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 2302.02801cs.LGcs.CL
收藏

LaMPP:语言模型作为感知和行动的概率先验

LaMPP: Language Models as Probabilistic Priors for Perception and Action

MIT CSAIL(麻省理工学院计算机科学与人工智能实验室)

Belinda Z. Li, William Chen, Pratyusha Sharma, Jacob Andreas

摘要:

在大型文本语料库上训练的语言模型编码了关于真实世界环境和动作序列的丰富分布信息。这些信息在当前的语言处理任务(如问答和指令生成)中起着至关重要的作用。我们描述了如何利用语言模型的 * 非语言 * 感知和控制任务。我们的方法将标签和决策作为概率图模型中的推理,其中语言模型参数化标签,决策和参数的先验分布,使得有可能以原则性的方式整合不确定的观察和不完整的背景知识。应用于语义分割、家庭导航和活动识别任务,该方法改进了对罕见、分布外和结构新颖的输入的预测。

英文摘要:

Language models trained on large text corpora encode rich distributional information about real-world environments and action sequences. This information plays a crucial role in current approaches to language processing tasks like question answering and instruction generation. We describe how to leverage language models for *non-linguistic* perception and control tasks. Our approach casts labeling and decision-making as inference in probabilistic graphical models in which language models parameterize prior distributions over labels, decisions and parameters, making it possible to integrate uncertain observations and incomplete background knowledge in a principled way. Applied to semantic segmentation, household navigation, and activity recognition tasks, this approach improves predictions on rare, out-of-distribution, and structurally novel inputs.

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。

↑