Independent Reading · 2025

Playing Repeated Stochastic Security Games Against Non-Stationary Attackers

Ling Chen and Runfa Zhang · Mathematics

Security GamesNon-StationarityOpponent Modelling

Full text

Read the Paper

Use View PDF or Download PDF to read the full paper on this device.

Paper Information

Authors
Ling Chen and Runfa Zhang
Venue
Mathematics · 2025
Open official paper

Summary

This paper studies repeated security games where an attacker’s latent behavior type changes stochastically and without notification.

Core Idea

A defender can reuse a library of policies while updating beliefs to detect opponent-type switches and select an appropriate response.

Method

The authors formulate defender strategies for five attacker types, apply Bayesian Policy Reuse, and compare it experimentally with EXP3-S.

Key Takeaways

  • Fixed-attacker assumptions can be inadequate for repeated security interactions.
  • Belief-based policy reuse can detect behavioral switches and improve defender utility.
  • Maintaining a relevant policy and opponent-type library is an important practical dependency.

Questions

  • How does the approach respond to attacker behaviors outside the predefined type library?
  • Can change detection and policy generation be learned jointly?