MedMNIST-NAS-Bench: A Tabular Neural Architecture Search Benchmark on MedMNIST v2
Fuente:
Zenodo
Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Recurso digital |
| Lingua: | inglese |
| Pubblicazione: |
Zenodo
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866902260558069760 |
|---|---|
| author | Wang, Wei (William) |
| author_facet | Wang, Wei (William) |
| contents | <p>MedMNIST-NAS-Bench is a tabular neural architecture search (NAS) benchmark covering four MedMNIST v2 medical-imaging datasets (BreastMNIST, PneumoniaMNIST, DermaMNIST, and BloodMNIST) on the native 28×28 image resolution. The benchmark is released across <strong>two search spaces of different cardinality</strong>:</p> <ul> <li><strong>tiny</strong>: a focused space of 768 architectures spanning three macro families (MobileNet, ResNet, and the Hybrid <code>HYB_RES_MB</code>), squeeze-and-excitation attention, and strided-convolution downsampling.</li> <li><strong>small</strong>: an extended space of 5,016 architectures that additionally includes ConvNeXt-style and <code>HYB_RES_NEXT</code> templates, bottleneck channel growth, ECA attention, patch-merge downsampling, and a wider range of width multipliers.</li> </ul> <p>For each (dataset, search space, architecture) combination, training is run from scratch for 100 epochs under three random seeds (777, 888, 999), producing <strong>9,216 fully trained runs in the tiny space</strong> and <strong>60,192 runs in the small space</strong>, for a total of <strong>69,408 runs</strong> across the benchmark.</p> <p>Each run records 60+ per-epoch metrics (accuracy, AUROC, F1, ECE, calibration error, loss, gradient-norm aggregates, learning rate, timing, and resource use) and validation/test logits at the final and best-validation epochs, enabling downstream users to (i) rank architectures by any standard metric, (ii) recompute custom metrics from saved predictions, and (iii) study optimisation dynamics. Data ships in both HDF5 (predictions + labels + metric arrays) and Apache Parquet (long-format tidy table) formats.</p> <p>Loader code, schema documentation, and the exact build script are available at <a href="https://github.com/jiywww/medmnist-nas-bench">https://github.com/jiywww/medmnist-nas-bench</a> under the Apache-2.0 licence. The MedMNIST labels redistributed in the HDF5 files are reused from MedMNIST v2 (Yang et al. 2023) under CC BY 4.0.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_20364514 |
| institution | Zenodo |
| language | eng |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | MedMNIST-NAS-Bench: A Tabular Neural Architecture Search Benchmark on MedMNIST v2 Wang, Wei (William) neural architecture search NAS tabular benchmark MedMNIST medical imaging deep learning benchmark <p>MedMNIST-NAS-Bench is a tabular neural architecture search (NAS) benchmark covering four MedMNIST v2 medical-imaging datasets (BreastMNIST, PneumoniaMNIST, DermaMNIST, and BloodMNIST) on the native 28×28 image resolution. The benchmark is released across <strong>two search spaces of different cardinality</strong>:</p> <ul> <li><strong>tiny</strong>: a focused space of 768 architectures spanning three macro families (MobileNet, ResNet, and the Hybrid <code>HYB_RES_MB</code>), squeeze-and-excitation attention, and strided-convolution downsampling.</li> <li><strong>small</strong>: an extended space of 5,016 architectures that additionally includes ConvNeXt-style and <code>HYB_RES_NEXT</code> templates, bottleneck channel growth, ECA attention, patch-merge downsampling, and a wider range of width multipliers.</li> </ul> <p>For each (dataset, search space, architecture) combination, training is run from scratch for 100 epochs under three random seeds (777, 888, 999), producing <strong>9,216 fully trained runs in the tiny space</strong> and <strong>60,192 runs in the small space</strong>, for a total of <strong>69,408 runs</strong> across the benchmark.</p> <p>Each run records 60+ per-epoch metrics (accuracy, AUROC, F1, ECE, calibration error, loss, gradient-norm aggregates, learning rate, timing, and resource use) and validation/test logits at the final and best-validation epochs, enabling downstream users to (i) rank architectures by any standard metric, (ii) recompute custom metrics from saved predictions, and (iii) study optimisation dynamics. Data ships in both HDF5 (predictions + labels + metric arrays) and Apache Parquet (long-format tidy table) formats.</p> <p>Loader code, schema documentation, and the exact build script are available at <a href="https://github.com/jiywww/medmnist-nas-bench">https://github.com/jiywww/medmnist-nas-bench</a> under the Apache-2.0 licence. The MedMNIST labels redistributed in the HDF5 files are reused from MedMNIST v2 (Yang et al. 2023) under CC BY 4.0.</p> |
| title | MedMNIST-NAS-Bench: A Tabular Neural Architecture Search Benchmark on MedMNIST v2 |
| topic | neural architecture search NAS tabular benchmark MedMNIST medical imaging deep learning benchmark |
| url | https://doi.org/10.5281/zenodo.20364514 |