A Reproduction of Green Recommender Systems

This paper presents a reproduction of the paper Green Recommender Systems. The goal of the paper was to find out if Recommender Systems could be trained with fewer data to save power while still achieving satisfying results. The authors of the original paper found that it depends on the type of algorithms used and divided those by their ability to be reduced. Our reproduction aims at verifying those results and tries to translate them to some general machine learning algorithms outside the realm of recommender systems. We tested the same idea on a set of additional algorithms and a wider range of datasets to test if those findings are not just restricted to the originally used algorithms and datasets. We found similar results for certain algorithms that tend to respond more positively to downsampling compared to others.