2021/05/09 by Noam Wies, Wies, Noam, Yoav Levine +5 · 4 citations
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Multimodal Machine Learning Applications #Natural Language Processing Techniques #Topic Modeling #cs.CL #cs.LG
paper · pdf · doi:10.48550/arxiv.2105.03928
ICML 2021
openalex publication_date 2021/05/09 · openalex created_date 2021/05/24 · arxiv created 2021/06/09 · arxiv updated 2021/06/10 · openalex updated_date 2026/07/28
After their successful debut in natural language processing, Transformer architectures are now becoming the de-facto standard in many domains. An obstacle for their deployment over new modalities is the architectural configuration: the optimal depth-to-width ratio has been shown to dramatically vary across data types (e.g., 10x larger over images than over language). We theoretically predict the existence of an embedding rank bottleneck that limits the contribution of self-attention width to the Transformer expressivity. We thus directly tie the input vocabulary size and rank to the optimal depth-to-width ratio, since a small vocabulary size or rank dictates an added advantage of depth over width. We empirically demonstrate the existence of this bottleneck and its implications on the depth-to-width interplay of Transformer architectures, linking the architecture variability across domains to the often glossed-over usage of different vocabulary sizes or embedding ranks in different domains. As an additional benefit, our rank bottlenecking framework allows us to identify size redundancies of 25%-50% in leading NLP models such as ALBERT and T5.