CyLab Talk
— 2:00pm
Location:
In Person
-
Mehrabian Collaborative Innovation Center, Room 2101
Speaker:
ROI BAR-ZUR
,
Ph.D. Candidate, Andrew and Erna Viterbi Faculty of Electrical & Computer Engineering, Technion, and, Member, The Initiative for Cryptocurrencies and Contracts
https://roibarzur.github.io/
Deep Reinforcement Learning: A New Frontier in Blockchain Security Analysis
Blockchain security is fundamentally anchored in the incentive system that promotes compliance among miners. However, the stability and security of these protocols face potential disruptions from factors such as variable transaction rewards and strategic deviations by miners. Deep Reinforcement Learning (RL), an advanced analytical tool, is deployed to dissect these intricate dynamics and understand their implications. Employing a novel deep reinforcement learning framework, the role of variable rewards in influencing miner behavior is effectively modeled and evaluated. This framework demonstrates that occasional large rewards can decrease the amount of computational power a miner needs to amass before it becomes advantageous for them to deviate from the protocol. An exploration of miner behavior also reveals that deviating miners may benefit from bribing others to act in their interest, further lowering the critical computational power ratio and posing a significant security concern. These findings underscore the value of sophisticated tools like deep RL in blockchain security analysis. They enable researchers and protocol designers to bring potential threats into focus and analyze modifications to bolster protocol resilience against exploitation.