Continuous strength, population-based representations in visual working memory

视觉工作记忆允许人们在短时间内记住信息，以便他们可以使用这些信息来执行任务（例如，当你在杂乱的橱柜中寻找它时，记住哪个杯子是你的）。这个记忆系统非常有限。例如，即使在很短的时间尺度内，我们也很难保留关于几个对象的准确信息。目前的建议使用新颖的计算模型和实验技术来更好地理解视觉工作记忆中存储的关于项目的信息。重要的是，一个人能够在脑海中保存的信息量（即工作记忆）已被证明与许多其他认知能力（包括智力）有关，工作记忆的中断在临床疾病中很常见，如注意力缺陷多动障碍和精神分裂症。因此，关于如何存储和使用这些信息的新知识对于理解个体之间重要差异的本质至关重要。提高我们对视觉工作记忆系统的理解也可能提高我们对视觉技能执行的理解，比如那些涉及驾驶和其他需要同时记忆和监控多个位置和物体的复杂任务（例如，跟踪其他车道上汽车的位置，以决定合并是否安全）。这项工作还可以通过更深入地了解哪些视觉元素对用户来说最容易（或最难）记住或导航，从而为显示和用户界面的设计提供信息。除了主要研究之外，该提案还包括一个重要的扩展组成部分，包括从代表性不足的群体中招募和培训第一代大学生，以及开发软件，使科学家能够更好地测量记忆表现，并通过允许招募更广泛，更具代表性的个人样本（即通过基于互联网的实验）的方法来做到这一点。这个研究项目旨在了解我们的工作记忆系统在脑海中保留了关于一个物体的哪些信息。现有的视觉工作记忆理论通常假设对一个特征的记忆是由一个单一的值来表示的（例如，对于颜色记忆，“我认为沙发是红色的”）。相比之下，这个提议使用连续再现任务的计算模型，参与者必须再现一组物品的确切颜色或形状，以测试连续强度人口代码的特征。这个项目提出的另一种记忆概念假设，记忆不是由一个单一的值来表示的，而是由一个特征的每个可能值的激活值来表示的（例如，对沙发颜色的记忆不仅仅是“红色”，而是每种颜色的一系列值）。这项提案中的研究测试的是人们是否能够获得特征的概率分布，而不仅仅是哪一个特征最能反映他们的记忆。它还测试了项目之间的可变性（精度和置信度）是否可以从这些表示中预测出来。这里提出的新理论框架——植根于信号检测理论——不仅允许理解像记忆错误这样的客观测量，而且还对记忆强度和与记忆强度相关的主观自信感做出预测。该框架也与工作记忆的神经种群编码模型有密切的相似之处。因此，这项工作在工作记忆容量、记忆精度和信心之间建立了新的联系，并在工作记忆的神经和认知模型之间架起了一座桥梁。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Visual working memory allows people to hold information in mind for a brief period of time so that they may use that information to perform a task (e.g., holding in mind which mug is yours as you look for it in a cluttered cabinet). This memory system is very limited. For example, we struggle to retain accurate information about just a few objects even over short timescales. The current proposal uses novel computational modeling and experimental techniques to better understand what information is stored about items in visual working memory. Importantly, the amount of information a person is able to hold in mind (i.e., in working memory) has been shown to be related to many other cognitive abilities (including intelligence), and disruptions of working memory are common in clinical disorders like attention deficit hyperactivity disorder and schizophrenia. New knowledge regarding how this information is stored and used is therefore critical for understanding the nature of important differences between individuals. Improving our understanding of the visual working memory system may also improve how we understand the execution of visual skills such as those involved in driving and other complex tasks which require simultaneous remembering and monitoring of multiple locations and objects (e.g., tracking the positions of cars in other lanes to decide whether it is safe to merge). This work may also inform the design of displays and user-interfaces by providing deeper knowledge of which visual elements are easiest (or hardest) for users to remember or navigate between. Beyond the primary research, this proposal also includes a significant outreach component, featuring both the recruitment and training of first-generation college students from underrepresented groups, as well as the development of software to allow scientists to better measure memory performance and to do so via methods that allow for recruitment of a wider, more representative sample of individuals (i.e., via internet-based experiments).This research project seeks to understand what information our working memory system holds in mind about an object. Existing theories of visual working memory commonly assume that a memory for a feature is represented by just a single value (e.g., for color memory, “I think the couch was ‘red’”). By contrast, this proposal uses computational models of continuous reproduction tasks, whereby participants must reproduce the exact color or shape of a set of items in order to test for signatures of continuous strength population-codes. The alternative conception of memory advanced by this project assumes that memory is not represented by a single value, but by an activation value for each possible value of a feature (e.g., memory for the color of the couch is not just ‘red’ but a range of values for each color). The research in this proposal tests whether people have access to a probability distribution over features, rather than simply what single feature best reflects their memory. It also tests whether the variability between items (in precision and confidence) is predictable from such representations. The new theoretical framework advanced here – which is rooted in signal detection theory – not only allows for an understanding of objective measures like memory error, but also makes predictions about memory strength and the subjective feeling of confidence associated with memory strength. The framework also shares close parallels with neural population coding models of working memory. Thus, this work makes novel connections between working memory capacity, memory precision and confidence, and provides a bridge between neural and cognitive models of working memory.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.