A Framework for Visual-Mathematical Literacy in Applied Machine Learning: Why Representation Choices Shape What Models Can Learn
Open Access
- 16 June 2026
- journal article
- Published by IOP Publishing in Machine Learning: Engineering
- Vol. 2 (2) , 025004
Abstract
Machine learning (ML) is increasingly applied to physical and engineering systems, yet the translation of physical phenomena into model inputs has received limited systematic scrutiny. The upstream decisions that determine what the model can learn and how its outputs should be interpreted are typically treated as implementation details, but they embed inductive bias in the model. A review of 28 applied ML studies from materials, mechanics and computational materials science (2007–2025) found that all reported basic representation parameters, but only 21% systematically compared alternative representations; the pattern is indicative of a structural rather than merely incidental gap, with the transparency deficit persisting within the most mature and comparison-friendly class (CNN image pipelines). To address this gap, the paper introduces visual-mathematical literacy: a framework for reasoning about and evaluating representation choices, drawing on representation learning, inductive bias analysis, physics-informed ML, and design epistemology. A case study in auxetic metamaterial classification is developed in two phases: a retrospective analysis of six representation decisions across a three-stage pipeline, and a prospective audit executing each of the ten protocol questions end-to-end against a reproduction of the pipeline, showing which decisions the protocol corroborates, refines, and opens. The paper concludes with a protocol of ten audit questions paired with quantitative diagnostics.Keywords
This publication has 50 references indexed in Scilit:
- Layer-wise anomaly detection and classification for powder bed additive manufacturing processes: A machine-agnostic algorithm for real-time pixel-wise semantic segmentationAdditive Manufacturing, 2020
- Finite‐Element‐Based Deep‐Learning Model for Deformation Behavior of Digital MaterialsAdvanced Theory and Simulations, 2020
- A deep learning-based model for defect detection in laser-powder bed fusion using in-situ thermographic monitoringProgress in Additive Manufacturing, 2020
- Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based LocalizationInternational Journal of Computer Vision, 2019
- VOLCO: A predictive model for 3D printed microarchitectureAdditive Manufacturing, 2018
- Context Aware Machine Learning Approaches for Modeling Elastic Localization in Three-Dimensional Composite MicrostructuresIntegrating Materials and Manufacturing Innovation, 2017
- Stochastic microstructure characterization and reconstruction via supervised learningActa Materialia, 2016
- A recurrent neural network for modeling crack growth of aluminium alloyNeural Computing and Applications, 2014
- Multiscale methodology for bone remodelling simulation using coupled finite element and neural network computationBiomechanics and Modeling in Mechanobiology, 2010
- Comparing partitionsJournal of Classification, 1985