transfer learning+EEG（一）

问题所在——the applications of BCls are seriously hindered by the time consuming calibrations repeated before each use.
EEG要进行校准（？？）
原因1：脑电维度高又有噪声（边缘分布和条件分布在小样本情况下计算不出） 机器学习还是要看的！
原因2：受试者在实验过程中的脑电信号不稳定（疲劳注意力集中等）
几种迁移学习的方法：
feature-representation-transfer：the knowledge transferred across domains is encoded into a new feature representation
instance-transfer：A important assumption in this case is that certain parts of data in the source domains can be reused to aid the target task. Accordingly, the approaches in this case often require that source and target domains have similar distributions of data.
classifer-transfer：domain adaption of classifer+ ensemble learing of classifers
Domain adaption of classifer is a promising method, which
is developed to overcome the changes in data fom one domain to another.
Ensemble learing of classifers combines multiple classifers
from multiple domains to obtain a fnal predictor.