The financial industry has evolved dramatically over the lastdecades and continues to face more and more challenges amidrapidly increased competitions, technologies developmentsand complex political and economic environments. To facethese challenges, how to apply Machine Learning methodsto develop the profitable winning strategies in the algorithmstock trading and portfolio management, framed in termsof risk-factor exposure, dynamic environment, as opposedto asset classes, is becoming a more important problem.The success of Deep Learning has achieved a lot of stateof the art performances in many industries, from computervision,natural language processing to autonomous drivingand robotics application. At the same time, reinforcementlearning also has gained a lot of attention from academiccommunities to various industries, combined with the tech-niques from deep learning, Deep Reinforcement Learninghave created an intelligent agent to interact with an unknowndynamic environment and aimed to maximize its cumulativerewards. By using the methods from deep learning, DeepReinforcement Learning is learning through trial and errorfrom simultaneously sequential and evaluative feedback andsampled by nonlinear function approximation. Based on thesecharacteristics, Deep Reinforcement Learning is a promisingapproach in algorithm trading and portfolio management.