HÀ NỘI — Viettel AI is set to present its latest breakthrough in artificial intelligence at the International Conference on Learning Representations (ICLR) 2025 — one of the world’s most prestigious gatherings for deep learning and AI research.

The company will officially introduce a pioneering model integration method known as CAMEx, a technology currently under active exploration by global tech giants such as Meta and xAI to address the challenge of scaling large AI models without compromising performance.

CAMEx, developed in collaboration between Viettel’s Data and Artificial Intelligence Services Centre (Viettel AI) and the National University of Singapore (NUS), enables the expansion of AI language models to up to five times larger while maintaining processing speed and accuracy.

In an era where the size and cost of training large language models continue to escalate, CAMEx represents a significant innovation — reducing the computational burden while unlocking broader applications for AI in complex domains.

Unlike conventional model merging techniques that often lead to data loss and reduced efficiency, CAMEx is designed to preserve training information while enhancing performance across a wide range of tasks, including language modelling, text classification, inference, content moderation, image classification and object detection.

A Viettel AI representative stated that trial results demonstrated CAMEx not only improves model performance but also offers immense potential for deployment in high-tech sectors, particularly in markets demanding both high speed and precision.

ICLR 2025, which is taking place in Singapore from April 24 to 28, will feature leading scientists and major technology firms such as Google, Meta and Microsoft, showcasing cutting-edge AI research and real-world applications.

A subsidiary of the Viettel Military Industry and Telecoms Group, Viettel AI is currently developing an expansive ecosystem spanning AI, Big Data, robotics and Digital Twin technologies. — VNS