South Korean Telecom LG U+ Says On-Device AI Optimization Tech Accepted at NeurIPS, Claims 4.5x Speedup
LG U+ announced that two papers on smartphone AI optimization were accepted at the NeurIPS 2026 workshop, with the company claiming ~4.5x faster processing and ~77% less battery consumption.
South Korean telecommunications operator LG U+ announced on October 29 that two of its research papers on artificial intelligence (AI) token optimization technology have been accepted for the "On-Device Intelligence" workshop at NeurIPS 2026. NeurIPS is one of the world's three top-tier AI academic conferences alongside ICML and ICLR, with this year's event scheduled for December in Sydney, Australia.
The research aims to address the constraints of limited memory and battery life in smartphones by improving the computational efficiency of tokens, the basic unit of data processed by AI, thereby reducing response latency and power consumption. LG U+ stated that it applied lightweighting and inference acceleration technologies to small language models (sLMs) used for call services. According to the company, processing speed increased by approximately 4.5 times, while battery consumption decreased by about 77% compared to existing methods.
Additionally, LG U+ developed technology for additional AI learning within the smartphone itself, reducing computation and memory usage to allow model training on the device without transmitting personal data to external servers. Verification on actual smartphones showed accuracy levels similar to server-trained models. Lee Sang-yeop, Chief Technology Officer (CTO) of LG U+, said the company is considering applying these findings to its AI services and plans to expand the optimization techniques validated on smartphones to server GPU environments.