A phase transition between positional and semantic learning in a solvable model of dot-product attention

• URL: http://arxiv.org/abs/2402.03902v2
• Date: Tue, 15 Oct 2024 19:54:06 GMT
• Title: A phase transition between positional and semantic learning in a solvable model of dot-product attention
• Authors: Hugo Cui, Freya Behrens, Florent Krzakala, Lenka Zdeborová,
• Abstract summary: Morelinear model dot-product attention is studied as a non-dimensional self-attention layer with trainable and low-dimensional query and key data. We show that either a positional attention mechanism (with tokens each other based on their respective positions) or a semantic attention mechanism (with tokens tied to each other based their meaning) or a transition from the former to the latter with increasing sample complexity.
• Score: 30.96921029675713
• License:
• Abstract: Many empirical studies have provided evidence for the emergence of algorithmic mechanisms (abilities) in the learning of language models, that lead to qualitative improvements of the model capabilities. Yet, a theoretical characterization of how such mechanisms emerge remains elusive. In this paper, we take a step in this direction by providing a tight theoretical analysis of the emergence of semantic attention in a solvable model of dot-product attention. More precisely, we consider a non-linear self-attention layer with trainable tied and low-rank query and key matrices. In the asymptotic limit of high-dimensional data and a comparably large number of training samples we provide a tight closed-form characterization of the global minimum of the non-convex empirical loss landscape. We show that this minimum corresponds to either a positional attention mechanism (with tokens attending to each other based on their respective positions) or a semantic attention mechanism (with tokens attending to each other based on their meaning), and evidence an emergent phase transition from the former to the latter with increasing sample complexity. Finally, we compare the dot-product attention layer to a linear positional baseline, and show that it outperforms the latter using the semantic mechanism provided it has access to sufficient data.

Related papers

• Learning Discrete Concepts in Latent Hierarchical Models [73.01229236386148]
  Learning concepts from natural high-dimensional data holds potential in building human-aligned and interpretable machine learning models. We formalize concepts as discrete latent causal variables that are related via a hierarchical causal model. We substantiate our theoretical claims with synthetic data experiments.
  arXiv  Detail & Related papers  (2024-06-01T18:01:03Z)
• Connectivity Shapes Implicit Regularization in Matrix Factorization Models for Matrix Completion [2.8948274245812335]
  We investigate the implicit regularization of matrix factorization for solving matrix completion problems. We empirically discover that the connectivity of observed data plays a crucial role in the implicit bias. Our work reveals the intricate interplay between data connectivity, training dynamics, and implicit regularization in matrix factorization models.
  arXiv  Detail & Related papers  (2024-05-22T15:12:14Z)
• Linear Noise Approximation Assisted Bayesian Inference on Mechanistic Model of Partially Observed Stochastic Reaction Network [2.325005809983534]
  This paper develops an efficient Bayesian inference approach for partially observed enzymatic reaction network (SRN) An interpretable linear noise approximation (LNA) metamodel is proposed to approximate the likelihood of observations. An efficient posterior sampling approach is developed by utilizing the gradients of the derived likelihood to speed up the convergence of Markov Chain Monte Carlo.
  arXiv  Detail & Related papers  (2024-05-05T01:54:21Z)
• On the Optimization and Generalization of Multi-head Attention [28.33164313549433]
  We investigate the potential optimization and generalization advantages of using multiple attention heads. We derive convergence and generalization guarantees for gradient-descent training of a single-layer multi-head self-attention model.
  arXiv  Detail & Related papers  (2023-10-19T12:18:24Z)
• Unsupervised and supervised learning of interacting topological phases from single-particle correlation functions [0.0]
  We show that unsupervised and supervised machine learning techniques are able to predict phases of a non-exactly solvable model when trained on data of a solvable model. In particular, we employ a training set made by single-particle correlation functions of a non-interacting quantum wire. We show that both the principal component analysis and the convolutional neural networks trained on the data of the non-interacting model can identify the topological phases of the interacting model with a high degree of accuracy.
  arXiv  Detail & Related papers  (2022-02-18T16:02:29Z)
• Scalable Intervention Target Estimation in Linear Models [52.60799340056917]
  Current approaches to causal structure learning either work with known intervention targets or use hypothesis testing to discover the unknown intervention targets. This paper proposes a scalable and efficient algorithm that consistently identifies all intervention targets. The proposed algorithm can be used to also update a given observational Markov equivalence class into the interventional Markov equivalence class.
  arXiv  Detail & Related papers  (2021-11-15T03:16:56Z)
• Learning Neural Causal Models with Active Interventions [83.44636110899742]
  We introduce an active intervention-targeting mechanism which enables a quick identification of the underlying causal structure of the data-generating process. Our method significantly reduces the required number of interactions compared with random intervention targeting. We demonstrate superior performance on multiple benchmarks from simulated to real-world data.
  arXiv  Detail & Related papers  (2021-09-06T13:10:37Z)
• Discovering Latent Causal Variables via Mechanism Sparsity: A New Principle for Nonlinear ICA [81.4991350761909]
  Independent component analysis (ICA) refers to an ensemble of methods which formalize this goal and provide estimation procedure for practical application. We show that the latent variables can be recovered up to a permutation if one regularizes the latent mechanisms to be sparse.
  arXiv  Detail & Related papers  (2021-07-21T14:22:14Z)
• On the Dynamics of Training Attention Models [30.85940880569692]
  We study the dynamics of training a simple attention-based classification model using gradient descent. We prove that training must converge to attending to the discriminative words when the attention output is classified by a linear classifier.
  arXiv  Detail & Related papers  (2020-11-19T18:55:30Z)
• Gradient Starvation: A Learning Proclivity in Neural Networks [97.02382916372594]
  Gradient Starvation arises when cross-entropy loss is minimized by capturing only a subset of features relevant for the task. This work provides a theoretical explanation for the emergence of such feature imbalance in neural networks.
  arXiv  Detail & Related papers  (2020-11-18T18:52:08Z)
• Multiplicative noise and heavy tails in stochastic optimization [62.993432503309485]
  empirical optimization is central to modern machine learning, but its role in its success is still unclear. We show that it commonly arises in parameters of discrete multiplicative noise due to variance. A detailed analysis is conducted in which we describe on key factors, including recent step size, and data, all exhibit similar results on state-of-the-art neural network models.
  arXiv  Detail & Related papers  (2020-06-11T09:58:01Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.