Adversarial Reinforcement Learning in Dynamic Channel Access and Power Control

Wang, Feng; Gursoy, M. Cenk; Velipasalar, Senem

doi:10.1109/wcnc49053.2021.9417271

“…Authors assume that the victim pattern is static during the learning phase. An extension to simultaneously attacks multiple channels has been proposed in [25], where the authors consider multiple dynamic channels. Their approach is based on a Deep Q-Network (DQN) approach with the main objective to reduce the sum-rate of the victim node.…”

Section: Related Literaturementioning

confidence: 99%

FOLPETTI: A Novel Multi-Armed Bandit Smart Attack for Wireless Networks

Bout,

Brighente,

Conti

et al. 2022

Preprint

0

View full text Add to dashboard Cite

Channel hopping provides a defense mechanism against jamming attacks in large scale Internet of Things (IoT) networks. However, a sufficiently powerful attacker may be able to learn the channel hopping pattern and efficiently predict the channel to jam.In this paper, we present FOLPETTI, a Multi-Armed Bandit (MAB)-based attack to dynamically follow the victim's channel selection in real-time. Compared to previous attacks implemented via Deep Reinforcement Learning (DRL), FOLPETTI does not require recurrent training phases to capture the victim's behavior, allowing hence a continuous attack. We assess the validity of FOLPETTI by implementing it to launch a jamming attack. We evaluate its performance against a victim performing random channel selection and a victim implementing a MAB defence strategy. We assume that the victim detects an attack when more than 20% of the transmitted packets are not received, therefore this represents the limit for the attack to be stealthy. In this scenario, FOLPETTI achieves a 15% success rate for the victim's random channel selection strategy, close to the 17.5% obtained with a genie-aided approach.Conversely, the DRL-based approach reaches a success rate of 12.5%, which is 5.5% less than FOLPETTI. We also confirm the results by confronting FOLPETTI with a MAB based channel hopping method. Finally, we show that FOLPETTI creates an additional energy demand independently from its success rate, therefore decreasing the lifetime of IoT devices. CCS CONCEPTS• Security and privacy → Mobile and wireless security; • Networks → Denial-of-service attacks.

show abstract

“…Authors assume that the victim pattern is static during the learning phase. An extension to simultaneously attack multiple channels has been proposed in [25], where the authors consider multiple dynamic channels. Their approach is based on a Deep Q-Network (DQN) approach with the main objective to reduce the sum-rate of the victim node.…”

Section: Related Literaturementioning

confidence: 99%

FOLPETTI: A Novel Multi-Armed Bandit Smart Attack for Wireless Networks

Bout

¹

,

Brighente

²

,

Conti

³

et al. 2022

Proceedings of the 17th International Conference on Availability, Reliability and Security

4

0

View full text Add to dashboard Cite

Channel hopping provides a defense mechanism against jamming attacks in large scale Internet of Things (IoT) networks. However, a sufficiently powerful attacker may be able to learn the channel hopping pattern and efficiently predict the channel to jam.In this paper, we present FOLPETTI, a Multi-Armed Bandit (MAB)-based attack to dynamically follow the victim's channel selection in real-time. Compared to previous attacks implemented via Deep Reinforcement Learning (DRL), FOLPETTI does not require recurrent training phases to capture the victim's behavior, allowing hence a continuous attack. We assess the validity of FOLPETTI by implementing it to launch a jamming attack. We evaluate its performance against a victim performing random channel selection and a victim implementing a MAB defence strategy. We assume that the victim detects an attack when more than 20% of the transmitted packets are not received, therefore this represents the limit for the attack to be stealthy. In this scenario, FOLPETTI achieves a 15% success rate for the victim's random channel selection strategy, close to the 17.5% obtained with a genie-aided approach.Conversely, the DRL-based approach reaches a success rate of 12.5%, which is 5.5% less than FOLPETTI. We also confirm the results by confronting FOLPETTI with a MAB based channel hopping method. Finally, we show that FOLPETTI creates an additional energy demand independently from its success rate, therefore decreasing the lifetime of IoT devices. CCS CONCEPTS• Security and privacy → Mobile and wireless security; • Networks → Denial-of-service attacks.

show abstract

“…Finally, we note that there is also a growing interest in developing defense strategies against adversarial attacks [24]- [26], and specifically jamming attacks [27], [28]. One of the most intriguing defense strategies is to ensemble several different policies, explore alternative strategies, and provide stable performance [17], [29].…”

Section: Introductionmentioning

confidence: 99%