Paper explained and visualized “Are Pre-trained Convolutions Better than Pre-trained Transformers?” Tune in to the epic fight of CNNs against transformers! Or at least, that’s how the paper is framed.
Because Ms. Coffee Bean also wonders what it takes for a transformer(-like) architecture to be named transformer and when does it become something else, e.g. a CNN. Join the comment section to discuss!
Referenced videos:
📺 Self-attention replaced with the Fourier Transform: https://youtu.be/j7pWPdGEfMA
📺 Ms. Coffee Bean explains the Transformer: https://youtu.be/FWFA4DGuzSc
Discussed paper:
📄 Tay, Y., Dehghani, M., Gupta, J., Bahri, D., Aribandi, V., Qin, Z., & Metzler, D. (2021). Are Pre-trained Convolutions Better than Pre-trained Transformers? https://arxiv.org/abs/2105.03322
Outline:
* 00:00 Are you tired of transformers?
* 01:12 What makes transformers so good?
* 05:13 CNN vs. Transformers
* 09:53 What makes a transformer a transformer? -- Discussion
Music 🎵 : Savior Search - DJ Freedem
🔗 Links:
YouTube: https://www.youtube.com/AICoffeeBreak
Twitter: https://twitter.com/AICoffeeBreak
Reddit: https://www.reddit.com/r/AICoffeeBreak/
#AICoffeeBreak #MsCoffeeBean #MachineLearning #AI #research
00:00 Are you tired of transformers?
01:12 What makes transformers so good?
05:13 CNN vs. Transformers
09:53 What makes a transformer a transformer? -- Discussion
