Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RoS-Guard: Robust and Scalable Online Change Detection with Delay-Optimal Guarantees

Zelin Zhu, Yancheng Huang, Kai Yang

TL;DR

RoS-Guard tackles online change detection in uncertain dynamic linear systems by formulating a robust generalized log-likelihood ratio (GLLR) and solving a robust optimization problem under polyhedral uncertainty. It introduces a tight MIQP formulation with a strong relaxation to a second-order cone program (SOCP) and leverages neural unrolling to enable GPU-accelerated, real-time inference. The framework provides theoretical guarantees on the false alarm period and worst-case detection delay, and empirical results on smart grids and MIMO channels show superior accuracy and substantial speedups over CPU-based baselines. This work enables scalable, robust OCD in large-scale systems with practical performance guarantees and accelerated computation suitable for real-time deployment.

Abstract

Online change detection (OCD) aims to rapidly identify change points in streaming data and is critical in applications such as power system monitoring, wireless network sensing, and financial anomaly detection. Existing OCD methods typically assume precise system knowledge, which is unrealistic due to estimation errors and environmental variations. Moreover, existing OCD methods often struggle with efficiency in large-scale systems. To overcome these challenges, we propose RoS-Guard, a robust and optimal OCD algorithm tailored for linear systems with uncertainty. Through a tight relaxation and reformulation of the OCD optimization problem, RoS-Guard employs neural unrolling to enable efficient parallel computation via GPU acceleration. The algorithm provides theoretical guarantees on performance, including expected false alarm rate and worst-case average detection delay. Extensive experiments validate the effectiveness of RoS-Guard and demonstrate significant computational speedup in large-scale system scenarios.

RoS-Guard: Robust and Scalable Online Change Detection with Delay-Optimal Guarantees

TL;DR

RoS-Guard tackles online change detection in uncertain dynamic linear systems by formulating a robust generalized log-likelihood ratio (GLLR) and solving a robust optimization problem under polyhedral uncertainty. It introduces a tight MIQP formulation with a strong relaxation to a second-order cone program (SOCP) and leverages neural unrolling to enable GPU-accelerated, real-time inference. The framework provides theoretical guarantees on the false alarm period and worst-case detection delay, and empirical results on smart grids and MIMO channels show superior accuracy and substantial speedups over CPU-based baselines. This work enables scalable, robust OCD in large-scale systems with practical performance guarantees and accelerated computation suitable for real-time deployment.

Abstract

Online change detection (OCD) aims to rapidly identify change points in streaming data and is critical in applications such as power system monitoring, wireless network sensing, and financial anomaly detection. Existing OCD methods typically assume precise system knowledge, which is unrealistic due to estimation errors and environmental variations. Moreover, existing OCD methods often struggle with efficiency in large-scale systems. To overcome these challenges, we propose RoS-Guard, a robust and optimal OCD algorithm tailored for linear systems with uncertainty. Through a tight relaxation and reformulation of the OCD optimization problem, RoS-Guard employs neural unrolling to enable efficient parallel computation via GPU acceleration. The algorithm provides theoretical guarantees on performance, including expected false alarm rate and worst-case average detection delay. Extensive experiments validate the effectiveness of RoS-Guard and demonstrate significant computational speedup in large-scale system scenarios.

Paper Structure

This paper contains 29 sections, 4 theorems, 84 equations, 6 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Statement
  4. Generalized Log-likelihood Ratio
  5. GLLR with System Uncertainty
  6. Relaxation for Efficient Optimization
  7. Neural Unrolling for GPU-acceleration
  8. Theoretical Analysis
  9. Condition to Meet Expected False Alarm Period Constraint
  10. Upper Bound on the Worst-Case Expected Detection Delay
  11. Experiments
  12. Datasets
  13. Performance Evaluation
  14. Evaluation of GPU Parallel Computing
  15. Numerical Verification of \ref{['theorem_2']} and \ref{['theorem_3']}
  16. ...and 14 more sections

Key Result

Theorem 1

Suppose $\mathbf{x}^{(t)}$ is upper bounded, i.e, $\| \mathbf{x}^{(t)}\|_2^2\leq \alpha$. The expected false alarm period of RoS-Guard is larger than $\gamma$ if :

Figures (6)

  • Figure 1: Illustration of the Unrolled Optimization Network
  • Figure 2: (a) Performance Comparison with Attack Injection Detection in Smart Grids System, (b) Performance Comparison with Various Uncertainty Sets.
  • Figure 3: Performance Comparison with Blockage Detection in Wireless MIMO System.
  • Figure 4: Detection Time Comparison between Neural Unrolling with GPU and MIQP with CPU in Different System Scale.
  • Figure 5: Numerical Verification of Theorems
  • ...and 1 more figures

Theorems & Definitions (10)

  • Theorem 1
  • proof : Proof Sketch
  • Theorem 2
  • proof : Proof Sketch
  • proof
  • Lemma 1
  • proof
  • Lemma 2
  • proof
  • proof