Fintech+AI

We apply deep reinforcement learning on portfolio management to model profits and risk with realistic transaction cost model. Using the financial domain-knowledge that our team has, we present special price representation method in order to extract financial market trend more effectively. Aside from the specialized data representation method, we also construct novel model structure that based on multi-agent reinforcement learning framework and combine risk-oriented reward function as the optimization criterion of agent. Our purpose is to learn an agent with risk-shifting behaviors as the building block of novel portfolio management system. In addition, we develop several key technology in FinTech field including graph embedding-based recommendation system, blockchain application platform, information security and biometric based on adversarial attack.

Our reasearch lab is devocated in the developments the technologies of Monte Carlo simulation and machine learning. We target to apply our technologies in the applications of FinTech in the banking, securities, and insurance industries.

This project integrates Behavioral Finance, Financial Engineering, Artificial Intelligence, Neural Networks methodologies into the applications of Robo-Advisor, portfolio management, financial asset trading strategies, valuation and hedging in complex bonds and interest rate derivatives, sports game forcasting.

We propose to construct a high performance computing environment based on Nvidia’s Compute Unified Device Architecture (CUDA). To the extent, we can train the deep market calibration models proposed by Sub-project 5 with the effectiveness and time-saving advantages.