my timesThe Korea Times
  1. Business
  2. Companies

LG Uplus advances smartphone AI with breakthrough papers at top global conference

Listen

Summary

LG Uplus said Thursday that two papers on faster, more battery-efficient smartphone AI were accepted at NeurIPS 2026’s On-Device Intelligence workshop. Its smartphone-optimized small language model ran about 4.5 times faster and used about 77 percent less battery than existing methods, while performing on par with the latest cloud-based AI, the company said. A second method trained AI on a smartphone without sending personal data outside the device. LG Uplus plans to apply the technology to its AI services and expand the research to servers.


Key Facts

  • NeurIPS 2026 will be held in Sydney from Dec. 6 to 12.
  • The On-Device Intelligence workshop focuses on AI that runs directly on phones rather than remote servers.
  • LG Uplus said its smartphone model performed on par with the latest cloud-based AI.
  • The company said its smartphone training method achieved accuracy on par with server-based training.
  • Lee Sang-yeob, LG Uplus’ chief technology officer, said the company would apply the technology to services and expand research to server environments.
By Jhoo Dong-chan
  • Published Oct 29, 2026 2:18 pm KST
LG Uplus workers discuss the efficiency of artificial intelligence-powered smartphone features. Courtesy of LG Uplus

LG Uplus workers discuss the efficiency of artificial intelligence-powered smartphone features. Courtesy of LG Uplus

LG Uplus, one of Korea’s major telecommunications carriers, said Thursday that two of its research papers on making mobile artificial intelligence (AI) run faster while consuming less smartphone battery have been accepted at NeurIPS 2026, the world’s premier AI research conference.

The papers were selected for the conference’s On-Device Intelligence workshop, which focuses on AI that runs directly on phones rather than on remote servers, according to the Korean mobile carrier.

NeurIPS, the Conference on Neural Information Processing Systems, is considered one of the top three AI conferences along with ICML and ICLR. This year’s event will be held in Sydney from Dec. 6 to 12.

The research centers on token optimization. Tokens are the units of data AI models use to understand questions and generate answers. Processing them more efficiently speeds up responses and cuts power use.

LG Uplus developed a small language model for its call services and optimized it for smartphones with limited memory and power. The model ran about 4.5 times faster and used about 77 percent less battery than existing methods, while performing on par with the latest cloud-based AI, the company said.

The company also developed a way to train AI on a smartphone without sending personal data outside the device, achieving accuracy on par with server-based training.

LG Uplus plans to apply the technology to its AI services and expand the work to servers.

“We will apply the technology to actual services and expand our research to server environments,” said Lee Sang-yeob, LG Uplus’ chief technology officer.

This article was published with the assistance of generative AI and edited by The Korea Times.

Explore More

  • Q.

  • Q.

  • Q.