Cleaning crowdsourced labels using oracles for statistical classification
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Nowadays, crowdsourcing is being widely used to collect training data for solving classification problems. However, crowdsourced labels are often noisy, and there is a performance gap between classification with noisy labels and classification with ground-truth labels. In this paper, we consider how to apply oracle-based label cleaning to reduce the gap. We propose TARS, a label-cleaning advisor that can provide two pieces of valuable advice for data scientists when they need to train or test a model using noisy labels. Firstly, in the model testing stage, given a test dataset with noisy labels, and a classification model, TARS can use the test data to estimate how well the model will perform w.r.t. ground-truth labels. Secondly, in the model training stage, given a training dataset with noisy labels, and a classification algorithm, TARS can determine which label should be sent to an oracle to clean such that the model can be improved the most. For the first advice, we propose an effective estimation technique, and study how to compute confidence intervals to bound its estimation error. For the second advice, we propose a novel cleaning strategy along with two optimization techniques, and illustrate that it is superior to the existing cleaning strategies. We evaluate TARS on both simulated and real-world datasets. The results show that (1) TARS can use noisy test data to accurately estimate a model's true performance for various evaluation metrics; and (2) TARS can improve the model accuracy by a larger margin than the existing cleaning strategies, for the same cleaning budget.
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.
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 | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 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 it