2020/12/19 by James Queeney, Ioannis Ch. Paschalidis, Queeney, James +3
Computer Science · #Advanced Neural Network Applications #Adversarial Robustness in Machine Learning #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Data Classification
paper · pdf · doi:10.48550/arxiv.2012.10791
openalex publication_date 2020/12/19 · openalex created_date 2022/09/08 · openalex updated_date 2026/07/28
In order for reinforcement learning techniques to be useful in real-world\ndecision making processes, they must be able to produce robust performance from\nlimited data. Deep policy optimization methods have achieved impressive results\non complex tasks, but their real-world adoption remains limited because they\noften require significant amounts of data to succeed. When combined with small\nsample sizes, these methods can result in unstable learning due to their\nreliance on high-dimensional sample-based estimates. In this work, we develop\ntechniques to control the uncertainty introduced by these estimates. We\nleverage these techniques to propose a deep policy optimization approach\ndesigned to produce stable performance even when data is scarce. The resulting\nalgorithm, Uncertainty-Aware Trust Region Policy Optimization, generates robust\npolicy updates that adapt to the level of uncertainty present throughout the\nlearning process.\n