Randomness In Neural Network Training: Characterizing The Impact of Tooling

2021/06/22 by Donglin Zhuang, Xingyao Zhang, Zhuang, Donglin +5 · 8 citations
Computer Science · #Adversarial Robustness in Machine Learning #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning and Data Classification #Neural and Evolutionary Computing (cs.NE)

paper · pdf · doi:10.48550/arxiv.2106.11872

openalex publication_date 2021/06/22 · openalex created_date 2021/07/05 · openalex updated_date 2026/07/28

Abstract

The quest for determinism in machine learning has disproportionately focused on characterizing the impact of noise introduced by algorithmic design choices. In this work, we address a less well understood and studied question: how does our choice of tooling introduce randomness to deep neural network training. We conduct large scale experiments across different types of hardware, accelerators, state of art networks, and open-source datasets, to characterize how tooling choices contribute to the level of non-determinism in a system, the impact of said non-determinism, and the cost of eliminating different sources of noise. Our findings are surprising, and suggest that the impact of non-determinism in nuanced. While top-line metrics such as top-1 accuracy are not noticeably impacted, model performance on certain parts of the data distribution is far more sensitive to the introduction of randomness. Our results suggest that deterministic tooling is critical for AI safety. However, we also find that the cost of ensuring determinism varies dramatically between neural network architectures and hardware types, e.g., with overhead up to 746%, 241%, and 196% on a spectrum of widely used GPU accelerator architectures, relative to non-deterministic training. The source code used in this paper is available at https://github.com/usyd-fsalab/NeuralNetworkRandomness.

Citations

• Gradient-based learning applied to document recognition
• ImageNet Large Scale Visual Recognition Challenge
• MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
• Random Erasing Data Augmentation
• Reproducibility in Machine Learning for Health
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Learning Multiple Layers of Features from Tiny Images
• Nondeterminism and Instability in Neural Network Optimization

Cited by

• MMA-Sim: Bit-Accurate Reference Model of Tensor Cores and Matrix Cores
• Efficient Heteroscedastic Bayesian Optimization for Risk-Aware AutoRL
• RepDL: Bit-level Reproducible Deep Learning Training and Inference
• Riddled basin geometry sets fundamental limits to predictability and reproducibility in deep learning
• Performance Consistency of Learning Methods for Information Retrieval Tasks
• Model Parallelism With Subnetwork Data Parallelism
• Understanding and Mitigating Numerical Sources of Nondeterminism in LLM Inference
• Leveraging Per-Instance Privacy for Machine Unlearning

Related

• Random Bias Initialization Improves Quantized Training
• On the Power and Limitations of Random Features for Understanding Neural Networks
• Noisy Machines: Understanding Noisy Neural Networks and Enhancing Robustness to Analog Hardware Errors Using Distillation
• Non-Determinism in TensorFlow ResNets
• Neural Redshift: Random Networks are not Random Functions