Abstract
When facing an unsatisfactory prediction from a machine learning model, users can be interested in investigating the underlying reasons and exploring the potential for reversing the outcome. We ask: To flip the prediction on a test point xt, how to identify the smallest training subset 𝒮t that we need to relabel?We propose an efficient algorithm to identify and relabel such a subset via an extended influence function for binary classification models with convex loss.We find that relabeling fewer than 2% of the training points can always flip a prediction.This mechanism can serve multiple purposes: (1) providing an approach to challenge a model prediction by altering training points; (2) evaluating model robustness with the cardinality of the subset (i.e., |𝒮t|); we show that |𝒮t| is highly related to the noise ratio in the training set and |𝒮t| is correlated with but complementary to predicted probabilities; and (3) revealing training points lead to group attribution bias. To the best of our knowledge, we are the first to investigate identifying and relabeling the minimal training subset required to flip a given prediction.- Anthology ID:
- 2024.findings-eacl.73
- Volume:
- Findings of the Association for Computational Linguistics: EACL 2024
- Month:
- March
- Year:
- 2024
- Address:
- St. Julian’s, Malta
- Editors:
- Yvette Graham, Matthew Purver
- Venue:
- Findings
- SIG:
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 1085–1098
- Language:
- URL:
- https://aclanthology.org/2024.findings-eacl.73
- DOI:
- Cite (ACL):
- Jinghan Yang, Linjie Xu, and Lequan Yu. 2024. Relabeling Minimal Training Subset to Flip a Prediction. In Findings of the Association for Computational Linguistics: EACL 2024, pages 1085–1098, St. Julian’s, Malta. Association for Computational Linguistics.
- Cite (Informal):
- Relabeling Minimal Training Subset to Flip a Prediction (Yang et al., Findings 2024)
- PDF:
- https://preview.aclanthology.org/nschneid-patch-1/2024.findings-eacl.73.pdf