认知地图欧式表征与拓扑表征的关系
文献类型：学位论文
| 作者 | 吴文雅 |
| 答辩日期 | 2024-06 |
| 文献子类 | 硕士 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院心理研究所 |
| 其他责任者 | 王亮 |
| 关键词 | 认知地图 欧式表征 拓扑表征 层级性 后继表征 |
| 学位名称 | 理学硕士 |
| 学位专业 | 认知神经科学 |
| 其他题名 | The relationship between Euclidean and topological representation of cognitive maps |
| 中文摘要 | Spatial navigation is an important cognitive function for humans and other animals living in complex environments. Cognitive maps are internal representations of spatial information, which help animals realize flexible navigation under complex and changeable environments, and to complete advanced cognitive behaviors such as route planning and shortcut selection. The representational formats of cognitive maps are still in dispute with two main theoretical hypotheses, namely Euclidean map and topological graph. On one hand, Euclidean map hypothesis assumes that cognitive maps adopt absolute and globally consistent Euclidean metrics, which are supported by the discovery of hippocampal place cells and entorhinal grid cells. On the other hand, topological graph hypothesis holds that cognitive maps represent the links or relationships between objects in the memory space which is not limited to physical space, but also includes abstract space (such as semantic concept space and social relationship space, etc.), while the hippocampus and the entorhinal cortex are both involved in the formation of such memory space. However, though supported by some experimental evidence, both Euclidean map and topological graph have certain limitations. Actually, they are at the two extremes in terms of accuracy for representing metric information, whereas cognitive maps in real environments tend to combine Euclidean and topological attributes together. Therefore, some researchers tried to integrate these two types of information, that is, to add Euclidean information to the topological graph so as to form hybrid cognitive maps, such as the reference frame network theory and the labeled graph hypothesis. In addition, some other research also proposed that the cognitive and computational mechanisms underlying Euclidean map and topological graph might be the same, such as the successor representation model. All the hypotheses above are reasonable to some extent. Based on this, Study 1 and 2 proposed the following hypotheses: Euclidean and topological representations are products of the same cognitive and computational mechanisms under different task backgrounds. Specifically, both of them adopt the way of successor representation, with Euclidean representations as the result of successor representation model defining and updating state distribution in the spatial scope, while topological representation as the result of defining and updating state sequence in the temporal course. Both of them follow the evolution from local to global (from egocentric to allocentric) in the construction of cognitive maps. Therefore, abilities of these two representations might be consistent within the same people, that is, there could be a positive correlation between them. Accordingly, behavioral tasks of Euclidean map construction and topological graph construction were designed respectively in study 1, and participants were asked to construct two types of cognitive maps from zero to one. The task performance showed that there was indeed a phenomenon of local expansion to the global in the process of cognitive map construction, and the boundary influence of local areas gradually declined. In study 2, the general navigation ability of participants was measured by SBSOD scale, and the relationship between Euclidean and topological representation ability was analyzed, as well as their contribution to general navigation ability. The results showed that Euclidean and topological representation ability were associated by a significant positive correlation and contributed to general navigation ability, indirectly indicating that there might be similar underlying mechanisms between them. We further simulated the construction of cognitive maps in Euclidean and topological tasks by using the successor representation model, focusing on the discount factor γ, a parameter representing the degree of local generalization, in order to find the best γ values in both tasks, and found that both were between 0.6 ~ 0.7. In