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 2311.05263cs.AIcs.CL
收藏

用于文本生成的基于模型的最小贝叶斯风险解码

Model-Based Minimum Bayes Risk Decoding for Text Generation

CyberAgent(赛博agent公司)

Yuu Jinnai, Tetsuro Morimura, Ukyo Honda, Kaito Ariu, Kenshi Abe

AI总结:

本文提出 MBMBR,用推理时可获得的模型概率替代 MBR 中的 Monte Carlo 概率估计,并在多类文本生成任务及模型上验证其性能优于 MBR。

摘要:

最小贝叶斯风险(Minimum Bayes Risk,MBR)解码已被证明在多种文本生成任务中是 beam search 解码的一种强有力替代方案。MBR 解码根据给定的效用函数,从假设池中选择在概率模型下具有最小期望风险的假设。由于在所有可能假设上精确计算期望风险并不现实,MBR 中通常采用两种近似。第一,它在一组采样得到的假设上进行积分,而不是在所有可能假设上积分。第二,它使用 Monte Carlo 估计器来估计每个假设的概率。虽然第一种近似对于使其在计算上可行是必要的,但第二种并非必不可少,因为在推理时我们通常可以获得模型概率。我们提出基于模型的 MBR(Model-Based MBR,MBMBR),这是 MBR 的一种变体,使用模型概率本身作为概率分布的估计,而不是使用 Monte Carlo 估计。我们通过分析和实验表明,在文本生成任务中,基于模型的估计比 Monte Carlo 估计更有前景。我们的实验表明,无论是使用 encoder-decoder 模型还是大型语言模型,MBMBR 在若干文本生成任务中都优于 MBR。

英文摘要:

Minimum Bayes Risk (MBR) decoding has been shown to be a powerful alternative to beam search decoding in a variety of text generation tasks. MBR decoding selects a hypothesis from a pool of hypotheses that has the least expected risk under a probability model according to a given utility function. Since it is impractical to compute the expected risk exactly over all possible hypotheses, two approximations are commonly used in MBR. First, it integrates over a sampled set of hypotheses rather than over all possible hypotheses. Second, it estimates the probability of each hypothesis using a Monte Carlo estimator. While the first approximation is necessary to make it computationally feasible, the second is not essential since we typically have access to the model probability at inference time. We propose Model-Based MBR (MBMBR), a variant of MBR that uses the model probability itself as the estimate of the probability distribution instead of the Monte Carlo estimate. We show analytically and empirically that the model-based estimate is more promising than the Monte Carlo estimate in text generation tasks. Our experiments show that MBMBR outperforms MBR in several text generation tasks, both with encoder-decoder models and with large language models.

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。

↑