all AI news
Double Trouble: How to not explain a text classifier's decisions using counterfactuals synthesized by masked language models?. (arXiv:2110.11929v3 [cs.CL] UPDATED)
Aug. 17, 2022, 1:11 a.m. | Thang M. Pham, Trung Bui, Long Mai, Anh Nguyen
cs.CL updates on arXiv.org arxiv.org
A principle behind dozens of attribution methods is to take the prediction
difference between before-and-after an input feature (here, a token) is removed
as its attribution. A popular Input Marginalization (IM) method (Kim et al.,
2020) uses BERT to replace a token, yielding more plausible counterfactuals.
While Kim et al. (2020) reported that IM is effective, we find this conclusion
not convincing as the DeletionBERT metric used in their paper is biased towards
IM. Importantly, this bias exists in Deletion-based …
More from arxiv.org / cs.CL updates on arXiv.org
Jobs in AI, ML, Big Data
Senior Machine Learning Engineer (MLOps)
@ Promaton | Remote, Europe
Data Analyst
@ SEAKR Engineering | Englewood, CO, United States
Data Analyst II
@ Postman | Bengaluru, India
Data Architect
@ FORSEVEN | Warwick, GB
Director, Data Science
@ Visa | Washington, DC, United States
Senior Manager, Data Science - Emerging ML
@ Capital One | McLean, VA