Overview: Locally Weighted Ensembling

Locally Weighted Ensembling is a transductive parameter transfer learning framework intended to improve the learning of a target task T_T on a testing domain D_T, transferring knowledge from k models trained on k labeled domains of interest. For any example x, we can weight the model predictions according to their performance in the neighborhood of other examples clustered near x, making an overall prediction by constructing an ensemble which is weighted according to structural similarity near x in the test domain.

Relevant Research

This repository implements and analyzes results proposed in .