Unlocking Transformer's Success
Kyunghyun Cho discusses the key ingredients that make Transformers successful, including the attention mechanism for variable-sized input, the residual connection to address the issue of vanishing gradients, and the importance of normalization layers. These concepts have roots in earlier papers by Jan McCune and have greatly influenced the evolution of deep learning.In this clip
From this podcast

The Gradient
Kyunghyun Cho: Neural Machine Translation, Language, and Doing Good Science
Related Questions