What is GOSS in LightGBM?

Question: What is the purpose of Gradient-based One-Side Sampling, or GOSS, in LightGBM? A) To remove all high-gradient examples B) To convert trees into neural networks C) To replace gradient boosting with bagging D) To keep more large-gradient examples while sampling smaller-gradient examples to speed up training Correct: D Explanation: GOSS keeps examples with large gradients because they are more informative for training, while sampling from smaller-gradient examples. This can reduce computation while preserving useful training signal. Topic: advanced ML / gradient boosting / LightGBM