Approved for presentation at the world’s premier AI conference, Viettel AI’s solution promises to scale AI models up to 5 times without compromising processing speed.

Recently, Viettel AI received an invitation to present its scientific research at ICLR 2025 (International Conference on Learning Representations) – one of the most prestigious global conferences on Artificial Intelligence (AI), taking place from April 24–28 in Singapore. The event will bring together top researchers, experts, and leading tech companies from around the world.



At the conference, Viettel AI will present a research paper titled “CAMEx: Curvature-aware Merging of Experts”, developed in collaboration with the National University of Singapore (NUS). The study focuses on optimizing performance and scalability of AI models, offering exceptional flexibility for large-scale models.

Specifically, the solution enables the expansion of language model size by up to five times (from 200 million parameters to 1 billion) while maintaining the same response speed. This allows the model not only to process data quickly but also to significantly improve accuracy, expressiveness, and efficiency in handling complex tasks. The solution opens up vast application potential in fields such as intelligent chatbots, traffic surveillance camera systems, and more.

ICLR 2025 is among the most reputable events worldwide in the fields of AI and Deep Learning, organized by leading scientists and experts and sponsored by tech giants like Google Research, Meta, and Microsoft. The conference not only showcases technological breakthroughs but also serves as a launchpad for pioneering research that brings AI into real-world applications across diverse fields such as data science, speech recognition, and computer vision.