ABSTRACT
Adversarial machine learning (ML) has been studied for more than a decade. Adversarial deep learning (DL) has also been researched for a decade. This paper gives a high-level overview of the trust and security of deep reinforcement learning (DRL) with its applications in automatic algorithmic trading to avoid financial instability. We first briefly introduce deep reinforcement learning (DRL). Then we address the most recent reviews of adversarial ML, DL, and DRL; we conceptually review the attacks and defenses of DRL systems to make the complex concepts much easier to understand.
To see how to preserve financial stability while applying AI techniques, particularly big data analytics for the FinTech application domain, the big data analytics pipeline has been extensively designed and developed on the public Cloud or the on-site private Cloud platforms. We present the robust DRL systems based on the secure Cloud platforms with possible proposed defense techniques against Cloud platform system attacks and adversarial DRL modeling attacks. Finally, we demonstrate a scenario of financial stability problems with an automatic algorithm trading system application by using a summarization table of previous proposed adversarial ML, DL, and DRL concerning this FinTech application. This study is an initial road map of trust and security DRL with its applications in the financial world.
