Multi-level Residual Networks from Dynamical Systems View
Authors:
- Bo ChangBCBo Chang
- Lili MengLMLili Meng
- Eldad HaberEldad Haber
- Frederick TungFrederick Tung
- David BegertDavid Begert
Preprint
Abstract
Deep residual networks (ResNets) and their variants are widely used in many computer vision applications and natural language processing tasks. However, the theoretical principles for designing and training ResNets are still not fully understood. Recently, several points of view have emerged to try to interpret ResNet theoretically, such as unraveled view, unrolled iterative estimation and dynamical systems view. In this paper, we adopt the dynamical systems point of view, and analyze the lesioning properties of ResNet both theoretically and experimentally. Based on these analyses, we additionally propose a novel method for accelerating ResNet training. We apply the proposed method to train ResNets and Wide ResNets for three image classification benchmarks, reducing training time by more than 40% with superior or on-par accuracy.Keywords
All Related Versions
This publication has 13 references indexed in Scilit:
- Reversible Architectures for Arbitrarily Deep Residual Neural NetworksProceedings of the AAAI Conference on Artificial Intelligence, 2018
- Stable architectures for deep neural networksInverse Problems, 2017
- Aggregated Residual Transformations for Deep Neural Networks2017
- A Proposal on Machine Learning via Dynamical SystemsCommunications in Mathematics and Statistics, 2017
- The microsoft 2016 conversational speech recognition system2017
- Dermatologist-level classification of skin cancer with deep neural networksNature, 2017
- Deep Networks with Stochastic Depth2016
- Large-Scale Item Categorization in e-Commerce Using Multiple Recurrent Neural Networks2016
- Deep Residual Learning for Image Recognition2016
- Microsoft COCO: Common Objects in Context2014