Fully Dynamic Algorithm for Constrained Submodular Optimization

  • Silvio Lattanzi ,
  • Slobodan Mitrović ,
  • Ashkan Norouzi-Fard ,
  • ,
  • Morteza Zadimoghaddam

NeurIPS 2020 |

PDF

The task of maximizing a monotone submodular function under a cardinality constraint is at the core of many machine learning and data mining applications, including data summarization, sparse regression and coverage problems. We study this classic problem in the fully dynamic setting, where elements can be both inserted and removed. Our main result is a randomized algorithm that maintains an efficient data structure with a poly-logarithmic amortized update time and yields a (1/2ϵ)-approximate solution. We complement our theoretical analysis with an empirical study of the performance of our algorithm.