Rethinking Perturbations in Encoder-Decoders for Fast Training
This work addresses training time efficiency for researchers and practitioners using encoder-decoder models, but it is incremental as it focuses on optimizing existing perturbation approaches rather than introducing a new paradigm.
The study tackled the computational inefficiency of existing perturbation methods like scheduled sampling and adversarial perturbations in encoder-decoder training by comparing them to simpler techniques such as word dropout and random token replacement, finding that these simpler methods achieve comparable or better scores while being faster.
We often use perturbations to regularize neural models. For neural encoder-decoders, previous studies applied the scheduled sampling (Bengio et al., 2015) and adversarial perturbations (Sato et al., 2019) as perturbations but these methods require considerable computational time. Thus, this study addresses the question of whether these approaches are efficient enough for training time. We compare several perturbations in sequence-to-sequence problems with respect to computational time. Experimental results show that the simple techniques such as word dropout (Gal and Ghahramani, 2016) and random replacement of input tokens achieve comparable (or better) scores to the recently proposed perturbations, even though these simple methods are faster. Our code is publicly available at https://github.com/takase/rethink_perturbations.