Speaker:

Yang Yu, Associate Professor, Nanjing University(LAMDA Lab )

Abstract:

Reinforcement learning achieved significant successes include being part of the AlphaGo system and playing Atari games. However, it is also criticized for applicability only in virtual worlds due to the requirement of huge amount of interaction data. In this talk, we will report our recent experience towards real-world reinforcement learning, including virtualizing real-world environments and reusing virtual-world policies in the real world.

Bio:

Yang Yu is an associate professor of computer science in Nanjing University, China. He joined the LAMDA Group as a faculty since he got his Ph.D. degree in 2011. His research area is in machine learning and reinforcement learning. He was recommended as AI’s 10 to Watch by IEEE Intelligent Systems in 2018, invited to have an Early Career Spotlight talk in IJCAI’18 on reinforcement learning, and received the Early Career Award of PAKDD in 2018.

Speaker :

Zeng Zhao, Senior Engineer, Fuxi AI Lab

Abstract:

Reinforcement Learning has achieved many milestones in recent years. However, the reliance on prior knowledge and engineering practice hinders its application and promotion in practical areas. We have implemented a reinforcement learning framework that combines traditional programming with visual programming. This framework abstracts the reinforcement learning process and the allocation & scheduling of computing resources at a higher level, effectively reducing the threshold of use, so that practitioners without domain specific knowledge can also easily use reinforcement learning to solve practical problems in many industrial fields such as games, autopilots, e-commerce, etc.

The main features of the framework include:

1. A visual modeling approach for industrial problem

2. Convenient for the decomposition of complicated problems

3. Scalable cloud computing resources

4. Integration with diverse reinforcement learning algorithms

Currently, this framework has been applied to a variety of game products developed by NetEase independently, so as to help traditional game developers build smarter game AI.

Bio:

Zeng Zhao is a senior deep learning system engineer at Netease's Fuxi Lab, focusing on designing a more versatile and efficient deep learning platform. He currently leads an AI cloud platform project for multiple domains, providing deeper learning algorithms with more efficient, stable and flexible computing resources through container and distributed computing. He received his Ph.D. in computer science from the University of Science and Technology of China, and his research direction is high performance computing. He has designed systems and optimized algorithms for several high performance computing platforms. In addition, Zhao Zeng was the co-founder of QXJoy and led the research and development of several products.

Speaker:

Jianye Hao, Associate Professor, Tianjin University

Abstract:

With the advance of deep learning, deep reinforcement learning have become quite popular in AI research community and industry as well. Deep reinforcement learning is viewed as one of the most promising way to achieve artificial general intelligence. Recent years have witnessed a wide range of DRL applications in robotics, e-commerce, and games and so on. Compared with single-agent learning, concurrent learning among multiple agents (aka multiagent learning) are more common in our human society and human-made systems and have received more attention in the last two years.

In this talk, I will first introduce the background of multiagent reinforcement learning, and then talk about recent research progresses in deep multiagent reinforcement learning and its applications.

Bio:

Jianye Hao is currently an associate professor at Tianjin University. Before that, he was a Postdoctoral Research Fellow in CSAIL at MIT and SUTD from 2013 to 2015. He received his Ph.D degree from The Chinese University of Hong Kong in 2013. Dr. Hao’s research interest mainly focuses on multiagent systems and reinforcement learning. He has published over 60 papers in top journals and conference proceedings including NIPS, AAMAS, IJCAI, AAAI, TPAMI, and authored two books on multiagent systems.

Speaker:

Mingjie Zhu, Founder and CEO of CraiditX

Abstract:

We will introduce our study and practice of reinforcement learning in finance industry like quantitative trading and anti-fraud detection. We have implemented the state-of-the-art reinforcement learning framework to address these problems. In this talk we will share the evaluation results comparing with traditional methods.

Bio:

Dr. Mingjie Zhu is the founder and CEO of CraiditX, a leading AI company in Chinese financial industry. Dr. Mingjie Zhu received his B.E. from the Special Class for Gifted Young, University of Science & Technology of China in 2004. He received his Ph.D. degree from Microsoft Research Asia in 2009. After that he worked as a researcher at Max Planck Institute with Dr. Gerhard Weikum. Then he moved to industry and worked as research scientist in Yahoo! Labs, eBay search. He then became the founding director of the big data department in Ctrip. He has over 15 years’ research and development experience in machine learning applications on massive data sets .In 2015 he started CraiditX． Dr. Mingjie Zhu is the visiting professor or graduate supervisor at University of Florida, University of Electronic Science and Technology of China and several other universities. He was also recognized as one of the 35 innovators under 35 by MIT Technology Review in 2017.

Speaker:

Zhiwei (Tony) Qin, Algorithm Researcher, Didi Research America

Abstract:

Multi-driver dispatching and repositioning in ride-sharing is a spatially extended and dynamic resource allocation problem that is intractable to solve exactly. Heuristic solutions construct an approximation to the true problem by ignoring the spatial extent, or the temporal dynamics, or both, and solve the approximate problem exactly. In this paper we develop a neural network architecture that uses a global representation of state and learns single-driver and multi-driver action-value-functions to continuously make non-myopic dispatching and repositioning assignments. We evaluate and demonstrate the advantages of these methods first on a static version of the dispatching problem, second on illustrative multi-driver dynamic dispatching and repositioning problems, and finally on data-driven dispatching and repositioning simulators.

Bio:

Tony Qin leads the reinforcement learning research at DiDi AI Labs. He received his Ph.D. in Operations Research from Columbia University and B.Sc. in Computer Science and Statistics from the University of British Columbia, Vancouver. Tony is broadly interested in research topics at the intersection of optimization and machine learning, and most recently in reinforcement learning and its applications in operational optimization, digital marketing, traffic signals control, and education. He has been on the program committees of a number of top-tier journals and conferences in machine learning and operations research, including AAAI, SDM, JMLR, TPAMI, TKDE.