Modified Group Generalized Binary Search with Near-Optimal Performance Guarantees

Group Generalized Binary Search (Group GBS) is an extension of the well known greedy algorithm GBS, for identifying the group of an unknown object while minimizing the number of binary questions posed about that object. This problem referred to as group identification or the Equivalence Class determination problem arises in applications such as disease diagnosis, toxic chemical identification, and active learning under persistent noise. Here, we propose a modified version of Group GBS and prove that it is competitive with the optimal algorithm. Our result holds even in the case where the queries have unequal costs.