28 citations · 45 across the 4 of their papers we have counts for
Showing 2023Show all
2 papers · 1 filter
stat.ML2023★ 3 cited
The Shaped Transformer: Attention Models in the Infinite Depth-and-Width Limit
Lorenzo Noci, Chuning Li, Mufan Bill Li +4
In deep learning theory, the covariance matrix of the representations serves as a proxy to examine the network's trainability. Motivated by the success of Transformers, we study th…
cs.LG2023★ 5 cited
Deep Transformers without Shortcuts: Modifying Self-attention for Faithful Signal Propagation
Bobby He, James Martens, Guodong Zhang +4
Skip connections and normalisation layers form two standard architectural components that are ubiquitous for the training of Deep Neural Networks (DNNs), but whose precise roles ar…