Group fairness in non-monotone submodular maximization

Summary: Many AI/Machine learning problems involve adaptively selecting a sequence of items with the goal of maximizing an adaptive submodular function. Existing studies mainly focus on monotone or non-monotone cases. This research generalizes previous results and shows that even for a non-monotone utility function, a random greedy policy can achieve an approximation ratio close to (1 - 1/e) if the function is close to a monotone function.

Summary: The article discusses approximation algorithms for k-submodular maximization problem under two types of constraints, providing corresponding time complexity and approximation ratio.