Budgeted Training: Rethinking Deep Neural Network Training Under Resource Constraints

Abstract

In most practical settings and theoretical analyses, one assumes that a model can be trained until convergence. However, the growing complexity of machine learning datasets and models may violate such assumptions. Indeed, current approaches for hyper-parameter tuning and neural architecture search tend to be limited by practical resource constraints. Therefore, we introduce a formal setting for studying training under the non-asymptotic, resource-constrained regime, i.e., budgeted training. We analyze the following problem: "given a dataset, algorithm, and fixed resource budget, what is the best achievable performance?" We focus on the number of optimization iterations as the representative resource. Under such a setting, we show that it is critical to adjust the learning rate schedule according to the given budget. Among budget-aware learning schedules, we find simple linear decay to be both robust and high-performing. We support our claim through extensive experiments with state-of-the-art models on ImageNet (image classification), Kinetics (video classification), MS COCO (object detection and instance segmentation), and Cityscapes (semantic segmentation). We also analyze our results and find that the key to a good schedule is budgeted convergence, a phenomenon whereby the gradient vanishes at the end of each allowed budget. We also revisit existing approaches for fast convergence and show that budget-aware learning schedules readily outperform such approaches under (the practical but under-explored) budgeted training setting.

arXiv commentICLR 2020. Project page with code is at http://www.cs.cmu.edu/~mengtial/proj/budgetnn/

References in corpus (13)

• Neural Architecture Search with Reinforcement Learning
• End to End Learning for Self-Driving Cars
• Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
• SGDR: Stochastic Gradient Descent with Warm Restarts
• On the Convergence of Adam and Beyond
• The Cityscapes Dataset for Semantic Urban Scene Understanding
• Efficient Neural Architecture Search via Parameter Sharing
• The Open Images Dataset V4: Unified image classification, object detection, and visual relationship detection at scale
• Efficient Architecture Search by Network Transformation
• Systematic evaluation of CNN advances on the ImageNet
• Essentially No Barriers in Neural Network Energy Landscape
• A Closer Look at Deep Learning Heuristics: Learning rate restarts, Warmup and Distillation
• Learnable Embedding Space for Efficient Neural Architecture Compression

Cited by in corpus (9)

• Train Large, Then Compress: Rethinking Model Size for Efficient Training and Inference of Transformers
• Multiplicative noise and heavy tails in stochastic optimization
• Normalized Attention Without Probability Cage
• Federated Action Recognition on Heterogeneous Embedded Devices
• REX: Revisiting Budgeted Training with an Improved Schedule
• Neighborhood-Aware Neural Architecture Search
• Towards Streaming Perception
• Visual Domain Adaptation for Monocular Depth Estimation on Resource-Constrained Hardware
• Gradient-based Hyperparameter Optimization Over Long Horizons