arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1906.04331cs.CLcs.LG
收藏

平行计划抽样

Parallel Scheduled Sampling

Google Brain(谷歌大脑)

Daniel Duckworth, Arvind Neelakantan, Ben Goodrich, Lukasz Kaiser, Samy Bengio

摘要:

自回归模型广泛应用于序列生成问题。输出序列通常以预定顺序生成,每次一个离散单元(像素或字或字符)。模型通过教师强制进行训练,其中地面实况历史作为输入被馈送到模型,在测试时被模型预测取代。计划抽样的目的是通过随机替换历史中的一些离散单元与模型的预测来减轻训练和测试时间之间的这种差异。虽然教师强制训练与ML加速器配合良好,因为计算可以跨时间并行化,但计划采样涉及不受欢迎的顺序处理。在本文中,我们介绍了一个简单的技术,并行化计划采样跨时间。在实验中,我们发现所提出的技术在图像生成,摘要,对话生成和翻译方面与教师强制培训相比具有等同或更好的性能。在对话响应生成任务中,并行预定采样比教师强制实现了1.6 BLEU分数(11.5%)的提高,而在图像生成中,它分别实现了20%和13.8%的提高,在弗雷歇初始距离(FID)和初始分数(IS)。此外,我们讨论了与预定采样相关的不同超参数对模型性能的影响。

英文摘要:

Auto-regressive models are widely used in sequence generation problems. The output sequence is typically generated in a predetermined order, one discrete unit (pixel or word or character) at a time. The models are trained by teacher-forcing where ground-truth history is fed to the model as input, which at test time is replaced by the model prediction. Scheduled Sampling aims to mitigate this discrepancy between train and test time by randomly replacing some discrete units in the history with the model's prediction. While teacher-forced training works well with ML accelerators as the computation can be parallelized across time, Scheduled Sampling involves undesirable sequential processing. In this paper, we introduce a simple technique to parallelize Scheduled Sampling across time. Experimentally, we find the proposed technique leads to equivalent or better performance on image generation, summarization, dialog generation, and translation compared to teacher-forced training. In dialog response generation task, Parallel Scheduled Sampling achieves 1.6 BLEU score (11.5%) improvement over teacher-forcing while in image generation it achieves 20% and 13.8% improvement in Frechet Inception Distance (FID) and Inception Score (IS) respectively. Further, we discuss the effects of different hyper-parameters associated with Scheduled Sampling on the model performance.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