CORTEXA
← Browse
openalexChemRxiv2026-07-23Cited by 0

Label Scarcity Reveals Architecture-Dependent Robustness in CNN-Based Casting Defect Detection

Tengxiao Liu, Hongtao Sun

Deep learning for automated visual inspection in manufacturing is often limited by the cost of labeling large image datasets. Using 7,348 casting images, we evaluated how three CNN architectures, ResNet18, ResNet50, and MobileNetV3-Small, tolerate progressively reduced labeling budgets, from 100% down to 1% of the training data. Under full supervision, all three architectures achieved mean test accuracies above 99%, appearing nearly interchangeable. This equivalence broke down as labeled data became scarce: ResNet18 and ResNet50 maintained mean accuracies above 99% down to 5% labels, whereas MobileNetV3-Small declined from 99.47% to 83.75% over the same range and exhibited markedly greater run-to-run variability. At the most severe scarcity tested (1% labels), all three architectures degraded substantially, reaching mean accuracies of 89.09%, 86.63%, and 65.71%, respectively. Class-specific error analysis, representative Grad-CAM visualization, and population-level analyses of attention entropy and prediction confidence further revealed increased misclassification, broader spatial attention, and reduced prediction confidence under severe label scarcity. These results demonstrate that robustness to label scarcity is an architecture-dependent property that cannot be inferred from fully supervised accuracy alone, providing a more realistic basis for selecting CNN architectures for data-constrained manufacturing inspection.

View free PDFSource page

Related papers

openalexChemRxiv2026-07-24

Capturing many-body effects for metal ions in aqueous phase through developing specialized machine learning force fields

Madelyn Smith, Pengfei Li

Metal ion–ligand interactions govern reactivity, bioavailability, and catalytic function across chemistry and biology, yet the many-body effects underlying water exchange around multivalent metal ions remain difficult to capture with classical force fields. Ab initio molecular dy…

View free PDFSource page
openalexChemRxiv2026-07-24

Cross-molecule attention unifies diverse mixture property predictions

Xiao Yang, Baojiang He, Lingbo Ji, 霍现宽, Jun Hu, Zhan Qu

The properties of a molecular mixture—the scent of a perfume, the thermodynamic behavior of a solvent blend—emerge from cross-molecule atomic interactions, not from whole-molecule properties alone. Yet dominant machine learning paradigms encode mixture components as independent m…

View free PDFSource page
openalexChemRxiv2026-07-23

Candidate Structure Prioritization in Non-Target Screening with Predicted Collision Cross-Sections and Uncertainty Quantification

Lucas Ferrando Plo, Anneli Kruve

Candidate structure prioritization in non-target screening (NTS) can be performed using collision cross-sections (CCSs) predicted by a machine-learning (ML) model, with a threshold on the relative difference between measured and in silico CCSs. Such thresholds generally are heuri…

View free PDFSource page
openalexChemRxiv2026-07-23

Beyond Multiscale: A Neural Network-Unified Molecular Model for Asymmetry-Free Dynamics Simulation

Takuma Ikeda, Hiroshi Watanabe, Haruyuki Nakano

Adaptive multiscale simulations, such as adaptive QM/MM, are indispensable for investigating complex solution dynamics but historically suffer from a fundamental dilemma: the dynamic exchange of identical solvent molecules across different model resolutions introduces a physical…

View free PDFSource page