Rationalizing Transformer Predictions via End-To-End Differentiable Self-Training
This addresses the need for interpretable AI models in domains requiring transparency, though it is incremental by simplifying and stabilizing an existing three-player-game approach.
The paper tackles the problem of training transformer classifiers to produce rationales for their predictions without explicit supervision, achieving state-of-the-art alignment with human annotations.
We propose an end-to-end differentiable training paradigm for stable training of a rationalized transformer classifier. Our approach results in a single model that simultaneously classifies a sample and scores input tokens based on their relevance to the classification. To this end, we build on the widely-used three-player-game for training rationalized models, which typically relies on training a rationale selector, a classifier and a complement classifier. We simplify this approach by making a single model fulfill all three roles, leading to a more efficient training paradigm that is not susceptible to the common training instabilities that plague existing approaches. Further, we extend this paradigm to produce class-wise rationales while incorporating recent advances in parameterizing and regularizing the resulting rationales, thus leading to substantially improved and state-of-the-art alignment with human annotations without any explicit supervision.