Adaptive Swin Transformer Partitioning over AI-RAN Networks

2026-04-26

AI Summary

This work addresses the challenges of deploying Transformer-based real-time video object detection in dynamic 5G AI-RAN networks, where large intermediate activation volumes incur high communication overhead and unstable latency. The study introduces, for the first time, a throughput-aware adaptive model splitting strategy tailored to Transformer architectures, enabling dynamic selection of optimal split points without retraining. It further proposes an accuracy-preserving activation compression technique, deeply co-designed with distributed User Plane Functions (dUPFs) to enable end-to-end inference optimization. Experiments on the NVIDIA Aerial AI-RAN platform demonstrate that the proposed approach significantly reduces uplink traffic and user-plane latency while enhancing system stability. The work also quantifies practical trade-offs among latency, energy consumption, and privacy in real-world deployment scenarios.

Technology Category

• Machine Learning: Hardware-aware ML
• Computer Vision: Large Vision Models
• Search and Optimization: Distributed Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

This paper demonstrates the feasibility of transformer-based split inference for real-time video object detection over dynamic 5G AI-RAN networks. We extend throughput-aware adaptive splitting from CNNs to a Swin Transformer backbone and show that practical split execution is achievable for transformer-based vision models without retraining. To address the large intermediate activations inherent to transformers, we introduce an efficient, accuracy-preserving activation compression pipeline that substantially reduces uplink payload. The complete system -- including adaptive split selection, transformer inference, and compression -- is implemented and validated end-to-end on a real-time detection workload, with distributed UPF (dUPF) integration further reducing user-plane latency and improving runtime stability. Extensive measurements on an NVIDIA Aerial-based AI-RAN testbed jointly account for inference and 5G communication energy, quantifying the latency-energy-privacy trade-offs in realistic deployments.

Problem

• split inference
• Swin Transformer
• AI-RAN
• activation compression
• real-time video object detection

Innovation

• Adaptive Splitting
• Swin Transformer
• Activation Compression
• AI-RAN
• Split Inference