Timezone: »

 
Poster
Only Train Once: A One-Shot Neural Network Training And Pruning Framework
Tianyi Chen · Bo Ji · Tianyu Ding · Biyi Fang · Guanyi Wang · Zhihui Zhu · Luming Liang · Yixin Shi · Sheng Yi · Xiao Tu

Wed Dec 08 12:30 AM -- 02:00 AM (PST) @
in Poster Session 3 »

Structured pruning is a commonly used technique in deploying deep neural networks (DNNs) onto resource-constrained devices. However, the existing pruning methods are usually heuristic, task-specified, and require an extra fine-tuning procedure. To overcome these limitations, we propose a framework that compresses DNNs into slimmer architectures with competitive performances and significant FLOPs reductions by Only-Train-Once (OTO). OTO contains two key steps: (i) we partition the parameters of DNNs into zero-invariant groups, enabling us to prune zero groups without affecting the output; and (ii) to promote zero groups, we then formulate a structured-sparsity optimization problem, and propose a novel optimization algorithm, Half-Space Stochastic Projected Gradient (HSPG), to solve it, which outperforms the standard proximal methods on group sparsity exploration, and maintains comparable convergence. To demonstrate the effectiveness of OTO, we train and compress full models simultaneously from scratch without fine-tuning for inference speedup and parameter reduction, and achieve state-of-the-art results on VGG16 for CIFAR10, ResNet50 for CIFAR10 and Bert for SQuAD and competitive result on ResNet50 for ImageNet. The source code is available at https://github.com/tianyic/onlytrainonce.

Keywords: Deep Learning · Optimization · Reinforcement Learning and Planning

Author Information

Tianyi Chen (Microsoft)
Bo Ji (National University of Singapore)
Tianyu Ding (Johns Hopkins University)
Biyi Fang (Microsoft)
Guanyi Wang (Georgia Institute of Technology)
Zhihui Zhu (University of Denver)
Luming Liang (Microsoft)
Yixin Shi (Microsoft)
Sheng Yi (North Carolina State University)
Xiao Tu (Microsoft)

More from the Same Authors