Rotation 1: The Computational Mechanism of Visual Working Memory Retrieval

第1轮：视觉工作记忆检索的计算机制

• 批准号: 2884559
• 金额: --
• 依托单位: University of Cambridge
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2884559
• 关键词: Rotation; Computational; Mechanism; Visual; Working

BBSRC strategic theme: Biosciences for an integrated understanding of healthVisual perception refers to the system of processes by which the brain organises and interprets visual information. Under the efficient coding hypothesis, the neural system for binding of visual features (e.g., colour and orientation) should be efficiently configured and optimised according to frequencies of patterns in the environment (i.e., natural statistics). The first research line will centre on systematic explorations of feature binding dynamics induced by manipulated natural statistics (e.g., coupling colours with orientations). Moreover, individual feature spaces will be also investigated in relation to altered natural statistics distributions. For these objectives, extended reality with binocular video passthrough will be used alongside traditional psychophysical methods. The second research line will focus on visual working memory (VWM), a flexible neural system enabling to store and retrieve internal representations of visual information over short time. The binding dynamics will be studied within the cued recall paradigm where participants encode a set of objects with unique feature values (e.g., colours and orientations). After a brief interval, the target item is cued by one of its features (e.g., colour) and another one must be reported (e.g., location). As such, the retrieval process can be broadly divided into cue-to-item matching and report feature readout. These stages will be explored in relation to trial-level dynamics across feature spaces, for example, testing the potential parallel nature of VWM retrieval and retrieval time components in relation to response errors. Importantly, previous computational work drew a parallel between VWM retrieval and visual decision making through the accumulation-to-bound principle where sensory information is continuously integrated towards decision bounds. Using the cued recall task, this project will further investigate the potential similarities building on robust findings in perceptual decision making, such as the effects of time pressure and task difficulty. Additionally, the trial-level dynamics will be examined through changes of mind, representing adaptive real-time decision updating. Follow-up experiments will extend the main findings to altered feature characteristics and dimensionality, targeting binding dynamics from the VWM retrieval perspective. The final experiments will strive to identify dynamics across trials and feature domains, for example, testing potential shifts in cue feature dominance within VWM retrieval by manipulating natural statistics of stimuli features. To extend behavioural analyses, the binding dynamics mechanism and VWM retrieval will be investigated using computational models within the accumulation-to-bound and population coding frameworks. In general, to advance our understanding of the dynamics of feature binding and retrieval in the visual brain, the project will utilise rigorous experimental manipulations in a multi-paradigm approach, focused on both visual perception and VWM. Whereas visual perception is typically framed as the encoding stage only preceding VWM storage, these cognitive and neural faculties might dynamically interact to navigate even through the most basic tasks. This work will initially attempt to map the key dynamics in each faculty to subsequently test their mutual dynamics, synthesised into testable computational models. Crucially, these fundamental principles may provide foundations for developments of diagnostic measures, sufficiently sensitive to detect emerging feature binding deficits, for example, in early stages of Alzheimer's disease.

BBSRC战略主题：综合理解健康的生物科学视觉感知是指大脑组织和解释视觉信息的过程系统。在有效编码假设下，结合视觉特征(例如颜色和方向)的神经系统应该根据环境中模式的频率(即自然统计)被有效地配置和优化。第一条研究路线将集中在系统探索由操纵的自然统计引起的特征结合动力学(例如，将颜色与方向相结合)。此外，还将调查与改变的自然统计分布有关的单个特征空间。为了实现这些目标，将结合传统的心理物理方法使用具有双目视频穿透功能的扩展现实技术。第二个研究线将集中在视觉工作记忆(VWM)上，这是一种灵活的神经系统，能够在短时间内存储和检索视觉信息的内部表征。绑定动态将在线索回忆范式中进行研究，在该范式中，参与者使用独特的特征值(例如，颜色和方向)对一组对象进行编码。在短暂的间隔之后，目标物品由其特征之一(例如，颜色)提示，并且必须报告另一个特征(例如，位置)。因此，检索过程可以大致分为线索对项目匹配和报告特征读出。这些阶段将与跨特征空间的试探级动态有关，例如，测试VWM检索和检索时间分量与响应错误相关的潜在并行性质。重要的是，以前的计算工作通过积累到界限的原理将VWM检索和视觉决策之间进行了比较，其中感觉信息被不断地整合到决策界限。使用线索回忆任务，本项目将进一步调查建立在知觉决策稳健结果基础上的潜在相似性，如时间压力和任务难度的影响。此外，将通过思维变化来检查试验级别的动态，这代表了自适应的实时决策更新。后续实验将把主要发现扩展到改变的特征特征和维度，从VWM检索的角度瞄准结合动力学。最后的实验将努力识别实验和特征领域的动态，例如，通过操纵刺激特征的自然统计数据，测试VWM检索中线索特征主导地位的潜在变化。为了扩展行为分析，将使用累积到结合和种群编码框架内的计算模型来研究结合动力学机制和VWM检索。总体而言，为了促进我们对视觉大脑中特征绑定和提取动力学的理解，该项目将使用严格的实验操作，采用多范式方法，重点放在视觉感知和VWM上。虽然视觉感知通常被认为是编码阶段，只是在VWM存储之前，但这些认知和神经能力可能会动态地相互作用，甚至在最基本的任务中导航。这项工作最初将试图映射每个教员的关键动态，以随后测试他们的相互动态，并合成到可测试的计算模型中。至关重要的是，这些基本原则可能为诊断措施的发展提供基础，这些措施足够灵敏，足以检测新出现的特征结合缺陷，例如，在阿尔茨海默病的早期阶段。