CDIO: Cross-Domain Inference Optimization with Resource Preference Prediction for Edge-Cloud Collaboration

TL;DR

CDIO addresses efficient real-time video inference in edge-cloud systems under dynamic resource conditions by coupling a spatiotemporal resource preference predictor with a cross-domain optimization algorithm. The predictor uses features $x_t=I_t+C_t+A_t+D_t$ and a multi-layer LSTM to output a binary resource preference, while the optimizer is framed as a combinatorial multi-armed bandit that minimizes $ \frac{1}{T}\sum_t (U_t + \varphi B_t)$ under accuracy and delay constraints. An explicit approximate regret $Reg(T)=T \alpha \beta R(S_{\max})-\mathbb{E}[\sum_t R(S_t)]$ guides learning, with a feedback mechanism that updates resource allocations. Empirical results on UA-DETRAC with edge and cloud hardware show CDIO reduces computing and bandwidth consumption by 20-40% and energy by over 40% while maintaining high accuracy and low latency, demonstrating robust performance under bandwidth fluctuations. This work offers a practical, adaptive framework for multi-resource coordination in real-time multimedia edge-cloud deployments.

Abstract

Currently, massive video tasks are processed by edge-cloud collaboration. However, the diversity of task requirements and the dynamics of resources pose great challenges to efficient inference, resulting in many wasted resources. In this paper, we present CDIO, a cross-domain inference optimization framework designed for edge-cloud collaboration. For diverse input tasks, CDIO can predict resource preference types by analyzing spatial complexity and processing requirements of the task. Subsequently, a cross-domain collaborative optimization algorithm is employed to guide resource allocation in the edge-cloud system. By ensuring that each task is matched with the ideal servers, the edge-cloud system can achieve higher efficiency inference. The evaluation results on public datasets demonstrate that CDIO can effectively meet the accuracy and delay requirements for task processing. Compared to state-of-the-art edge-cloud solutions, CDIO achieves a computing and bandwidth consumption reduction of 20%-40%. And it can reduce energy consumption by more than 40%.

Figures (9)

• Figure 1: The illustration of edge-cloud collaborative inference architecture.
• Figure 2: The inference results for different tasks on edge and cloud. If Task 1 is more sensitive to delay, it is suitable for uploading to the edge processing. If Task 2 requires better accuracy, it is suitable for processing in the cloud.
• Figure 3: The workflow of the proposed CDIO framework.
• Figure 4: The overview of the multi-layer LSTM model.
• Figure 5: The delay analysis within CDIO framework.