Diminishing Simplicity Bias using GAN Generated UnbiasedAugmentations
Bibliographic record
Abstract
Simplicity Bias (SB) characterizes the propensity of neural networks (NN) to excessively favor simpler features, while neglecting potentially informative complex features. This bias hinders a network's capacity to generalize effectively, contributing to reduced robustness outside of distributions (OOD). The challenge arises as more intricate features are prone to being overshadowed by simpler features, leading to feature selection bias and spurious correlation. These spurious correlations can give an inflated view of the model's performance on a test set is good, but the model may be sub-optimal and non-robust because of its over reliance on simple features. In this paper, we investigate the adverse effects of SB on both robustness and model generalization. To address this issue, we introduce a Generative Adversarial Network (GAN) designed to generate unbiased examples by deceiving the original NN exhibiting SB. We demonstrate that the incorporation of these generated unbiased examples encourages learning from complex features previously overlooked by the biased model. Our approach hinges on leveraging models exhibiting SB, and we substantiate its efficacy through controlled datasets showcasing SB. We used four datasets in this paper and show that our approach is effective in mitigating SB across all four datasets.
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 | 0.002 | 0.009 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 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 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".