Autonomous Intelligent Machines & Systems

自主智能机器

• 批准号: 2880664
• 金额: --
• 依托单位: University of Oxford
• 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-2880664
• 关键词: Autonomous; Intelligent; Machines; Systems

Research Proposal: Decoding Language from Non-Invasive Brain Activity using Machine LearningMy research focuses on advancing machine learning models to decode language from brain activity, specifically using non-invasive techniques such as functional Magnetic Resonance Imaging (fMRI) and Magnetoencephalography (MEG). Despite recent advancements, current models struggle with poor generalization to unseen linguistic contexts, especially across novel participants and diverse language inputs. This limitation highlights the need for more robust models capable of handling complex and variable brain data. Decoding language from brain signals presents significant challenges, because of the high-dimensional and noisy nature of the data, as well as substantial individual variability in brain responses. Therefore, my research specifically aims to address the following fundamental questions: How and why do brain signals differ between individuals? How can decoding models account for these individual differences while ensuring performance on unseen participants? I am particularly interested in developing methods that learn meaningful representations of participants, enabling the clustering and understanding of similar participants, improving our understanding of differences and the model's ability to generalize across diverse populations. Through this, I hope to contribute to the development of more accurate and adaptive brain decoding models that bridge the gap between brain activity and language.

研究建议：使用机器学习从非侵入性大脑活动中解码语言我的研究重点是推进机器学习模型从大脑活动中解码语言，特别是使用功能性磁共振成像（fMRI）和脑磁图（MEG）等非侵入性技术。尽管最近取得了一些进展，但目前的模型对看不见的语言背景的泛化能力很差，特别是在新的参与者和不同的语言输入中。这种限制凸显了对能够处理复杂且可变的大脑数据的更强大模型的需求。从大脑信号解码语言提出了重大挑战，因为数据的高维和嘈杂的性质，以及大脑反应的巨大个体差异。因此，我的研究特别旨在解决以下基本问题：大脑信号如何以及为什么在个体之间存在差异？解码模型如何在确保对看不见的参与者的性能的同时解释这些个体差异？我特别感兴趣的是开发学习有意义的参与者表示的方法，使相似参与者的聚类和理解，提高我们对差异的理解和模型在不同人群中推广的能力。通过这一点，我希望有助于开发更准确和自适应的大脑解码模型，弥合大脑活动和语言之间的差距。