Zero Order Optimization Methods with Applications to Reinforcement Learning ｜Jorge Nocedal скачать в хорошем качестве

Zero Order Optimization Methods with Applications to Reinforcement Learning ｜Jorge Nocedal

Jorge Nocedal explained Zero-Order Optimization Methods with Applications to Reinforcement Learning. In applications such as adversarial training and reinforcement learning, it is necessary to optimize a function whose derivatives are not available. In most of these cases function evaluations contain noise. The question is then how to best perform the optimization in these circumstances. After reviewing the main approaches in the literature , he focused on two strategies that aim at estimating gradients. One is based on Gaussian smoothing, and he illustrated its performance on some reinforcement learning tasks. The second strategy makes use of a noise estimation procedure due to Hamming, and constructs quadratic models of the noisy objective. He concluded with some open questions in this field. BAAI (Beijing Academy of Artificial Intelligence, https://www.baai.ac.cn/) is a non-profit research institute, encouraging scientists to promote revolutionary or disruptive breakthroughs in AI theories, methodologies, tools, systems and applications. The BAAI Conference (https://2020.baai.ac.cn/) is committed to promoting international exchange and cooperation for the development of Artificial Intelligence, and building a platform for all our AI friends from all around the world to exchange ideas, discuss the challenges in front of the technology innovation and look into a better future driven by AI technology innovation. More About BAAI Twitter：  / baaizhiyuan   LinkedIn：  / baaibeijing   Youtube 链接：   • The New Science of Cause and Effect with R...