Infinite attention: NNGP and NTK for deep attention networks

Abstract

There is a growing amount of literature on the relationship between wide neural networks (NNs) and Gaussian processes (GPs), identifying an equivalence between the two for a variety of NN architectures. This equivalence enables, for instance, accurate approximation of the behaviour of wide Bayesian NNs without MCMC or variational approximations, or characterisation of the distribution of randomly initialised wide NNs optimised by gradient descent without ever running an optimiser. We provide a rigorous extension of these results to NNs involving attention layers, showing that unlike single-head attention, which induces non-Gaussian behaviour, multi-head attention architectures behave as GPs as the number of heads tends to infinity. We further discuss the effects of positional encodings and layer normalisation, and propose modifications of the attention mechanism which lead to improved results for both finite and infinitely wide NNs. We evaluate attention kernels empirically, leading to a moderate improvement upon the previous state-of-the-art on CIFAR-10 for GPs without trainable kernels and advanced data preprocessing. Finally, we introduce new features to the Neural Tangents library (Novak et al., 2020) allowing applications of NNGP/NTK models, with and without attention, to variable-length sequences, with an example on the IMDb reviews dataset.

arXiv commentICML 2020

Cited by in corpus (16)

• Tensor Programs II: Neural Tangent Kernel for Any Architecture
• Finite Versus Infinite Neural Networks: an Empirical Study
• Dataset Distillation with Infinitely Wide Convolutional Networks
• Tensor Programs III: Neural Matrix Laws
• Inductive Biases and Variable Creation in Self-Attention Mechanisms
• Tensor Programs IIb: Architectural Universality of Neural Tangent Kernel Training Dynamics
• Towards Deepening Graph Neural Networks: A GNTK-based Optimization Perspective
• Exact posterior distributions of wide Bayesian neural networks
• The Limitations of Large Width in Neural Networks: A Deep Gaussian Process Perspective
• Precise characterization of the prior predictive distribution of deep ReLU networks
• On the Dynamics of Training Attention Models
• Scale Mixtures of Neural Network Gaussian Processes
• Kernel-Based Smoothness Analysis of Residual Networks
• Scaling Neural Tangent Kernels via Sketching and Random Features
• Uncertainty-aware Cardinality Estimation by Neural Network Gaussian Process
• Enhanced Recurrent Neural Tangent Kernels for Non-Time-Series Data