kNN
K-Nearest Neighbors, an instance-based machine learning method that makes predictions by finding the K historical samples most similar to a new sample.
Timeline (last 90 days)
The kNN method based on bge-base embeddings (predicting using the outcomes of the 50 most similar historical cases) showed flat performance across all quintiles, with no discriminative power.
A 2023 paper demonstrated that gzip + kNN outperformed some BERT variants on few-shot text classification tasks.
Subsequent research pointed out that the advantage of gzip + kNN diminishes significantly on large datasets.
All Facts (3)
The kNN method based on bge-base embeddings (predicting using the outcomes of the 50 most similar historical cases) showed flat performance across all quintiles, with no discriminative power.
50%UnverifiedA 2023 paper demonstrated that gzip + kNN outperformed some BERT variants on few-shot text classification tasks.
50%UnverifiedSubsequent research pointed out that the advantage of gzip + kNN diminishes significantly on large datasets.
50%