’ cloud disk and personal users’ synchronization respectively. Yesterday, Google announced that it would integrate the two services later this year, in order to make it easier for business managers to synchronize file distribution and avoid confusion among employees. In addition, Google will design a new Google drive desktop client to unify the two services.

4. Microsoft released win10 cloud configuration: flexible and secure deployment of Win32 applications

In 2021, the trend of intelligent data collection and annotation industry of four major industries is predicted; Tsinghua University proposes depth aligned clustering for new intention map discovery

February 3 news   Microsoft announced today that it will launch the cloud configuration version of windows 10. Cloud config provides a simple way to apply a set of unified configuration to windows 10 devices and turn them into simple and easy to manage terminals.It should be noted that the cloud configuration is not a new version or mode of windows. It is just the device configuration of windows 10 recommended by Microsoft, and it is cloud optimized for users with specific workflow requirements.Through windows 10 cloud configuration, end users can enjoy familiar windows interfaces and applications, while it administrators will get familiar management experience. If you have a Windows 10 professional or enterprise PC and endpoint manager, you can use the cloud to configure applications and settings. In addition to Microsoft teams, Microsoft edge and onedrive for business, cloud configuration gives you the flexibility to deploy Win32 and line of business (LOB) applications to devices directly or through virtualization.

five . Faster and clearer! NVIDIA realizes SDF real-time rendering for the first time, and the speed is improved by 3 orders of magnitude

In 2021, the trend of intelligent data collection and annotation industry of four major industries is predicted; Tsinghua University proposes depth aligned clustering for new intention map discovery

Recently, NVIDIA published a new research result, which improves the real-time rendering speed by 2-3 orders of magnitude. In terms of rendering quality, it can also better handle complex style and scale graphics data, and even synchronize the shadow that may be formed by ambient lighting in real time. This latest research is a paper entitled “neural geometry level of detail: real time rendering of implicit 3D shapes”. It is a research result jointly published by NVIDIA and researchers from the University of Toronto and McGill University. It has been submitted to arXiv.

In the paper, the researchers said,By introducing an efficient neural network representation method, they realized 3D high fidelity real-time rendering based on SDF for the first time, and achieved the most advanced quality of geometric reconstruction. More importantly, compared with other studies, it improves rendering speed by 2-3 orders of magnitude.

6. In depth learning performance measurement of NVIDIA A100: the training speed can reach 3.5 times of V100

In 2021, the trend of intelligent data collection and annotation industry of four major industries is predicted; Tsinghua University proposes depth aligned clustering for new intention map discovery

NVIDIA’s latestThe new architecture of A100 accelerated computing card,Its official website publicizes: automatic mixing precision and fp16 can bring twice the performance improvement for A100; moreover, without changing the code, the performance of A100 with tf32 is 20 times higher than that of NVIDIA Volta. Compared with V100, A100’s single precision floating-point computing power is improved from 15.7 tflops to 19.5 tflops, while double precision floating-point computing power is improved from 7.8 tflops to 9.7 tflops.

Academic news

1. AAAI 2021 Tsinghua proposes depth aligned clustering for new intention discovery

In 2021, the trend of intelligent data collection and annotation industry of four major industries is predicted; Tsinghua University proposes depth aligned clustering for new intention map discovery

The deep aligned algorithm successfully transfers knowledge with a few known intention priorsDeep pair clustering learning high quality intention feature representation is used for new intention map discovery.This method can use a small number of labeled intention samples to predict the number of new intention maps, and provide consistent pseudo labels as self-monitoring signals for representation learning through alignment strategy. Experiments are carried out on two benchmark datasets. Deepalaligned achieves significant improvement compared with the existing methods under different known intentions and cluster number settings.

*Link to the paper:https://arxiv.org/pdf/2012.08…

2. ACM mm 3D dense connection network based on time frequency space characteristics of emotional EEG signals

In 2021, the trend of intelligent data collection and annotation industry of four major industries is predicted; Tsinghua University proposes depth aligned clustering for new intention map discovery

In this paper, we put forward some suggestionsA EEG emotion recognition model, SST emotionnet,The model uses dual stream structure to capture complementary spatial, frequency and temporal features of EEG signals at the same time; meanwhile, a space-frequency / time attention mechanism is designed to enable the model to dynamically focus on some time-frequency space local patterns that are more recognizable for emotion recognition tasks. Experiments on the open EEG emotion recognition datasets seed and seed-iv show that the model has excellent performance, which is superior to the traditional baseline method. In addition, the model proposed in this paper is simpleA general framework of multivariate physiological time series can be applied to sleep staging, fatigue driving monitoring and other related fields in the future.

* Link to the paper:https://dl.acm.org/doi/abs/10…

The above information is from the Internet, compiled by the Jingdong technology developer’s official account, which does not represent Jingdong’s technology position.

Recommended reading

Welcome to click【Jingdong Technology】, learn about the developer community

More wonderful technical practice and exclusive dry goods analysis

Welcome to official account of Jingdong technology developer.

In 2021, the trend of intelligent data collection and annotation industry of four major industries is predicted; Tsinghua University proposes depth aligned clustering for new intention map discovery