Quickest Change Detection with Cost-Constrained Experiment Design

2025-09-17

AI Summary

This paper addresses the quickest change detection problem under multi-experiment selection: at each time step, a decision-maker dynamically selects an experiment from multiple heterogeneous observation sources—differing in information quality and sampling cost—to minimize the worst-case average detection delay, subject to constraints on false alarm rate and total sampling cost. To this end, we propose 2E-CUSUM—the first sequential detection framework supporting experiment selection, data skipping, and zero-observation decisions. By integrating a data-efficiency mechanism and a cost-aware CUSUM statistic, 2E-CUSUM achieves joint asymptotic optimality in balancing detection delay, false alarm control, and cost efficiency. Theoretically, it attains the minimal achievable delay order under composite pre- and post-change hypotheses. Extensive simulations demonstrate that 2E-CUSUM significantly outperforms baseline methods, reducing detection delay while saving over 30% of observational resources.

Technology Category

• Reasoning under Uncertainty: Sequential Decision Making
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

In the classical quickest change detection problem, an observer performs only one experiment to monitor a stochastic process. This paper considers the case where, at each observation time, the decision-maker needs to choose between multiple experiments with different information qualities and costs. The goal is to minimize the worst-case average detection delay subject to false alarm and cost constraints. An algorithm called the 2E-CUSUM Algorithm has been developed to achieve this goal for the two-experiment case. Extensions to multiple-experiment designs are also studied, and 2E-CUSUM is extended accordingly. Data efficiency, where the observer has the choice not to perform an experiment, is explored as well. The proposed algorithms are analyzed and shown to be asymptotically optimal.

Problem

• Optimizing experiment selection with varying costs and information qualities
• Minimizing worst-case detection delay under false alarm constraints
• Developing efficient algorithms for cost-constrained change detection scenarios

Innovation

• Cost-constrained multiple experiment selection
• 2E-CUSUM algorithm for detection optimization
• Asymptotically optimal multi-experiment design