經典論文區
新論文區
Data
1. [Video Recognition] Supervised Learning Using YouTube! New Video Recognition Framework OmniSource!
https://ai-scholar.tech/en/articles/image-recognition/omnisource?fbclid=IwAR3wMPkFLCvTrApjcRxatzLzAbGwatTWzBDN8EumYtQCQvR-xuUWCnTMOow
Training
1. You Can Now EASILY Train 10x Bigger Models On Your GPU Using 'ZERO-Offload' !!
- How to use low precision or mixed precision numbers
- How to trade memory for computation by recalculating from checkpoints
- This method uses the CPU's memory.ZeRO-Offload is based on the third method.
https://ai-scholar.tech/en/articles/deep-learning/zero_offload?fbclid=IwAR2OjQUwLRsZQTImlVGxZgTfO9EH3S93YhF8QpBIS1dG9i8v3eOHST3XHXw
Explainable AI
1. New Grad-CAM With Integrated Gradients
- 通常都用最後一層feature map來獲取有差別的gradient
- 改良:藉由不斷比較全黑圖與原圖的輸入,獲得差分化的熱點圖
https://ai-scholar.tech/en/articles/explainable.ai/Integrated-Gradients?fbclid=IwAR3cHt_RDJeyfEu_N1ASldW0Yp4O7PJN4xjfA1y8-OA3QQZnWhl_qsla59s
Methodology
1. When And How Should CNN Ensembles Be Used?
- new metrics for evaluating one big CNN model or ensemble many CNN models