2023/08/31 by Davoud Ataee Tarzanagh, Yingcong Li, Tarzanagh, Davoud Ataee +5 · 5 voices · 14 citations
Computer Science · Mathematics · #Algorithm #Applied mathematics #Artificial intelligence #Artificial neural network #Computer science #Deep learning #Discrete mathematics #Domain Adaptation and Few-Shot Learning #Equivalence (formal languages) #Gradient descent #Machine Learning and Data Classification #Mathematical optimization #Mathematics #Nonlinear system #Pairwise comparison #Parameterized complexity #Softmax function #Support vector machine #Topic Modeling #Transformer #Voltage
paper · pdf · doi:10.48550/arxiv.2308.16898
published in arXiv (Cornell University) (Cornell University)
openalex publication_date 2023/08/31 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/01
Since its inception in "Attention Is All You Need", transformer architecture has led to revolutionary advancements in NLP. The attention layer within the transformer admits a sequence of input tokens X and makes them interact through pairwise similarities computed as softmax(XQK^\top X^\top), where (K,Q) are the trainable key-query parameters. In this work, we establish a formal equivalence between the optimization geometry of self-attention and a hard-margin SVM problem that separates optimal input tokens from non-optimal tokens using linear constraints on the outer-products of token pairs. This formalism allows us to characterize the implicit bias of 1-layer transformers optimized with gradient descent: (1) Optimizing the attention layer with vanishing regularization, parameterized by (K,Q), converges in direction to an SVM solution minimizing the nuclear norm of the combined parameter W=KQ^\top. Instead, directly parameterizing by W minimizes a Frobenius norm objective. We characterize this convergence, highlighting that it can occur toward locally-optimal directions rather than global ones. (2) Complementing this, we prove the local/global directional convergence of gradient descent under suitable geometric conditions. Importantly, we show that over-parameterization catalyzes global convergence by ensuring the feasibility of the SVM problem and by guaranteeing a benign optimization landscape devoid of stationary points. (3) While our theory applies primarily to linear prediction heads, we propose a more general SVM equivalence that predicts the implicit bias with nonlinear heads. Our findings are applicable to arbitrary datasets and their validity is verified via experiments. We also introduce several open problems and research directions. We believe these findings inspire the interpretation of transformers as a hierarchy of SVMs that separates and selects optimal tokens.