Training Neural Networks Using Features Replay

Abstract

Training a neural network using backpropagation algorithm requires passing error gradients sequentially through the network. The backward locking prevents us from updating network layers in parallel and fully leveraging the computing resources. Recently, there are several works trying to decouple and parallelize the backpropagation algorithm. However, all of them suffer from severe accuracy loss or memory explosion when the neural network is deep. To address these challenging issues, we propose a novel parallel-objective formulation for the objective function of the neural network. After that, we introduce features replay algorithm and prove that it is guaranteed to converge to critical points for the non-convex problem under certain conditions. Finally, we apply our method to training deep convolutional neural networks, and the experimental results show that the proposed method achieves {faster} convergence, {lower} memory consumption, and {better} generalization error than compared methods.

arXiv commentNeurIPS 2018 Spotlight, Training deep learning faster, Convergence guarantee for Pipeline-based methods

References in corpus (9)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Continuous control with deep reinforcement learning
• Playing Atari with Deep Reinforcement Learning
• Revisiting Distributed Synchronous SGD
• Direct Feedback Alignment Provides Learning in Deep Neural Networks
• Training Neural Networks Without Gradients: A Scalable ADMM Approach
• Decoupled Parallel Backpropagation with Convergence Guarantee
• Distributed optimization of deeply nested systems

Cited by in corpus (14)

• Efficient and Robust Parallel DNN Training through Model Parallelism on Multi-GPU Platform
• The Internet of Federated Things (IoFT): A Vision for the Future and In-depth Survey of Data-driven Approaches for Federated Learning
• XPipe: Efficient Pipeline Model Parallelism for Multi-GPU DNN Training
• Fully Decoupled Neural Network Learning Using Delayed Gradients
• LayerPipe: Accelerating Deep Neural Network Training by Intra-Layer and Inter-Layer Gradient Pipelining and Multiprocessor Scheduling
• Pipelined Backpropagation at Scale: Training Large Models without Batches
• Parallel Training of Deep Networks with Local Updates
• Toward Model Parallelism for Deep Neural Network based on Gradient-free ADMM Framework
• Alternating Synthetic and Real Gradients for Neural Language Modeling
• Ouroboros: On Accelerating Training of Transformer-Based Language Models
• Accumulated Decoupled Learning: Mitigating Gradient Staleness in Inter-Layer Model Parallelization
• Towards Quantized Model Parallelism for Graph-Augmented MLPs Based on Gradient-Free ADMM Framework
• Scalable Semi-Supervised SVM via Triply Stochastic Gradients
• Associated Learning: Decomposing End-to-end Backpropagation based on Auto-encoders and Target Propagation