Bibliographic record
Abstract
Transfer learning involves leveraging knowledge gained from solving one task and then using that knowledge to improve performance and reduce subsequent training time on a different but related task. Despite its advantages, recent attention has been directed towards a critical concern relating to the fairness of models trained with transfer learning. A previous study has demonstrated that transfer learning can preserve biases (that are intentionally planted) from the source task, transferring them to the target task. In this paper, we question a different but equally critical problem: whether transfer learning can introduce new biases or lead to greater biases in the target task compared to models trained from scratch. Our investigation reveals that transfer learning has the potential to introduce varying degrees of bias in the target task that were not originally present in the source task. Specifically, in an Alzheimer’s Disease classification task, we show that the use of transfer learning introduces greater bias with respect to sex and age, compared to an equivalent non-transfer learning approach and a simpler model, both trained from scratch and almost as accurate. These findings underscore the need for a comprehensive understanding of the inherent limitations and risks associated with the application of transfer learning, particularly in high-risk applications, e.g. healthcare. This result also suggests the need for further research into how and when transfer learning introduces and amplifies bias.
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 | 0.007 | 0.025 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.003 |
| Scholarly communication | 0.003 | 0.005 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.002 | 0.004 |
| Insufficient payload (model declined to judge) | 0.007 | 0.003 |
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".