A Hierarchical Spectral Method for Extreme Classification
share
Extreme classification problems are multiclass and multilabel classification problems where the number of outputs is so large that straightforward strategies are neither statistically nor computationally viable. One strategy for dealing with the computational burden is via a tree decomposition of the output space. While this typically leads to training and inference that scales sublinearly with the number of outputs, it also results in reduced statistical performance. In this work, we identify two shortcomings of tree decomposition methods, and describe two heuristic mitigations. We compose these with an eigenvalue technique for constructing the tree. The end result is a computationally efficient algorithm that provides good statistical performance on several extreme data sets.
READ FULL TEXT