K-means Clustering

Large datasets often contain overly similar examples that consume training cycles without contributing to learning. A new paper identifies similar training examples, even if they’re not labeled.

If you’re standing close to others at a party, it’s likely you have something in common. This is the idea behind using k-means clustering to split data points into groups. Whether the groups formed via human agency or some other force, this algorithm will find them.

School teachers may not like to hear this, but sometimes you get the best answer by peeking at your neighbor’s paper. A new language model framework peeks at the training data for context when making a prediction.