Neural architectures and their connectivity

神经架构及其连接性

• 批准号: 2744000
• 金额: --
• 依托单位: University of Warwick
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2022
• 资助国家: 英国
• 起止时间: 2022 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2744000
• 关键词: Neural; architectures; their; connectivity

One focus is on effective connectivity algorithms and how they can be applied to understanding human brain function. An aim is to measure how effective connectivity changes with different task demands to study information flow in the brain. To investigate neural architectures, one route will utilise genetic algorithms to create biologically plausible architectures for defined computational tasks. A related route will investigate how the network design relates to the invariances in the input, using for example memory trace learning rules or geometric approaches to deep learning. Understanding the relationship between environmental statistics and neural network design may lead real world applications in the fields of unsupervised learning and neural architecture search. Finally, knowledge of both the connectivity and architecture of the brain is not only important for computational neuroscience, but may lead to applications in medical science.

其中一个焦点是有效的连接算法以及如何将它们应用于理解人脑功能。目的是测量有效连接如何随不同任务要求而变化，以研究大脑中的信息流。为了研究神经架构，一种途径是利用遗传算法为定义的计算任务创建生物学上合理的架构。相关路线将研究网络设计如何与输入的不变性相关，例如使用记忆跟踪学习规则或深度学习的几何方法。了解环境统计和神经网络设计之间的关系可能会引导无监督学习和神经架构搜索领域的实际应用。最后，关于大脑连接和结构的知识不仅对计算神经科学很重要，而且可能会导致在医学科学中的应用。