Utility Learning

Utility learning focuses on inferring an agent's preferences, represented as a utility function, from observed behavior or data, aiming to predict future choices and optimize decision-making. Current research emphasizes efficient algorithms for learning utility functions from various data sources, including demonstrations, pairwise comparisons, and reward signals, often employing techniques like policy gradients, Bayesian optimization, and graph neural networks. This field is crucial for advancing reinforcement learning, recommender systems, and decision support systems by enabling personalized and efficient optimization under uncertainty and multiple objectives. The ability to accurately model and learn utility functions has significant implications for various applications, including healthcare, economics, and human-computer interaction.