# Adaptive Learning for Moving Target defence: Enhancing Cybersecurity Strategies

arXiv: https://arxiv.org/abs/2508.17945 · HTML: https://mlsanity.com/paper/2508.17945/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Adversarial robustness and RL in finance, one of 15 directions in Deep reinforcement learning methods.
- Builds on 5 of its 15 references that are in the map.
- First posted on arXiv 25 Aug 2025.

## Placed in

- Deep reinforcement learning methods / [Adversarial robustness and RL in finance](https://mlsanity.com/deeprl/adversarial-robustness-and-rl-in-finance/)

## Builds on (most cited first, 5 of 5)

- [On the convergence of policy gradient methods to Nash equilibria in general stochastic games](https://mlsanity.com/paper/2210.08857/) (2022)
- [Multi-agent Reinforcement Learning in Bayesian Stackelberg Markov Games for Adaptive Moving Target Defense](https://mlsanity.com/paper/2007.10457/) (2020)
- [Learning Near-Optimal Intrusion Responses Against Dynamic Attackers](https://mlsanity.com/paper/2301.06085/) (2023)
- [A Survey of Moving Target Defenses for Network Security](https://mlsanity.com/paper/1905.00964/) (2019)
- [RL and Fingerprinting to Select Moving Target Defense Mechanisms for Zero-day Attacks in IoT](https://mlsanity.com/paper/2212.14647/) (2022)

## Abstract (arXiv)

In this work, we model Moving Target Defence (MTD) as a partially observable stochastic game between an attacker and a defender. The attacker tries to compromise the system through probing actions, while the defender minimizes the risk by reimaging the system, balancing between performance cost and security level. We demonstrate that the optimal strategies for both players follow a threshold structure. Based on this insight, we propose a structure-aware policy gradient reinforcement learning algorithm that helps both players converge to the Nash equilibrium. This approach enhances the defender's ability to adapt and effectively counter evolving threats, improving the overall security of the system. Finally, we validate the proposed method through numerical simulations.