英文是纯手打的!论文原文的summarizing and paraphrasing。可能会出现难以避免的拼写错误和语法错误,若有发现欢迎评论指正!文章偏向于笔记,谨慎食用
目录
2.3.1. Multi-Voxel Pattern Analysis and Deep Learning
2.3.2. Region of Interest and Feature Selection
2.3.3. Unsupervised Learning and Prior Knowledge
2.4. Limited Paired Fmri and Image Date
2.4.3. Graph Convolutional Networks
2.5.2. Brain Cognitive Limitation
1. 心得
(1)考古2021的brain decoding
(2)篇幅太短了,图也没有,快速扫过去看个乐了
2. 论文逐段精读
2.1. Abstract
①Challenges of fMRI decoding: a) mapping ability, b) limited data, c) noise
2.2. Introduction
①比较简单地介绍了深度学习在fMRI解码怎么用的,暂时不赘述
2.3. Model Mapping Capability
2.3.1. Multi-Voxel Pattern Analysis and Deep Learning
①Multi - voxel pattern analysis (MVPA) with linear kernel is easily affected by artifact and holds worse explaination
2.3.2. Region of Interest and Feature Selection
①Curse of dimensionality ocurs when limit data meets high dimensional feature and overfitting ocurs when limit data meets low dimensional feature
axiomatic adj.公理;不证自明
2.3.3. Unsupervised Learning and Prior Knowledge
①无监督吗??学习大脑表示?在fMRI解码?
2.4. Limited Paired Fmri and Image Date
2.4.1. Few-Shot Learning
①Few-shot methods: representation-based paradigm, initialization-based paradigm, and illusion-based paradigm
2.4.2. Transfer Learning
①Learning the representation of images in ImageNet, then tuning medical image
2.4.3. Graph Convolutional Networks
①GNN performs good in limited data
2.5. The Effect of Fmri Noise
2.5.1. Hemodynamic Delay
①Consider the hemodynamic delay when decodes
2.5.2. Brain Cognitive Limitation
①EEG decoding may have more data
2.6. Conclusion
~