Year: 2018
Type: preprint
Abstract: Batch Normalization (BatchNorm) is a widely adopted technique that enables faster and more stable training of deep neural networks (DNNs). Despite its pervasiveness, the exact reasons for BatchNorm's effectiveness are still poorly understood. The popular belief is that this effectiveness stems from controlling the change of the layers' input distributions during training to reduce the so-called "internal covariate shift". In this work, we demonstrate that such distributional stability of layer inputs has little to do with the success of BatchNorm. Instead, we uncover a more fundamental impact of BatchNorm on the training process: it makes the optimization landscape significantly smoother. This smoothness induces a more predictive and stable behavior of the gradients, allowing for faster training.
Source: arXiv (Cornell University)
Institution: Massachusetts Institute of Technology
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Awards: US-Israel Collaboration: Collaborative Research: New Tools for Extracting Neuronal Phenotypes from a Volumetric Set of Cerebral Cortex Images, CAREER: Fast Graph Algorithms and Continuous Optimization, SaTC: CORE: Small: Towards Adversarially Robust Machine Learning, BIGDATA: F: Testing High Dimensional Distributions without the Curse of Dimensionality, TWC: TTP Option: Frontier: Collaborative: MACS: A Modular Approach to Cloud Security, SHF: Medium: Collaborative Research: Run-Time Support for Scalable Concurrent Programming, AF: SMALL: Frontiers in Algorithmic Game Theory, BIGDATA: IA: DKA: Collaborative Research: High-Thoughput Connectomics
OpenAlex ID: W6982273903
