GradInit: Learning to Initialize Neural Networks for Stable and Efficient Training
This addresses a foundational issue in machine learning by providing a portable initialization scheme that improves training stability and efficiency across diverse architectures, though it is incremental as it builds on existing initialization heuristics.
The paper tackles the problem of training instability and hyperparameter sensitivity in novel neural architectures by introducing GradInit, an automated and architecture-agnostic initialization method that adjusts layer norms to minimize loss after a single optimization step, resulting in accelerated convergence and improved test performance for convolutional networks and enhanced stability for Transformers without requiring learning rate warmup.
Innovations in neural architectures have fostered significant breakthroughs in language modeling and computer vision. Unfortunately, novel architectures often result in challenging hyper-parameter choices and training instability if the network parameters are not properly initialized. A number of architecture-specific initialization schemes have been proposed, but these schemes are not always portable to new architectures. This paper presents GradInit, an automated and architecture agnostic method for initializing neural networks. GradInit is based on a simple heuristic; the norm of each network layer is adjusted so that a single step of SGD or Adam with prescribed hyperparameters results in the smallest possible loss value. This adjustment is done by introducing a scalar multiplier variable in front of each parameter block, and then optimizing these variables using a simple numerical scheme. GradInit accelerates the convergence and test performance of many convolutional architectures, both with or without skip connections, and even without normalization layers. It also improves the stability of the original Transformer architecture for machine translation, enabling training it without learning rate warmup using either Adam or SGD under a wide range of learning rates and momentum coefficients. Code is available at https://github.com/zhuchen03/gradinit.