Elastic weight consolidation for better bias inoculation

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 957–964 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 957 Elastic weight consolidation for better bias inoculation James Thorne Department of Computer Science University of Cambridge jt719@cam.ac.uk Andreas Vlachos Department of Computer Science University of Cambridge av308@cam.ac.uk Abstract The biases present in training datasets have been shown to affect models for sentence pair classification tasks such as natural language inference (NLI) and fact verification. While fine-tuning models on additional data has been used to mitigate them, a common issue is that of catastrophic forgetting of the original train- ing dataset. In this paper, we show that elas- tic weight consolidation (EWC) allows fine- tuning of models to mitigate biases while be- ing less susceptible to catastrophic forgetting. In our evaluation on fact verification and NLI stress tests, we show that fine-tuning with EWC dominates standard fine-tuning, yielding models with lower levels of forgetting on the original (biased) dataset for equivalent gains in accuracy on the fine-tuning (unbiased) dataset. 1 Introduction A number of recent works have illustrated short- comings in sentence-pair classification models that are used for Natural Language Inference (NLI). These arise from limited or biased train- ing data and the lack of suitable inductive bias in models. Naik et al. (2018) demonstrated that phenomena such as the presence of negation or a high degree of lexical overlap induce misclassifi- cations on models trained on the MultiNLI dataset (Williams et al., 2018). Poliak et al. (2018) and Gururangan et al. (2018) identified biases intro- duced during dataset construction that were ex- ploited by models to learn associations between the label and only one of the two input sen- tences without considering the other – known as hypothesis-only bias. Such biases also affect fact verification (Schus- ter et al., 2019), typically modeled as a text-pair classification between a claim and evidence re- trieved from a trusted source (Pomerleau and Rao, 2017; Thorne et al., 2018). Symmetric (Counterfactual) Evidence (Schuster et al., 2019) His debut solo studio album [...] was released on 28 April 2011 Evidence Model Predicts SUPPORTS ⚠ 1) Train model using FEVER dataset only 2) Illustrate test set accuracies on FEVER + Symmetric Before mitigating biases High accuracy on original task Low accuracy on counterfactual task Mitigating biases by fine-tuning on Symmetric Higher accuracy on counterfactual task Original task exhibits catatstophic forgetting Fine-tuning with EWC on Symmetric Higher accuracy on counterfactual task Catastrophic forgetting mitigated FEVER Dataset (Thorne et al., 2018) Damon Albarn 's debut album was released in 2014 His debut solo studio album [...] was released on 28 April 2014 Evidence Model Predicts SUPPORTS ✔ Figure 1: Hypothesis only bias in FEVER contributes to low accuracy when testing against counterfactual ev- idence. This is mitigated by fine-tuning on counter- factual evidence. Catastrophic forgetting from fine- tuning is reduced when using elastic weight consoli- dation (EWC), preserving the original task accuracy. To mitigate these undesirable behaviors Liu et al. …

Content truncated to 500 words