Change point detection in multi-agent systems based on higher-order features.

Area of Science:

• Complex Systems Science
• Network Science
• Data Analysis

Background:

• Multi-agent systems generate high-dimensional N x d x T data, challenging traditional multivariate change point detection (CPD).
• Existing CPD methods struggle with the complexity and scale of data from numerous interacting agents.

Purpose of the Study:

• To develop novel methods for dimensionality reduction in multi-agent system data for effective change point detection.
• To address the challenge of missing data in high-dimensional time series analysis within multi-agent systems.

Main Methods:

• Constructing topological structures using Vietoris-Rips complexes on time-slice snapshots.
• Extracting higher-order features, specifically Betti numbers and persistence, to compress d-dimensional data into a single dimension.
• Applying general CPD methods to the reduced feature space.

Main Results:

• Demonstrated significant performance improvements in change point detection for multi-agent systems.
• Successfully compressed high-dimensional features using topological invariants (Betti numbers and persistence).
• Showcased the robustness of the proposed methods in handling missing data due to feature independence across snapshots.

Conclusions:

• The proposed Betti number and persistence feature extraction methods offer a new approach for dimensionality reduction and missing data handling in multi-agent systems.
• These topological feature-based methods provide a valuable tool for enhancing change point detection in complex systems.
• The techniques show potential for broader applications in fields like complex network analysis.