MétaCan
Menu
Back to cohort
Record W4220918566 · doi:10.18280/isi.270114

Detecting and Mitigating Bias in Data Using Machine Learning with Pre-Training Metrics

2022· article· en· W4220918566 on OpenAlex

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.

venuePublished in a venue whose home country is Canada.
no affNo Canadian affiliation: this work is invisible to an affiliation-only frame.
No Canadian affiliation. An affiliation-only frame, the usual design, would never have seen this work. It is one of the works that make the case for inverting the frame.

Bibliographic record

VenueIngénierie des systèmes d information · 2022
Typearticle
Languageen
FieldComputer Science
TopicImbalanced Data Classification Techniques
Canadian institutionsnot available
Fundersnot available
KeywordsMetric (unit)Divergence (linguistics)Computer sciencePerformance metricSampling biasSample (material)Artificial intelligenceStatisticsPattern recognition (psychology)False positive rateWork (physics)Machine learningSample size determinationMathematicsEngineering

Abstract

fetched live from OpenAlex

In this paper, the proposed algorithm to detect the bias from the datasets and to mitigate the bias in the datasets was observed. The consequences of this work shows that not only bias in a model can be decreased without forfeiting model performance rate, but improving the performance. Class imbalance, KL divergence, sample disparity and Kolmogorov-Smirnov (KS) are the pre-training metrics used in the work. Each metric is given weightage and the features are detected based on the maximum weightage. The model is trained to learn the unbiased data and shows the significant improvement in the performance of the system. ROC curve, False Positive Rate and False Negative Rate is used for bias trade-off. The comparison between FPR and FNR before mitigating bias and after mitigating bias is performed and its results are significantly improved.

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 imitation

Not 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.

metaresearch head score (Codex)0.002
metaresearch head score (Gemma)0.001
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Simulation or modeling · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: none
Teacher disagreement score0.971
Threshold uncertainty score0.572

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0020.001
Meta-epidemiology (narrow)0.0000.000
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0010.002
Science and technology studies0.0010.000
Scholarly communication0.0000.006
Open science0.0010.001
Research integrity0.0000.000
Insufficient payload (model declined to judge)0.0000.000

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.

Opus teacher head0.081
GPT teacher head0.284
Teacher spread0.203 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it