Gradient surgery: A necessity for robust test-time adaptation for detecting casting defects
Bibliographic record
Abstract
Casting defects pose a significant challenge in the manufacturing industry, leading to material waste, production inefficiencies, and compromised product quality. While deep learning models have shown promise in automating defect detection, their effectiveness is often constrained by domain shifts and variability in real-world data distributions. In this work, we propose Bayesian Test-Time Adaptation (BTTA), a novel framework designed to enhance the robustness and adaptability of machine learning models in such dynamic environments. Unlike traditional Test-Time Adaptation (TTA) methods, our approach employs gradient-guided diversification with Stein Variational Gradient Descent (SVGD) to explore diverse optimization paths. Experimental results on benchmark datasets, including CIFAR-10-C , Casting Defects , and GDXray , demonstrate significant performance improvements across key metrics. Notably, the framework achieves an average accuracy improvement of 2-3% under severe corruption levels and excels in cross-domain generalization, highlighting its ability to handle diverse and unseen defect categories. This dynamic adaptability not only addresses the limitations of static models but also offers a practical and cost-effective solution for real-time defect detection in industrial settings. Our study underscores the potential of BTTA to transform quality assurance processes, ensuring reliable performance across varying operational conditions without the need for extensive retraining or large annotated datasets. The codebase for BTTA is available on: https://github.com/afsharshamsi/GradSurgery .
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".