气候变化领域集成服务门户(CCPortal): Transforming task representations to perform novel tasks

CCPortal

DOI	10.1073/PNAS.2008852117
	Transforming task representations to perform novel tasks
	Lampinen A.K.; McClelland J.L.
发表日期	2021
ISSN	00278424
起始页码	32970
结束页码	32981
卷号	117 期号:52
英文摘要	An important aspect of intelligence is the ability to adapt to a novel task without any direct experience (zero shot), based on its relationship to previous tasks. Humans can exhibit this cognitive flexibility. By contrast, models that achieve superhuman performance in specific tasks often fail to adapt to even slight task alterations. To address this, we propose a general computational framework for adapting to novel tasks based on their relationship to prior tasks. We begin by learning vector representations of tasks. To adapt to new tasks, we propose metamappings, higher-order tasks that transform basic task representations. We demonstrate the effectiveness of this framework across a wide variety of tasks and computational paradigms, ranging from regression to image classification and reinforcement learning. We compare to both human adaptability and language-based approaches to zero-shot learning. Across these domains, metamapping is successful, often achieving 80 to 90% performance, without any data, on a novel task, even when the new task directly contradicts prior experience. We further show that metamapping can not only generalize to new tasks via learned relationships, but can also generalize using novel relationships unseen during training. Finally, using metamapping as a starting point can dramatically accelerate later learning on a new task and reduce learning time and cumulative error substantially. Our results provide insight into a possible computational basis of intelligent adaptability and offer a possible framework for modeling cognitive flexibility and building more flexible artificial intelligence systems. © 2020 National Academy of Sciences. All rights reserved.
英文关键词	Artificial intelligence; Cognitive science; Transfer; Zero-shot
语种	英语
scopus关键词	article; artificial intelligence; cognitive flexibility; controlled study; human; human experiment; language; reinforcement learning (machine learning); adaptation; biological model; cognition; learning; vision; Adaptation, Physiological; Artificial Intelligence; Cognition; Humans; Language; Learning; Models, Neurological; Visual Perception
来源期刊	Proceedings of the National Academy of Sciences of the United States of America
文献类型	期刊论文
条目标识符	http://gcip.llas.ac.cn/handle/2XKMVOVA/179648
作者单位	Department of Psychology, Stanford University, Stanford, CA 94305, United States
推荐引用方式 GB/T 7714	Lampinen A.K.,McClelland J.L.. Transforming task representations to perform novel tasks[J],2021,117(52).
APA	Lampinen A.K.,&McClelland J.L..(2021).Transforming task representations to perform novel tasks.Proceedings of the National Academy of Sciences of the United States of America,117(52).
MLA	Lampinen A.K.,et al."Transforming task representations to perform novel tasks".Proceedings of the National Academy of Sciences of the United States of America 117.52(2021).