Essay.08-XingpingYu-2100017812
Keywords: utility, reward function, reinforcement learning
Abstract: Utility function plays an important role in human decision making. But these utility functions are internal to humans, which are hard to observe and represent in a specific way. Moreover, utility functions can vary from different individuals. In another word they are quite subjective and hard to measure with a consistent standard. When it comes to different tasks, there are more kinds of utility functions
behind the decision-making process. For computational models, we are trying to build some general frameworks to learn and represent the utility functions that are close to human utility. Some possible ways are discussed in this essay. This essay mainly proposes some possible utility functions like cost, human preference etc and learning policies like deep learning and reinforcement learning. For further analysis, this essay discusses some advantages and disadvantages of the ways in data collection, generalization and efficiency.
Submission Number: 163
Loading