III: Small: Collaborative Research: Towards End-to-End Knowledge Discovery in Complex Brain Networks

III：小：协作研究：复杂大脑网络中的端到端知识发现

• 批准号: 1718310
• 负责人: Xiangnan Kong
• 金额: $ 26.68万
• 依托单位: Worcester Polytechnic Institute
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2017
• 资助国家: 美国
• 起止时间: 2017-08-01 至 2020-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1718310&HistoricalAwards=false
• 关键词: III; Small; Collaborative; Research; Towards

Modeling the brain as a networked system has become a popular paradigm in neuroscience. Knowledge discovery on brain network data has the potential to revolutionize the way we understand the brain and provide a basis for developing new treatments for neurological disorders. However, today knowledge discovery systems on brain networks are still extremely hard to build. Researchers must be taught in tedious details how to design the pipeline of data processing and adjust the parameter settings in different processing stages. To address this problem, the project will aim to build an end-to-end system for knowledge discovery in brain networks that learns to integrate different processing stages and self-adjust with minimal human intervention. Educational activities will include curriculum development and training of students in the areas of data science and neuroscience. Result dissemination is planned via publication in relevant peer-reviewed conferences and journals.To make fundamental contributions to realizing this vision, it is necessary to go beyond the pipeline-based approaches, where the multiple stages of the knowledge discovery are studied in isolation. It has been shown that end-to-end systems can significantly alleviate the problem of laborious efforts in building pipelines and making adjustments. In the planned system, the four stages of the knowledge discovery on brain networks (i.e., node discovery, edge discovery, feature mining and statistical learning) will be integrated and studied together instead of being studied in isolation. This project will investigate innovative ideas on unifying different stages of knowledge discovery in brain networks: (1) collective brain network discovery, where the node discovery and edge discovery stages are deeply integrated into a unified framework; (2) label-contrasting edge discovery, which unifies edge discovery with model inference stages; (3) collective edge discovery unifies the edge discovery and statistical learning stages; (4) node-compressing sub-graph mining, node discovery stage and feature mining stage; (5) an end-to-end system across all four stages, which allows the learners in different stages to work together and form a unified learner for brain network analysis. The project will takes a next logical step in brain network research. It will build intellectual and formal connections between data mining and neuroscience. The project results have the potential for automating brain network analysis. In terms of broader impacts, the project will deliver new analytic tools that are broadly relevant to the study of the brain in cognitive neuroscience. These tools also have broad clinical applicability for early diagnosis of neurological injury, for evaluation of treatment response, etc. Educational activities will include curriculum development and training of students in the areas of data science and neuroscience.

将大脑建模为网络系统已成为神经科学中的流行范式。大脑网络数据的知识发现有可能彻底改变我们理解大脑的方式，并为开发神经系统疾病的新疗法提供基础。然而，今天，在大脑网络上构建知识发现系统仍然非常困难。研究人员必须从繁琐的细节中学习如何设计数据处理的流水线，以及如何调整不同处理阶段的参数设置。为了解决这个问题，该项目将致力于在大脑网络中建立一个端到端的知识发现系统，该系统将学习整合不同的处理阶段，并在最少的人为干预下进行自我调整。教育活动将包括数据科学和神经科学领域的课程开发和学生培训。计划通过在相关同行评审会议和期刊上发表来传播成果，为实现这一愿景作出根本性贡献，有必要超越基于管道的方法，即孤立地研究知识发现的多个阶段。已经表明，端到端系统可以显著减轻在构建管道和进行调整方面的费力工作的问题。在计划系统中，大脑网络上的知识发现的四个阶段（即，节点发现、边缘发现、特征挖掘和统计学习）将被整合在一起，而不是孤立地进行研究。本项目将探讨统一脑网络知识发现不同阶段的创新思路：（1）集体脑网络发现，将节点发现和边发现深度整合到一个统一的框架中;（2）标签对比边发现，将边发现与模型推理阶段统一起来;（3）集体边发现，将边发现与统计学习阶段统一起来;（4）集体边发现，将边发现与统计学习阶段统一起来;（5）集体边发现，将边发现与统计学习阶段统一起来。（4）节点压缩子图挖掘、节点发现阶段和特征挖掘阶段;（5）跨越所有四个阶段的端到端系统，其允许不同阶段的学习者一起工作并形成用于脑网络分析的统一学习者。该项目将在大脑网络研究中迈出合乎逻辑的下一步。它将在数据挖掘和神经科学之间建立知识和正式的联系。该项目的结果具有自动化大脑网络分析的潜力。就更广泛的影响而言，该项目将提供与认知神经科学中的大脑研究广泛相关的新分析工具。这些工具还具有广泛的临床应用性，可用于神经损伤的早期诊断、治疗反应的评估等。教育活动将包括数据科学和神经科学领域的课程开发和学生培训。

项目成果

Learning to Selectively Update State Neurons in Recurrent Networks

• DOI: 10.1145/3340531.3412018
• 发表时间: 2020-10
• 期刊: Proceedings of the 29th ACM International Conference on Information & Knowledge Management
• 作者: Thomas Hartvigsen;Cansu Sen;Elke A. Rundensteiner

Recurrent Networks for Guided Multi-Attention Classification

• DOI: 10.1145/3394486.3403083
• 发表时间: 2020-07
• 期刊: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining
• 作者: Xin Dai;Xiangnan Kong;Tian Guo;J. B. Lee;Xinyue Liu;C. Moore

Deep Parametric Model for Discovering Group-cohesive Functional Brain Regions

• DOI: 10.1137/1.9781611976236.71
• 发表时间: 2020
• 作者: J. B. Lee;Xiangnan Kong;C. Moore;Nesreen Ahmed

Dual-Attention Recurrent Networks for Affine Registration of Neuroimaging Data

• DOI: 10.1137/1.9781611976236.43
• 发表时间: 2020
• 作者: Xin Dai;Xiangnan Kong;Xinyue Liu;J. B. Lee;C. Moore