A Multi-FPGA Scalable Framework for Deep Reinforcement Learning Through Neuroevolution

Laserna, J.J.; Otero, Andrés; Torre, Eduardo de la

doi:10.1007/978-3-031-19983-7_4

Cited by 2 publications

(1 citation statement)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As artificial intelligence (AI) advances quickly, more academics have begun utilizing deep reinforcement learning to implement algorithmic trading. Through its excellent feature representation ability of the deep neural network to fit the state, action, value, and other functions, deep rein-forcement learning has demonstrated strong learning ability and decision-making ability in autonomous driving [9], [10], job scheduling [11], [12], and game playing [13], [14]. In addition to being more advanced than human experts in particular domains, it also has the potential to achieve artificial general intelligence [15].…”

Section: Introductionmentioning

confidence: 99%

A Mean-VaR Based Deep Reinforcement Learning Framework for Practical Algorithmic Trading

Jin

2023

IEEE Access

View full text Add to dashboard Cite

It is difficult to automatically produce trading signals based on previous transaction data and the financial status of assets because of the significant noise and unpredictability of capital markets. This paper proposes an innovative algorithm to solve the optimal portfolio problem in stock market trading activities. Our novel portfolio trading strategy utilizes three features to outperform other benchmark strategies in a real-market environment. First, we propose a mean-VaR portfolio optimization model, the solution of which is based on the actor-critic architecture. Unlike the existing literature that learns the expectation of cumulative returns, the critic module learns the cumulative returns distribution by quantile regression, and the actor module outputs the optimal portfolio weight by maximizing the objective function of the optimization model. Secondly, we use a linear transformation function to realize short selling to ensure investors have profit opportunities in the bear market. Third, A multi-process method, called Ape-x, was used to accelerate the speed of deep reinforcement learning training. To validate our proposed approach, we conduct backtesting for two representative portfolios and observe that the proposed model in this work is superior to the benchmark strategies.

show abstract