addition, the behavioral modeling of successor representation was conducted according to the task performance of participants to find the best fitting parameter γ in each of three sessions. The results showed that participants tended to adopt a higher γ (higher than the optimal value simulated by the model) in the early stage of cognitive map construction, that is, they adopted an over-generalization strategy. However, with the gradual expansion of cognitive maps from the local to the global, the γ from the behavioral optimal fitting gradually decreased to approach the task-specific optimal γ simulated above, indicating that the strategies adopted under the two steps of representations during construction of cognitive maps were to gradually optimize the local generalization degree to achieve the optimal memory performance, supporting the hypothesis we proposed above. Study 3 further changed the experimental task to a simpler version on the basis of study 1 and 2, and explored the temporal dynamics of encoding and integration of Euclidean and topological information via brain activities recording during the construction of cognitive maps through EEG technology. Participants needed to learn Euclidean and topological information respectively, and complete three types of test tasks, namely estimation of Euclidean distance, topological distance and path distance between two randomly given objects. The first two tasks only needed the corresponding Euclidean or Topological information, while the path distance estimation required both kinds of information. According to representational similarity analysis, participants encoded and integrated two types of information at the same time during each type of learning task in the learning stage, while whether they tended to extract task-specific information in the test stage was unknown. To sum up, based on previous theoretical assumptions of representation formats of cognitive maps, this study proposed the commonalities of Euclidean and topological representations in the cognitive map construction process and verified them through behavioral experiments and computational modeling methods. Besides, the temporal dynamics of encoding and integration of these two representations were reflected by EEG activity, which would be helpful for researchers in the field of cognitive maps to broaden their research ideas and deepen their understanding of the relationship between Euclidean and topological representations. |
| 英文摘要 | 空间导航对于人和其他生活在复杂环境中的动物而言是一项重要的认知功 能。认知地图是对空间信息的内部表征，有助于动物在复杂多变的环境下实现灵 活导航，完成如路线规划、捷径选择等高级认知行为。认知地图的表征形式目前 仍存在争议，两个主流的理论假设分别是欧式地图和拓扑图。一方面，欧式地图 假设认为认知地图采取绝对的、全局一致的欧式度量结构，海马位置细胞和内嗅 网格细胞的发现都为此假设提供证据支持。另一方面，拓扑图结构假设则认为认 知地图表征的是记忆空间中物体之间的连接关系，并且记忆空间不仅限于物理空 间，还包括广义的抽象空间（如语义概念空间和社会关系空间等），而海马和内 嗅参与此类记忆空间的形成。 然而，欧式地图和拓扑图两种表征方式虽然都有实验证据支持，但也都存在 一定的局限，二者处于度量性信息表征精度的两个极端，真实环境下的认知地图 往往同时具备欧式和拓扑这两种属性。因此，后续有研究者试图将二者融合起来， 即将欧式信息添加到拓扑结构上形成混合型的认知地图，如参考框架网络理论和 有标签图假设；也有研究指出欧式地图和拓扑图的底层认知和计算机制可能是一 致的，如后继表征模型。 上述这些假设都有一定的合理性，基于此，研究一和研究二提出如下假设： 欧式表征和拓扑表征是不同任务背景下相同认知和计算机制的产物，即二者均采 取后继表征的方式，不过欧式表征是后继表征模型在空间范围内定义和更新状态 分布的结果，而拓扑表征则是在时间进程上定义和更新状态序列的结果，二者在 认知地图的构建过程中均遵循从局部扩展到全局（从自我中心到环境中心）的演 变。由此，这两种表征的能力可能在被试个体内部具有一致性，即存在正相关关 系。因此，研究一分别设计了欧式地图和拓扑图构建的行为任务，让被试从 0 到 1 构建两种表征，任务表现显示认知地图构建过程中确实存在从局部扩展到全局 的现象，局部区域的边界影响逐渐减小。研究二通过 SBSOD 问卷测量被试的一 般性导航能力，并分析欧式表征、拓扑表征能力之间的关系以及二者对一般性导 航能力的贡献，结果显示欧式表征和拓扑表征的能力存在显著的正相关，且能预 测一般性导航能力，间接表明二者可能存在相似的底层机制。研究二进一步运用 后继表征模型模拟欧式任务和拓扑任务中的认知地图构建过程，关注折扣因子这 一代表局部泛化程度的参数，寻找两种任务下的最佳γ值，发现二者均介于 0.6-0.7 之间；此外还根据被试在三个 session 中的任务表现进行后继表征的行为 建模，寻找每个 session 中最佳的γ拟合参数值，结果显示被试在认知地图构建 初期倾向于采取较高的γ（高于上述模型模拟出的最优值），即采取过度泛化的 策略，但随着认知地图逐步从局部扩展至全局，最优拟合被试行为的γ逐渐下降 以趋近上述模拟出的任务特定的最优γ值，表明在认知地图构建过程中两种表征下采取的策略均为逐渐优化局部的泛化程度，以达到最优的记忆表现，支持了前 述假设。 研究三在研究一、二的基础上进一步将实验任务改为更简单的版本，通过脑 电技术探究认知地图构建过程中脑活动对欧式和拓扑信息的编码及其整合的时 间动态性。被试需要分别学习欧式和拓扑信息，并完成三类测试任务，分别是欧 式距离、拓扑距离和路径距离估计，其中前两种任务仅需要相应的欧式或拓扑信 息即可，而路径距离估计则同时需要两种信息。通过表征相似性分析发现，被试 在学习阶段每类学习任务期间会同时编码两类信息并将其整合起来，但没有验证 是否在测试阶段更倾向于提取任务特定的信息。 综上，本研究在梳理和总结以往认知地图表征方式的理论假设基础上，提出 了欧式表征和拓扑表征在认知地图构建过程中存在的共性并通过行为实验和建 模进行验证。此外，还通过脑电活动反映两种表征在编码和整合阶段的时间动态 性，有助于认知地图领域研究者拓宽研究思路、加深对欧式表征和拓扑表征关系 的理解。 |
| 语种 | 中文 |
| 源URL | [http://ir.psych.ac.cn/handle/311026/47966]  |
| 专题 | 心理研究所_健康与遗传心理学研究室 |
| 推荐引用方式 GB/T 7714 | 吴文雅. 认知地图欧式表征与拓扑表征的关系[D]. 中国科学院心理研究所. 中国科学院大学. 2024. |
入库方式： OAI收割
来源：心理研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。