SoftBank Corp. develops a foundational Large Telecom Model (LTM)

SoftBank Corp. announced the development of a new Large Telecom Model (LTM), a Generative AI (GenAI) foundation for the telecom industry. The LTM utilizes diverse datasets, including SoftBank’s expansive network data and operational insights, to facilitate advanced inference in the management and design of cellular networks. The company aims to integrate this model into its own operations to enhance network efficiency.

Specialized Artificial Intelligence (AI) models have been created through fine-tuning the LTM, particularly for optimizing base station configurations. Notably, these fine-tuned models have demonstrated over 90% accuracy in predicting configurations for base stations excluded from training, significantly reducing the time required for adjustments from days to minutes, thus minimizing operational costs and human error.

The LTM also supports the “AI for RAN” initiative designed to improve Radio Access Network (RAN) performance. Expected future developments include using the LTM as a guideline for network design and the creation of AI agents for network optimization.

Leveraging NVIDIA technologies, specifically the NIM system, has improved the performance of the LTM, yielding up to a fivefold enhancement in specified metrics. This collaboration grants deployment flexibility across on-premises (on-prem) and cloud environments.

This work aligns with SoftBank's “Human AI” concept, spearheaded by its Research Institute of Advanced Technology (RIAT), which aims to use AI for improving mobile networks. SoftBank plans to combine various AI models developed from the LTM with its AI-RAN solution currently in progress.

Ryuji Wakikawa, Vice President at SoftBank RIAT, stated that the LTM can address various operational processes and enhance network performance. Chris Penrose from NVIDIA highlighted the model's potential to streamline and expedite network operations through specialized AI agents.

SoftBank intends to further explore the LTM’s use in its services while continuing to advance Research and Development (R&D) efforts and collaborations to progress next-generation networks.