arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2004.14444cs.LGcs.CLstat.ML
收藏

自然分布偏移对问答模型的影响

The Effect of Natural Distribution Shift on Question Answering Models

UC Berkeley(加州大学伯克利分校)

John Miller, Karl Krauth, Benjamin Recht, Ludwig Schmidt

摘要:

我们建立了四个新的测试集斯坦福问答数据集(SQuAD)和评估问答系统的能力,以推广到新的数据。我们的第一个测试集来自原始维基百科域,并测量现有系统过拟合原始测试集的程度。尽管几年来大量的测试集重复使用,我们没有发现自适应过拟合的证据。剩下的三个测试集是从《纽约时报》的文章、Reddit的帖子和亚马逊的产品评论中构建的,并测量了对自然分布偏移的鲁棒性。在广泛的模型中,我们观察到平均性能分别下降3.8,14.0和17.4 F1点。相比之下,强大的人类基线匹配或超过SQuAD模型在原始域上的性能,并且在新域中几乎没有下降。总之,我们的研究结果证实了令人惊讶的弹性的坚持方法,并强调需要走向评估指标,将鲁棒性自然分布偏移。

英文摘要:

We build four new test sets for the Stanford Question Answering Dataset (SQuAD) and evaluate the ability of question-answering systems to generalize to new data. Our first test set is from the original Wikipedia domain and measures the extent to which existing systems overfit the original test set. Despite several years of heavy test set re-use, we find no evidence of adaptive overfitting. The remaining three test sets are constructed from New York Times articles, Reddit posts, and Amazon product reviews and measure robustness to natural distribution shifts. Across a broad range of models, we observe average performance drops of 3.8, 14.0, and 17.4 F1 points, respectively. In contrast, a strong human baseline matches or exceeds the performance of SQuAD models on the original domain and exhibits little to no drop in new domains. Taken together, our results confirm the surprising resilience of the holdout method and emphasize the need to move towards evaluation metrics that incorporate robustness to natural distribution shifts.

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。

↑