Reviews: Efficient Loss-Based Decoding on Graphs for Extreme Classification

This paper proposes an algorithm to solve extreme multi-class classification problems using Error Correcting Output Coding. During training, the algorithm simply learns l (logK) independent binary classifiers. Main contribution of the paper is in the inference algorithm. It reduces the costly loss based decoding framework for ECOC to that of finding shortest path on a weighted trellis graph. The paper is well written and easy to understand.