Towards the social implementation of IOWN DCI technologies: a real-time AI analysis platform at EXPO2025 OSAKA, KANSAI, Japan.

We'll share our journey, technical advancements, and future evolution from our project on a real-time AI analysis platform supporting various use cases at the NTT pavilion at Expo 2025 Osaka, Kansai, Japan. This platform extends our recognized PoC of the CPS Area Management Use Case, fully leveraging IOWN (Innovative Optical and Wireless Network) DCI (Data-Centric Infrastructure) technologies to improve latency, scalability, and energy efficiency.

The highlights of our presentation are as follows:

• AI use cases at the NTT pavilion at Expo 2025 Osaka, Kansai, Japan, including controlling wall movement based on visitors' facial expressions, detecting visitors' falls for immediate rescue, and applying blur processing to maintain privacy


• An overview of our AI analysis platform with a three-tiered geographically distributed processing model (customer edge, edge cloud, and public cloud) leveraging RDMA (Remote Direct Memory Access) over IOWN APN (All Photonics Network) to efficiently collect a large volume of sensor data and flexibly utilize GPU resources at remote sites with minimal communication overheads


• Accelerated data pipelines for AI inference utilizing advanced offloading techniques, capable of completing all data processing for each AI inference within a single MIG instance of GPUs


• RDMA-based chaining of the accelerated data pipelines to support value-added but complex AI use cases combining results from multiple AI models, and
  
• Scaling of the accelerated data pipelines using Kubernetes and CDI (Composable Disaggregated Infrastructure) to optimize hardware resource utilization.