Noisy Self-Knowledge Distillation for Text Summarization
This work addresses text summarization, a key task in natural language processing, by improving model training robustness, though it appears incremental as it builds on existing distillation techniques.
The paper tackles the problem of training text summarization models on single reference and noisy datasets by applying self-knowledge distillation with smoothed labels and multiple noise signals, achieving state-of-the-art results on three benchmarks.
In this paper we apply self-knowledge distillation to text summarization which we argue can alleviate problems with maximum-likelihood training on single reference and noisy datasets. Instead of relying on one-hot annotation labels, our student summarization model is trained with guidance from a teacher which generates smoothed labels to help regularize training. Furthermore, to better model uncertainty during training, we introduce multiple noise signals for both teacher and student models. We demonstrate experimentally on three benchmarks that our framework boosts the performance of both pretrained and non-pretrained summarizers achieving state-of-the-art results.