Significance Comment

Being able to train in genetic programming effectively while not requiring constant reevaluation of the entire training set is significant. The results stimulate further research into the question, "what are the optimal instances to train on?" The results show increase in efficiency when using random sampling, so more selected sampling could yield even better results.

Quality Comment

The paper is highly readable, and provides a good overview of the importance of increasing generalization and reducing overfitting in Genetic Programming. The contribution is explained in sufficient detail, enough to allow replication.

Summary

The publication is a good introductory article into consider generalization in genetic programming. It should be easily understandable, even if your knowledge in genetic programming is limited. Their contribution is simple, with good results.

My own experience is as a PhD student in computer science, currently focusing on a thesis related to increasing the efficiency of genetic programming through exploitation of knowledge acquired during training that is normally ignored.