Experience the world’s 1st Redstonic Neural Network in Minecraft; Shenzhen holds drivers of self-driving cars responsible for crashes; AI brings back a decades-old concert | by Recode China AI

China AI News weekly from July 4 to July 10

The creators of Bilibili create the world’s 1st neural network in Minecraft

What’s new: While most artificial neural networks are embedded in coding, a team of five Chinese recently implemented a neural network in the sandbox video game Minecraft – the most popular most played game – using a type of element from game named redstones. They demonstrated the network in a video uploaded 10 days ago to Bilibili, a Chinese YouTube-like site, which quickly garnered over a million views and caught the attention of Yann LeCun, chief scientist of the IA and Turing Award winner, who hailed the work on Twitter as “very meta”.

Principles: Billed as the world’s first redstone neural network, the network uses LeNet, a CNN architecture created by LeCun in 1989, as its backbone and can recognize 15×15 handwritten digits with up to 80% accuracy. The input device comprises a single-pulse pressure plate tablet and a 15×15 coordinate screen. The input figure data is then passed through a 3×3 convolution layer with ReLU and three fully connected layers with tanh.

Thanks for reading Recode China AI! Subscribe for free to receive new posts and support my work.

The creators said they used stochastic computing to make the network, which made it much simpler in design and layout compared to traditional high-precision computing. They pre-trained the neural network in PyTorch and built the network in Minecraft, the recognition time is theoretically 5 minutes per digit. However, limited by Minecraft’s computing capacity, the actual running time exceeds 20 minutes.

Video: You can watch the first video【Minecraft】世界首个纯红石神经网络! and the second deep dive video【Minecraft】红石卷积神经网络——原理. The code is available on GitHub StochasticNet.

Shenzhen’s new regulations hold drivers of self-driving cars liable in crashes

What’s new: Last week, Shenzhen approved a regulation that says drivers of self-driving cars will be held liable for accidents.

As the first-of-its-kind regulation in China (and possibly the world) that clarifies responsibilities for cars with self-driving capabilities, the Shenzhen Special Economic Zone Regulation on Connected Intelligent Vehicle Management will enter effective August 1, 2022. Intelligent connected vehicles are referred to as autonomous vehicles that are operated by an automated system. Below are noteworthy highlights.

Paired with drivers: The new regulations require vehicles with conditional or high-level autonomous driving – comparable to L3 and L4 in SAE driving automation levels – to be paired with drivers behind the wheel and an associated manual driving mode. Fully autonomous vehicles are allowed to remove drivers and manual driving systems, but can only operate in designated areas.

Responsibilities: The driver of a self-driving car – with conditional or high-level automation – will be held liable for damages resulting from the accident if they are at fault, according to the new regulations. If the accident is caused by the vehicle due to systematic failures, the driver, owner or operator of the vehicle can claim compensation from the manufacturer and seller of the vehicle.

The owner and operator of fully autonomous vehicles and are responsible for a car accident, on the other hand.

Communications with other road users: Autonomous car manufacturers must equip the vehicle with an external device that can indicate its autonomous driving mode.

Chinese TikTok restores decades-old concert using AI

What’s new: An AI-enhanced video of a concert performed by prominent Hong Kong rock band Beyond three decades ago was released on TikTok’s Chinese counterpart Douyin last week, attracting more than 140 million viewers to watch. .

Who is beyond? Beyond were a Hong Kong rock band formed in 1983 and quickly became a household name in East Asia in the 1980s-90s, led by Wong Ka Kui. to party, beyond Live 1991 Life Contact Concert, was a concert held by Beyond at Hung Hom Stadium in Hong Kong in September 1991, also the most performed.

Technical details: Researchers at Volcano Engine, the enterprise technology unit of TikTok’s parent company, ByteDance, refreshed the decade-old video by upgrading its horizontal resolution from 540p to 4K, a 16x upgrade. They also used:

  • Multiframe deinterlacing remove complex artifacts from early interlaced videos;
  • Adaptive Portrait Augmentation to restore Wong’s facial details such as locks of hair and figure;
  • Improved color brightness to rebalance overexposure and dark scenes;
  • Audio denoising, audio superresolution and loudness normalization to improve audio quality.

Researchers from Tsinghua University, the University of Cambridge and the University of Illinois at Chicago have come up with CHEF, China’s first fact-checking dataset based on evidence from 10,000 real-world claims . The dataset covers multiple areas, ranging from politics to public health, and provides annotated evidence culled from the internet. In addition, we develop established baselines and a new approach capable of modeling evidence retrieval as a latent variable, enabling joint training with the end-to-end veracity prediction model. Extensive experiments show that CHEF will provide a challenging test bed for the development of fact-checking systems designed to retrieve and reason about non-English assertions.

Researchers from Changping Lab, Peking University, Huawei and others presented the first million-level protein structure prediction dataset with high coverage and diversity, named PSP. This dataset consists of 570,000 real structure sequences (10 TB) and 745,000 complementary distillation sequences (15 TB). They further provide the reference training procedure for the SOTA protein structure prediction model on this dataset. They validate the usefulness of this dataset for training by participating in the CAMEO competition in which the model won first place.

Alibaba’s fintech arm Ant Group has opened up SecretFlow, a unified framework for privacy-preserving data intelligence and machine learning. The framework provides

  • An abstract device layer consists of simple devices and secret devices that encapsulate various cryptographic protocols.
  • A device stream layer modeling higher algorithms as device object streams and DAGs.
  • A layered algorithm for performing data analysis and machine learning with horizontally or vertically partitioned data.
  • A workflow layer that seamlessly integrates data processing, model training, and hyperparameter tuning.

DMAIa cognitive AI company, raised 500 million RMB ($75 million) in its Series B funding round. Founded in 2017 by renowned computer vision luminary Song-Chun Zhu, the company based in Guangzhou develops cognitive AI assistants and platforms in the education and gaming sectors.

MetaX, an integrated circuit design company specializing in GPUs, raised RMB 1 billion ($150 million) in its pre-B funding round. Founded in 2020, the Shanghai-based company develops high-performance GPU silicon for use in artificial intelligence, video and graphics rendering, and more.

Autowise.ai, a self-driving technology company, raised RMB 200 million ($30 million) in its Series B2 funding round. Founded in 2017 by former driving chief Didi Chuxing, Chao Huang, the Shanghai-based company develops self-driving sweepers and scrubber driers for indoor and outdoor cleaning.






Leave a Reply

Your email address will not be published. Required fields are marked *