Automated Numerical Stability Analysis of Deep Learning Operators

DOI10.48550/arXiv.2607.25494MaRDI QIDQ7284448arXivFDO

Authors Xinye Chen

Publication date 28 July 2026

Published in arXiv

Tldr We introduce a unified software tool, called AutoStabDL, that integrates CESTAC to detect numerical stability of deep learning operators with single computation pass while providing sources of instability and monitoring during training and inference.

Full work available at URL https://doi.org/10.48550/arXiv.2607.25494

arXiv classification

cs.AImath.NA

Cites work

• A new analysis of iterative refinement and its application to accurate solution of ill-conditioned sparse linear systems
• A New Approach to Probabilistic Rounding Error Analysis
• Attention Is All You Need
• BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
• CADNA: a library for estimating round-off error propagation
• Computing $k$-means in mixed precision
• Deep learning
• Discrete stochastic arithmetic for validating results of numerical software
• Gradient-based learning applied to document recognition
• ImageNet classification with deep convolutional neural networks
• Neural Network Precision Tuning
• Numerical Error Analysis of Large Language Models
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Scikit-learn: machine learning in Python

This page was built for publication: Automated Numerical Stability Analysis of Deep Learning Operators

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7284448)