In local search, gradient descent, one update is very costly in large datasets
https://podcast.ucsd.edu/watch/fa23/cse151a_a00/11