Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Prioritized Information Bottleneck Theoretic Framework with Distributed Online Learning for Edge Video Analytics

Zhengru Fang, Senkang Hu, Jingjing Wang, Yiqin Deng, Xianhao Chen, Yuguang Fang

TL;DR

The paper addresses the challenge of efficient, low-latency edge-based video analytics with multiple cameras by introducing the Prioritized Information Bottleneck (PIB) framework. PIB prioritizes data sharing using SNR and RoI-aware weights, employs a deterministic information bottleneck with variational bounds, and uses a gate mechanism based on distributed online learning to select edge servers, achieving sublinear regret. It fuses information across cameras with a multi-frame correlation model and optimizes communication via a CMAB formulation, providing theoretical regret bounds and scalable encoding/decoding losses. Real-world experiments on heterogeneous edge devices demonstrate substantial improvements in mean object detection accuracy (MODA) and large reductions in communication cost under poor channel conditions, enabling robust, low-latency collaborative perception without video reconstruction at the edge. The work significantly advances practical, scalable, task-focused edge analytics for dense multi-camera deployments.

Abstract

Collaborative perception systems leverage multiple edge devices, such surveillance cameras or autonomous cars, to enhance sensing quality and eliminate blind spots. Despite their advantages, challenges such as limited channel capacity and data redundancy impede their effectiveness. To address these issues, we introduce the Prioritized Information Bottleneck (PIB) framework for edge video analytics. This framework prioritizes the shared data based on the signal-to-noise ratio (SNR) and camera coverage of the region of interest (RoI), reducing spatial-temporal data redundancy to transmit only essential information. This strategy avoids the need for video reconstruction at edge servers and maintains low latency. It leverages a deterministic information bottleneck method to extract compact, relevant features, balancing informativeness and communication costs. For high-dimensional data, we apply variational approximations for practical optimization. To reduce communication costs in fluctuating connections, we propose a gate mechanism based on distributed online learning (DOL) to filter out less informative messages and efficiently select edge servers. Moreover, we establish the asymptotic optimality of DOL by proving the sublinearity of their regrets. To validate the effectiveness of the PIB framework, we conduct real-world experiments on three types of edge devices with varied computing capabilities. Compared to five coding methods for image and video compression, PIB improves mean object detection accuracy (MODA) while reducing 17.8% and reduces communication costs by 82.65% under poor channel conditions.

Prioritized Information Bottleneck Theoretic Framework with Distributed Online Learning for Edge Video Analytics

TL;DR

The paper addresses the challenge of efficient, low-latency edge-based video analytics with multiple cameras by introducing the Prioritized Information Bottleneck (PIB) framework. PIB prioritizes data sharing using SNR and RoI-aware weights, employs a deterministic information bottleneck with variational bounds, and uses a gate mechanism based on distributed online learning to select edge servers, achieving sublinear regret. It fuses information across cameras with a multi-frame correlation model and optimizes communication via a CMAB formulation, providing theoretical regret bounds and scalable encoding/decoding losses. Real-world experiments on heterogeneous edge devices demonstrate substantial improvements in mean object detection accuracy (MODA) and large reductions in communication cost under poor channel conditions, enabling robust, low-latency collaborative perception without video reconstruction at the edge. The work significantly advances practical, scalable, task-focused edge analytics for dense multi-camera deployments.

Abstract

Collaborative perception systems leverage multiple edge devices, such surveillance cameras or autonomous cars, to enhance sensing quality and eliminate blind spots. Despite their advantages, challenges such as limited channel capacity and data redundancy impede their effectiveness. To address these issues, we introduce the Prioritized Information Bottleneck (PIB) framework for edge video analytics. This framework prioritizes the shared data based on the signal-to-noise ratio (SNR) and camera coverage of the region of interest (RoI), reducing spatial-temporal data redundancy to transmit only essential information. This strategy avoids the need for video reconstruction at edge servers and maintains low latency. It leverages a deterministic information bottleneck method to extract compact, relevant features, balancing informativeness and communication costs. For high-dimensional data, we apply variational approximations for practical optimization. To reduce communication costs in fluctuating connections, we propose a gate mechanism based on distributed online learning (DOL) to filter out less informative messages and efficiently select edge servers. Moreover, we establish the asymptotic optimality of DOL by proving the sublinearity of their regrets. To validate the effectiveness of the PIB framework, we conduct real-world experiments on three types of edge devices with varied computing capabilities. Compared to five coding methods for image and video compression, PIB improves mean object detection accuracy (MODA) while reducing 17.8% and reduces communication costs by 82.65% under poor channel conditions.
Paper Structure (24 sections, 6 theorems, 46 equations, 13 figures, 3 tables, 2 algorithms)

This paper contains 24 sections, 6 theorems, 46 equations, 13 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background
  3. State-of-the-Art
  4. Edge Video Analytics
  5. Learning-Based Transmission Scheduling
  6. Our Contributions
  7. System Model
  8. Communication Model
  9. Priority Weight Formulation
  10. Problem Formulation
  11. Prioritized Information Bottleneck Analysis
  12. Combinatorial Multi-armed Bandit (CMAB) Problem
  13. Methodology
  14. Architecture Summary
  15. Variational Approximation Method
  16. ...and 9 more sections

Key Result

Proposition 1

The probabilistic model of decoder $p(Y|Z)$ maps a representation $Z \in \mathbb{Z}$ into task inference $Y \in \mathbb{Y}$. Let $q(Y|Z)$ denote the variational approximation of decoder $p(Y|Z)$. We can obtain

Figures (13)

  • Figure 1: Comparison of compression methods: (a) Traditional compression method with redundant data, (b) Information bottleneck method for task-specific compression.
  • Figure 2: System model.
  • Figure 3: The procedure of video encoding.
  • Figure 4: The procedure of video decoding.
  • Figure 5: The visualization of the edge video system. Left: We use contour lines to display the perception coverage of different cameras. The small dots in the grid represent pedestrians, with different colors of the dots indicating the number of cameras covering each pedestrian. It can be observed that areas closer to the perception center of cameras are covered by more cameras. Right: The visualization of the raw video data and the legend for different numbers of covered cameras.
  • ...and 8 more figures

Theorems & Definitions (12)

  • Proposition 1
  • Proof 1
  • Proposition 2
  • Proof 2
  • Lemma 1
  • Proof 3
  • Theorem 1
  • Proof 4
  • Proposition 3
  • Proof 5
  • ...and 2 more