Turn a published ML replication into a bounded next check
Which specific claim in an existing label-smoothing or noisy-label replication can be mapped to accessible code, data and an affordable evaluation?
Make this plan your own ↓Read, edit and export without an account. This is preparation; no run or result is claimed.
Make the question your own.
5 sources · Exact preparation context
- [Re] When Does Label Smoothing Help? →
Wagner, Sven Dominik; Kurowski, Yann Philipp; Holländer, Leonhard Moritz; Uelwer, Tobias. [Re] When Does Label Smoothing Help?. ReScience C 10(1), #1; 10.5281/zenodo.14953654.
Record revision: 7d572b4aeced63c459f3327eaca77679891a46ac7446f970e3556601edf98b2b
- Learning with Noisy Labels [~Re]visited →
Hudovernik, Valter; Rot, Žiga; Vovk, Klemen; Škodnik, Luka; Zajc, Luka Čehovin. Learning with Noisy Labels [~Re]visited. ReScience C 11(1), #2; 10.5281/zenodo.18401497.
Record revision: dab275d6c82daad20f1c75a15462fef45d2bea00baaf33a84600e13c79eb744c
- [Re] Numerical influence of ReLU'(0) on backpropagation →
Martorella, Tommaso; Contreras, Héctor Manuel Ramirez; García, Daniel Cerezo. [Re] Numerical influence of ReLU'(0) on backpropagation. ReScience C 9(2), #17; 10.5281/zenodo.8173692.
Record revision: d16ef4b304b9e8aa47a9080a4ba63091c0c82562a63f74da6b40922d6a41914a
- 1.16. Probability calibration →
scikit-learn documentation. 1.16. Probability calibration. https://scikit-learn.org/stable/modules/calibration.html; observed version 1.9.1 (observed documentation version; stable URL is mutable).
Record revision: 050be5bd8cb0531752bdbb0497b88170e005285a7bf26884b3f238fb913c6149
- 3.4. Metrics and scoring: quantifying the quality of predictions →
scikit-learn documentation. 3.4. Metrics and scoring: quantifying the quality of predictions. https://scikit-learn.org/stable/modules/model_evaluation.html; observed version 1.9.1 (observed documentation version; stable URL is mutable).
Record revision: 14c631ef986b838e43bdf04a5e3fd3a3be03cb58771e82dd8b05cc39b4aa6282
Source links open the current listing. This plan retains the revisions above; listing review does not grant permission to copy or adapt source files.
Before carrying out this plan
- Full-text claim review, exact artifact revisions and file-specific rights remain necessary.
- GPU/dependency/model/data requirements are outside this pilot; qualify and quote before execution.
Limits to keep in view
- No replication was performed by importing these references.
- Do not rebrand an upstream article as a new Executable Science publication.
Edit freely. Export a copy before reloading or leaving. Your text stays in this tab until you choose to copy, download or save it privately.
What to compare
Read the full selected paper before defining the claim. A README, abstract, original author's score, or catalog entry cannot substitute for an independently specified evaluation plan.
Useful outputs
- A claim-to-source map preserving original authors, venue and version DOI.
- Separate code, data, weights and manuscript license findings.
- One scoped protocol with environment, input digests, metric, tolerances and a maximum execution quote.
Inputs and prerequisites
- Full-text claim review, exact artifact revisions and file-specific rights remain necessary.
- GPU/dependency/model/data requirements are outside this pilot; qualify and quote before execution.
What this work would not establish
- No replication was performed by importing these references.
- Do not rebrand an upstream article as a new Executable Science publication.
Start with these sources.
[Re] When Does Label Smoothing Help?
ReScience C
A replication concerning label smoothing in neural-network training.
Investigate How does smoothing strength alter accuracy, log loss and calibration on a fixed validation protocol?
Learning with Noisy Labels [~Re]visited
ReScience C
A replication report on learning from labels that contain errors, with source metadata identifying a Python deep-learning implementation.
Investigate How does a prespecified label-corruption schedule affect held-out accuracy and calibration across fixed seeds?
[Re] Numerical influence of ReLU'(0) on backpropagation
ReScience C
A replication about the numerical influence of the derivative convention for ReLU at zero during backpropagation.
Investigate On a controlled network with exact-zero activations, how do alternative zero-derivative conventions change gradients and subsequent updates?
1.16. Probability calibration
scikit-learn documentation
A probability-calibration guide covering reliability curves and sigmoid, isotonic and temperature-scaling approaches.
Investigate How do calibration methods trade off log loss and reliability when the calibration sample is small?
3.4. Metrics and scoring: quantifying the quality of predictions
scikit-learn documentation
A broad scoring reference that distinguishes classification, regression, ranking and clustering metrics and their scorer interfaces.
Investigate Can a scoring implementation reproduce independently computed confusion-matrix and probability-loss examples, including undefined edge cases?