Computer vision technology, which is widely used in the industry, is becoming increasingly advanced thanks to the development of AI. A technique with more sophisticated predictive power than the active learning technique that verifies labeled data will be announced at the upcoming International Conference on Computer Vision.

Global video technology company Hyperconnect announced on the 25th that it will present a machine learning-related technical paper at the 'International Conference on Computer Vision (ICCV) 2023', the world's top computer science conference.

ICCV is one of the three major international academic conferences representing the field of computer vision, along with the Computer Vision and Pattern Recognition Conference (CVPR) and the European Computer Vision Conference (ECCV), and is considered an authoritative conference in the field of artificial intelligence. Hyperconnect will present the paper 'TiDAL: Learning Training Dynamics for Active Learning' at the ICCV 2023 to be held in Paris, France in October.

This was evaluated as an achievement following Hyperconnect's publication of the paper 'Learning with Noisy Labels by Efficient Transition Matrix Estimation to Combat Label Miscorrection' at ECCV last year.

Hyperconnect will introduce 'TiDAL', a technique designed to improve the data learning efficiency of machine learning, and 'active learning', the topic of the paper, is a technique to intelligently select and efficiently label the data that is most useful for improving model performance from the data pool required for advanced machine learning to train the model. In the labeling task of naming data such as images, videos, and audio, active learning is a method to improve classification performance by extracting data problems that are difficult for AI to judge on its own, and has the advantage of minimizing the manpower and cost required for labeling work.

Data labeled in this way through AI technology needs to be verified using an algorithm that measures the error between the predicted value and the actual value through active learning. 'TiDAL' can detect model behavior that changes during learning and predict data more precisely than existing active learning methods, and as a result, it can increase the accuracy of data labeling tasks and design machine learning models with improved performance at a lower cost than existing methods.

“Continuous investment and ceaseless research in machine learning technology, which is attracting attention in various industries, have led to encouraging results recognized by global academic societies, proving Hyperconnect’s outstanding technological prowess,” said Ha Sung-joo, Director of Hyperconnect AI Lab. “This research is expected to be applied to various Hyperconnect services in the future and greatly contribute to enhancing user experience value.”

Meanwhile, Hyperconnect is working to develop innovative technologies that can be utilized in the actual service development and operation stages through its own AI lab, and is receiving high evaluations for its research achievements and technological prowess worldwide by developing and announcing various technologies such as AI and deep learning. In addition, it has proven its own technological competitiveness by securing approximately 280 global registered patents in Korea, the US, Japan, Europe, and other countries around the world.