Huawei Cloud AI labels cloud desktop, the security guard behind artificial intelligence

With the continuous development of artificial intelligence technology, from smart homes to smart driving, from weather diagnosis to financial risk control, the application scope of AI is becoming more and more extensive, and the benefits it brings are becoming more and more significant. The training of large models requires enterprises to provide massive training data, which has extremely high requirements for security and privacy protection. Currently, the data of large models requires a large amount of manual annotation.

On September 21, at the Huawei Full Connection Conference 2023, Huawei Cloud CTO Zhang Yuxin shared that Huawei Cloud AI Annotation Cloud Desktop provides safe and efficient solutions and application practices for the annotation industry.

Zhang Yuxin, Chief Technology Officer of Huawei Cloud

Take intelligent car driving as an example. During the intelligent driving process, the intelligent driving system makes specific behavioral decisions based on surrounding environment information, such as data feedback from other vehicles, traffic lights, pedestrians, etc., such as braking, changing lanes, overtaking, etc. . To train a set of qualified perception algorithms and allow cars to handle more and more complex scenes, it is necessary to have massive, accurate, and high-quality road scene data. Professionals will label the data and convert it into supporting algorithm training. The data.

However, intelligent driving annotation data contains a large amount of geographical information 3D point clouds, geographical information, trajectories, etc., involving a large number of geographical elements and location information, and national regulations require strict control. Currently, car companies can already intelligently label these sensitive data through large models. However, at least 10% of the data still needs to be fine-tuned and manually labeled by annotators. During this entire process, data security, annotation efficiency, and annotation quality are undoubtedly the three core elements that car companies are most concerned about.

In terms of data security, the massive geographical data labeled by intelligent driving is sensitive data. In the past, data labeling was performed through the browser or client, and there is still the risk of data leakage caused by the browser and local cached data. Now, Huawei Cloud provides a secure and compliant cloud automotive zone, which can realize data storage, annotation, model training, simulation, and deployment all on the cloud. Using AI to annotate cloud desktops further ensures that sensitive data in the data annotation process does not fall to the ground.

Data annotation is a labor-intensive industry, and annotation tasks are usually completed through third-party professional annotation outsourcing. In order to meet data security requirements, companies require annotators to work together in an office area and deploy an isolated network, or open a private intranet line for outsourcing teams in remote locations, which will bring high costs. Through Huawei Cloud AI Annotation Cloud Desktop, annotators scattered across the country can access a secure annotation environment and conduct safe annotation through the annotation platform. For labeling scenarios such as intelligent driving point cloud labeling that require graphics cards, GPU cloud applications can also be used to reduce the cost of enterprise labeling terminals

In terms of annotation efficiency, 3D point clouds, pictures, videos, etc. must be cached when loaded locally. Annotation data of about 20M often requires 4 to 5 seconds to be loaded. Especially for 3D point clouds, once dragged, some data needs to be reloaded. Now data is interconnected on Huawei Cloud and loaded in real time on the cloud. Combined with millisecond-level transmission of Huawei Cloud HDP high-definition display protocol, high-definition images and videos can be loaded instantly, reducing the local display delay of images by more than 50%, allowing annotators to annotate quickly and smoothly.

In terms of annotation quality, when encountering fuzzy data, such as road conditions images collected in bad weather, multiple people are often required to make judgments to ensure accurate data annotation. In order to reduce the rework rate, annotators usually ask colleagues for help through instant messaging software, and sometimes screenshots are needed to solve annotation problems. This is not only inefficient, but also poses data security risks. This problem can also be solved through Huawei Cloud's multi-user collaborative desktop. Its collaboration capability allows annotators to initiate collaboration with one click, allowing experienced colleagues or experts to perform annotation operations to quickly resolve data doubts and achieve accurate annotation.

Huawei Cloud AI Annotation Cloud Desktop has been successfully used in Huawei’s smart car autonomous driving data annotation project. Through this technology, more than 2,000 annotators perform annotation work in the cloud, achieving regional collaboration across 7 cities, ensuring the efficiency and safety of the annotation work

Targeted at smart home, security, finance, Internet and other industries, in manual annotation scenarios such as pictures, videos, and voices, the AI ​​annotation cloud desktop will also help artificial intelligence continue to move forward with its safe, efficient, and collaborative capabilities.

The above is the detailed content of Huawei Cloud AI labels cloud desktop, the security guard behind artificial intelligence. For more information, please follow other related articles on the PHP Chinese website!