講演情報
[BT-1-02]Name-Based In-Network Inference for Low-Latency Edge Intelligence
〇Htet Htet Hlaing1, Hitoshi Asaeda1 (1. National Institute of Information and Communications Technology (NICT))
キーワード：
Information-centric networking、In-network computing、AI inference、Edge intelligence
AI workloads, built on models with billions of parameters and large data volumes, outpace server-centric infrastructure in scalability and responsiveness. While serverless approaches offer elastic, on-demand execution, they struggle with compute-intensive, latency-sensitive inference. This paper presents a name-based in-network inference framework for low-latency edge intelligence, where AI tasks are represented as named invocable functions using information-centric networking (ICN). Through intent-aware resolution and in-network function discovery, requests such as object detection or anomaly analysis are discovered, routed, and executed at intermediate nodes. Evaluation results show that the proposal reduces task completion time compared with baselines and improves inference throughput across evaluated tasks.