工作记忆
对比度(视觉)
认知心理学
短时记忆
干涉理论
视觉短时记忆
对象(语法)
视觉记忆
等值
心理学
期限(时间)
长期记忆
内存错误
计算机科学
视觉对象识别的认知神经科学
标志性记忆
识别记忆
认知
人工智能
神经科学
发展心理学
召回
物理
拉什模型
量子力学
作者
Mark Schurgin,Jonathan Flombaum
摘要
Human visual memory is tolerant, meaning that it supports object recognition despite variability across encounters at the image level. Tolerant object recognition remains one capacity in which artificial intelligence trails humans. Typically, tolerance is described as a property of human visual long-term memory (VLTM). In contrast, visual working memory (VWM) is not usually ascribed a role in tolerant recognition, with tests of that system usually demanding discriminatory power-identifying changes, not sameness. There are good reasons to expect that VLTM is more tolerant; functionally, recognition over the long-term must accommodate the fact that objects will not be viewed under identical conditions; and practically, the passive and massive nature of VLTM may impose relatively permissive criteria for thinking that two inputs are the same. But empirically, tolerance has never been compared across working and long-term visual memory. We therefore developed a novel paradigm for equating encoding and test across different memory types. In each experiment trial, participants saw two objects, memory for one tested immediately (VWM) and later for the other (VLTM). VWM performance was better than VLTM and remained robust despite the introduction of image and object variability. In contrast, VLTM performance suffered linearly as more variability was introduced into test stimuli. Additional experiments excluded interference effects as causes for the observed differences. These results suggest the possibility of a previously unidentified role for VWM in the acquisition of tolerant representations for object recognition. (PsycINFO Database Record
科研通智能强力驱动
Strongly Powered by AbleSci AI