Uncertain but Useful: Leveraging CNN Variability into Data Augmentation

2025-09-05

AI Summary

This work addresses the long-overlooked issue of numerical instability during deep learning training in neuroimaging analysis. Contrary to prevailing approaches that treat model variability induced by iterative stochastic optimization solely as noise to be suppressed, we propose— for the first time—to harness it as a valuable resource. We introduce “variability-driven data augmentation”: leveraging floating-point perturbations and controlled random seed variation, we generate robust model ensembles from FastSurfer CNN; their aggregated outputs serve as augmented features for brain-age regression. Experiments show comparable performance to baselines on the primary task while significantly improving robustness on downstream tasks. Our results demonstrate that training-induced variability can be systematically exploited to enhance generalization and practical utility, thereby broadening the conceptual and methodological boundaries for characterizing and leveraging uncertainty in deep learning.

Technology Category

• Machine Learning: Ensemble Methods
• Reasoning under Uncertainty: Stochastic Optimization
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Deep learning (DL) is rapidly advancing neuroimaging by achieving state-of-the-art performance with reduced computation times. Yet the numerical stability of DL models -- particularly during training -- remains underexplored. While inference with DL is relatively stable, training introduces additional variability primarily through iterative stochastic optimization. We investigate this training-time variability using FastSurfer, a CNN-based whole-brain segmentation pipeline. Controlled perturbations are introduced via floating point perturbations and random seeds. We find that: (i) FastSurfer exhibits higher variability compared to that of a traditional neuroimaging pipeline, suggesting that DL inherits and is particularly susceptible to sources of instability present in its predecessors; (ii) ensembles generated with perturbations achieve performance similar to an unperturbed baseline; and (iii) variability effectively produces ensembles of numerical model families that can be repurposed for downstream applications. As a proof of concept, we demonstrate that numerical ensembles can be used as a data augmentation strategy for brain age regression. These findings position training-time variability not only as a reproducibility concern but also as a resource that can be harnessed to improve robustness and enable new applications in neuroimaging.

Problem

• Investigating training-time variability in deep learning models
• Exploring CNN susceptibility to numerical instability during training
• Leveraging training variability as data augmentation resource

Innovation

• Leveraging CNN training variability via perturbations
• Using floating point and random seed perturbations
• Repurposing numerical ensembles for data augmentation