Striking a Balance: Alleviating Inconsistency in Pre-trained Models for Symmetric Classification Tasks
This work solves inconsistency issues in symmetric classification tasks for NLP practitioners, though it is incremental as it builds on existing fine-tuning paradigms.
The paper addresses inconsistency in pre-trained models for symmetric classification tasks where predictions should be invariant to input order, applying a consistency loss function to improve prediction consistency on three paraphrase detection datasets without significant accuracy drops.
While fine-tuning pre-trained models for downstream classification is the conventional paradigm in NLP, often task-specific nuances may not get captured in the resultant models. Specifically, for tasks that take two inputs and require the output to be invariant of the order of the inputs, inconsistency is often observed in the predicted labels or confidence scores. We highlight this model shortcoming and apply a consistency loss function to alleviate inconsistency in symmetric classification. Our results show an improved consistency in predictions for three paraphrase detection datasets without a significant drop in the accuracy scores. We examine the classification performance of six datasets (both symmetric and non-symmetric) to showcase the strengths and limitations of our approach.