Learning to Deceive Knowledge Graph Augmented Models via Targeted Perturbation
Abstract
Knowledge graphs (KGs) have helped neural models improve performance on various knowledge-intensive tasks, like question answering and item recommendation. By using attention over the KG, such KG-augmented models can also "explain" which KG information was most relevant for making a given prediction. In this paper, we question whether these models are really behaving as we expect. We show that, through a reinforcement learning policy (or even simple heuristics), one can produce deceptively perturbed KGs, which maintain the downstream performance of the original KG while significantly deviating from the original KG's semantics and structure. Our findings raise doubts about KG-augmented models' ability to reason about KG information and give sensible explanations.
Cite
Text
Raman et al. "Learning to Deceive Knowledge Graph Augmented Models via Targeted Perturbation." International Conference on Learning Representations, 2021.Markdown
[Raman et al. "Learning to Deceive Knowledge Graph Augmented Models via Targeted Perturbation." International Conference on Learning Representations, 2021.](https://mlanthology.org/iclr/2021/raman2021iclr-learning/)BibTeX
@inproceedings{raman2021iclr-learning,
title = {{Learning to Deceive Knowledge Graph Augmented Models via Targeted Perturbation}},
author = {Raman, Mrigank and Chan, Aaron and Agarwal, Siddhant and Wang, PeiFeng and Wang, Hansen and Kim, Sungchul and Rossi, Ryan and Zhao, Handong and Lipka, Nedim and Ren, Xiang},
booktitle = {International Conference on Learning Representations},
year = {2021},
url = {https://mlanthology.org/iclr/2021/raman2021iclr-learning/}
}