JoMA: Demystifying Multilayer Transformers via JOint Dynamics of MLP and Attention

arXiv:2310.00535v3 Announce Type: replace
Abstract: We propose Joint MLP/Attention (JoMA) dynamics, a novel mathematical framework to understand the training procedure of multilayer Transformer architectures. This is achieved by integrating out the self-attention layer in Transformers, producing a modified dynamics of MLP layers only. JoMA removes unrealistic assumptions in previous analysis (e.g., lack of residual connection) and predicts that the attention first becomes sparse (to learn salient tokens), then dense (to learn less salient tokens) in the presence of nonlinear activations, …

arxiv attention cs.ai cs.cl cs.lg dynamics mlp transformers type via