IdentifyMix: An efficient two-stage learning approach to combating label noise
Bibliographic record
Abstract
Deep neural networks require correct label annotation during supervised learning.It is inevitable, however, that some labels are noisy during the labeling process.A deep neural network retains incorrect labels during training, resulting in a degradation of performance.Therefore, it is essential to identify samples with potentially correct labels.In state-of-the-art methods, small-loss samples are chosen for subsequent training through a sample selection strategy.Howerver, it typically ignores the imbalance in noise ratios between mini-batches when performing sample selection within each minibatch.Further, numerous valuable samples with high losses are discarded, which adversely affects the generalization performance of the model, particularly under conditions of high noise ratios.To this end, this paper proposes IdentifyMix, an effective two-stage learning approach for noisy robust learning that combines an unique sample selection strategy and the semi-supervised learning technique.By observing how the dynamics of network training are changing, AUM (Area Under the Margin) provides a criterion that is applied in this research to identify mislabeled data.Moreover, by combining semi-supervised learning with contrastive learning and data augmentation, it is possible to extract more useful information from mislabeled samples.Experiments on several synthetic and real-world noise benchmarks demonstrate the effectiveness of IdentifyMix compared with state-of-the-art methods.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.005 | 0.010 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.004 | 0.005 |
| Research integrity | 0.003 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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 source (direct Gemma or distilled Codex), 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".