Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EdgeSync: Faster Edge-model Updating via Adaptive Continuous Learning for Video Data Drift

Peng Zhao, Runchu Dong, Guiqin Wang, Cong Zhao

TL;DR

EdgeSync addresses the challenge of maintaining accuracy for lightweight edge models amid streaming video data drift by enabling cloud-based retraining of updated edge models while minimizing update latency. It combines a sample filtering module that scores timeliness and adaptability to select informative frames for upload with a training management module that uses offline Bayesian optimization and online early stopping to determine retraining timing and hyperparameters. The approach reduces bandwidth and latency while improving accuracy over strong baselines in complex scenes, demonstrating practical viability for large-scale edge deployments. This work enables scalable, real-time edge analytics across multiple cameras by supporting fast, adaptive updates.

Abstract

Real-time video analytics systems typically place models with fewer weights on edge devices to reduce latency. The distribution of video content features may change over time for various reasons (i.e. light and weather change) , leading to accuracy degradation of existing models, to solve this problem, recent work proposes a framework that uses a remote server to continually train and adapt the lightweight model at edge with the help of complex model. However, existing analytics approaches leave two challenges untouched: firstly, retraining task is compute-intensive, resulting in large model update delays; secondly, new model may not fit well enough with the data distribution of the current video stream. To address these challenges, in this paper, we present EdgeSync, EdgeSync filters the samples by considering both timeliness and inference results to make training samples more relevant to the current video content as well as reduce the update delay, to improve the quality of training, EdgeSync also designs a training management module that can efficiently adjusts the model training time and training order on the runtime. By evaluating real datasets with complex scenes, our method improves about 3.4% compared to existing methods and about 10% compared to traditional means.

EdgeSync: Faster Edge-model Updating via Adaptive Continuous Learning for Video Data Drift

TL;DR

EdgeSync addresses the challenge of maintaining accuracy for lightweight edge models amid streaming video data drift by enabling cloud-based retraining of updated edge models while minimizing update latency. It combines a sample filtering module that scores timeliness and adaptability to select informative frames for upload with a training management module that uses offline Bayesian optimization and online early stopping to determine retraining timing and hyperparameters. The approach reduces bandwidth and latency while improving accuracy over strong baselines in complex scenes, demonstrating practical viability for large-scale edge deployments. This work enables scalable, real-time edge analytics across multiple cameras by supporting fast, adaptive updates.

Abstract

Real-time video analytics systems typically place models with fewer weights on edge devices to reduce latency. The distribution of video content features may change over time for various reasons (i.e. light and weather change) , leading to accuracy degradation of existing models, to solve this problem, recent work proposes a framework that uses a remote server to continually train and adapt the lightweight model at edge with the help of complex model. However, existing analytics approaches leave two challenges untouched: firstly, retraining task is compute-intensive, resulting in large model update delays; secondly, new model may not fit well enough with the data distribution of the current video stream. To address these challenges, in this paper, we present EdgeSync, EdgeSync filters the samples by considering both timeliness and inference results to make training samples more relevant to the current video content as well as reduce the update delay, to improve the quality of training, EdgeSync also designs a training management module that can efficiently adjusts the model training time and training order on the runtime. By evaluating real datasets with complex scenes, our method improves about 3.4% compared to existing methods and about 10% compared to traditional means.
Paper Structure (27 sections, 4 equations, 5 figures, 3 tables, 2 algorithms)

This paper contains 27 sections, 4 equations, 5 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Video Analytics Systems
  4. Continuous Learning
  5. Unsupervised Adaptation Methods
  6. Method
  7. System Architecture
  8. Sample Filtering Module
  9. Training Manager
  10. Select a model for retraining
  11. Adapt update frequency for retraining
  12. Send retrained parameters back to edge
  13. experiments
  14. Experiment Settings
  15. Datasets
  16. ...and 12 more sections

Figures (5)

  • Figure 1: Changes of class distributions and accuracy when a car camera enters downtown of ideal fast retraining and normal retraining.
  • Figure 2: Overall architecture of EdgeSync
  • Figure 3: Impact of accuracy with varying numbers of cameras.
  • Figure 4: Impact of hyper-parameters profiling in retraining windows.
  • Figure 5: Sensitivity analysis of filter percentage parameters.