Github Code:https://github.com/moskomule/ewc.pytorch/tree/master
Github Code2:https://github.com/kuc2477/pytorch-ewc
國外筆記:https://www.khoury.northeastern.edu/home/hand/teaching/cs7150-summer-2020/Continual_Learning_and_Catastrophic_Forgetting.pdf
Medium with code:https://pub.towardsai.net/overcoming-catastrophic-forgetting-a-simple-guide-to-elastic-weight-consolidation-122d7ac54328
另一篇數學推導論文:https://arxiv.org/pdf/2105.04093.pdf
https://blog.csdn.net/qq_43428929/article/details/125868016
yzu1103309 changed a year agoView mode Like Bookmark
題目說明
VoxCeleb1 資料集
利用 Transformer Encoder 做語者辨識
69438 labeled data for training
切十分之一給 Validation
6000 unlabeled data for testing
可分成 600 個類別(600 位語者)
成果
Screenshot_20240207_103458
yzu1103309 changed a year agoView mode Like Bookmark
基本概念
Screenshot_20240119_083253
專門用在圖像辨識上
每個神經元不需要看整張圖片(No need to be Fully connected)
Screenshot_20240119_083736
所以將圖片拆分成一個一個 Receptive field
每個 Receptive field 連接多個 Neurons
yzu1103309 changed a year agoView mode Like Bookmark