Top 10 Tech Trends for 2022; How to Make Composite Image Look Authentic? Add Shadows; New Chinese Benchmark Tests NLU & NLG Models
China’s AI news in the week of January 16, 2022
Alibaba, Tencent Releases 10 Tech Trends of 2022
Despite the continued uncertainties of Covid-19, 2021 ended with numerous meaningful research breakthroughs in AI. As we bid farewell to 2021 and welcome 2022 with high hopes, Chinese tech companies are revealing their predictions of tech trends in 2022.
Damo Academy, the research wing of Alibaba, began compiling an annual selection of the year's most important technology trends since 2018. In December, the research institute unveiled its list of top 10 technology trends. The results were analyzed from 7.7 million publicly available papers in the past three years.
AI for Science
Co-evolution of Large- and Small-scale AI Models
Silicon Photonic Chips
AI for Renewable Energy
Perceptive Soft Robotics
High-precision Medicine
Privacy-preserving Computation
Satellite-terrestrial Integrated Computing
Cloud-Network-Device Convergence
Extended Reality (XR)
Jeff Zhang, President of Alibaba Cloud Intelligence and Head of Alibaba DAMO Academy, “The boundary of technologies is extended from the physical world to mixed reality. More and more cutting-edge technologies find their way to industrial applications. Digital technologies power a green and sustainable future.”
Tencent Research Institute also released a list of 10 technology trends that spans from cloud computing and digital avatars.
Cloud-Native IT
Quantum Computing in the NISQ era
Industrialization and Generalization of AI
Cloud-Empowered Connections
Cloud-Native Security
Digital Twins
Extended Reality (XR)
Multi-modal Service Robots
Energy network
Satellite-terrestrial Cooperation
On the other side of the Pacific Ocean, Google AI Chief Jeff Dean released his highly-anticipated annual recap of Google research of the past year, and forecasted five important directions and progresses of AI in the next five years.
More Capable, General-Purpose ML Models
Continued Efficiency Improvements for ML
ML Is Becoming More Personally and Communally Beneficial
Growing Benefits of ML in Science, Health, and Sustainability
Deeper and Broader Understanding of ML
Shadow Generation for Composite Image in Real-world Scenes
Look at the composite image above that mixes a foreground object with a different background image and tell me why it looks unauthentic. Something is missed, right? It’s the shadow.
A team of researchers from Shanghai Jiao Tong University and Tokyo-based RIKEN Center for Advanced Intelligence Project recently proposed a method to generate shadows of foreground objects for background images. The research team piled up a real-world shadow generation dataset DESOBA by generating synthetic composite images based on paired real images and deshadowed images and proposed a novel shadow generation network SGRNet.
Image composition can be applied to a wide range of applications like virtual social media, artistic creations, and automatic advertising. The shadow generation task is to solve the shadow inconsistency between the foreground object and the background image, which is an essentially image-to-image translation problem
Researchers borrowed the SOBA dataset, which has 840 training images with 2,999 object-shadow pairs and 160 test images with 624 object-shadow pairs, as the base dataset. The SGRNet consists of two stages: a shadow mask prediction stage and a shadow filling stage.
CUGE: A Chinese Language Understanding and Generation Evaluation Benchmark
China’s top AI institute BAAI released CUGE, a Chinese Language Understanding and Generation Evaluation benchmark, as the counterpart of GLUE and GEM for Chinese NLP models.
CUGE covers 7 important language capabilities, 17 mainstream NLP tasks, and 19 representative datasets. The benchmark has the following features: (1) Hierarchical benchmark framework, where datasets are principally selected and organized with a language capability-task-dataset hierarchy. (2) Multi-level scoring strategy, where different levels of model performance are provided based on the hierarchical framework. CUGE is now publicly available with a leaderboard.
The creator of CUGE is the Beijing Academy of Artificial Intelligence (BAAI), a non-profit research institute dedicated to promoting collaboration among academia and industries, as well as fostering top talents and a focus on long-term research on the fundamentals of AI technology.
BAAI made headlines last year with the release of Wu Dao 2.0, a 1.75-trillion-parameter MoE model trained on China’s supercomputer Sunway. Wu Dao 2.0 offers various capabilities for downstream tasks, including question answering, language generation, semantic understanding, captioning, and text-to-image/video generation.
Investment News:
Smarter Eye, a Beijing-based developer of computer vision technology designed for automotive, has raised nearly RMB300 million yuan (~$47 million) in its Series C funding round. Founded in 2014, Smarter Eye advances 3D perception solutions for ADAS and active safety systems like height limit warning systems, based on binocular cameras and vision algorithms.
Waytous, one of the largest autonomous mining solution providers in China, has raised almost RMB300 million yuan (~$47 million) in its Series C funding round. Founded in 2014, the company provides a total autonomous solution for the mining industry to improve its productivity, sustainability and safety.
DeepGlint, a Beijing-based computer vision and facial recognition developer, has received approvals to go public on the Shanghai STAR market. Founded in 2013, the company plans to raise RMB1 billion yuan (~$157 million) for technology development and algorithm advancement. The company is on the U.S. entity list, which restricts US companies from doing business with it without a special license.