Joint learning of interpretation and distillation

TL;DR

This paper explores a new approach to jointly learn model interpretation and distillation, demonstrating that imitating explanations can enhance both interpretability and predictive accuracy, especially for models with different structures.

Contribution

It introduces a method for simultaneous learning of model explanations and predictions, improving both interpretability and distillation performance for heterogeneous models like GBDT2NN.

Findings

Improved prediction accuracy through explanation imitation

Enhanced interpretability of GBDT2NN models

Better performance on benchmark datasets

Abstract

The extra trust brought by the model interpretation has made it an indispensable part of machine learning systems. But to explain a distilled model's prediction, one may either work with the student model itself, or turn to its teacher model. This leads to a more fundamental question: if a distilled model should give a similar prediction for a similar reason as its teacher model on the same input? This question becomes even more crucial when the two models have dramatically different structure, taking GBDT2NN for example. This paper conducts an empirical study on the new approach to explaining each prediction of GBDT2NN, and how imitating the explanation can further improve the distillation process as an auxiliary learning task. Experiments on several benchmarks show that the proposed methods achieve better performance on both explanations and predictions.