Section: New Results

A factorized version space algorithm for interactive database exploration

One challenge in building an interactive database exploration system is that existing active learning (AL) techniques experience slow convergence when learning the user interest on large datasets. To address this slow convergence problem, we augmented version space-based AL algorithms, which have strong theoretical results on convergence but are very costly to run, with additional insights obtained in the user labeling process. These insights lead to a novel algorithm that factorizes the version space to perform active learning in a set of subspaces, with provable results on optimality, as well as optimizations for better performance. Evaluation results using real world datasets show that our algorithm significantly outperforms state-of-the-art version space algorithms, as well as our previous data exploration algorithm DSM (Huang et al., PVLDB 2018), for large database exploration.

The above work was accepted as a conference paper at ICDM 2019 [14]. In addition, we have presented a demonstration of our software at NeurIPS 2019 [26], where people could interact with our system over two real-world datasets, and also observe how our system compares against traditional AL algorithms.