Minimizing Model Misclassification Using Regularized Loss Interpretability
Bibliographic record
Abstract
In the field of machine learning, the increasing use of complex deep learning models like Deep Neural Networks is leading to a decrease in the transparency of decision-making processes. This presents significant challenges, especially in sensitive applications where specific misclassifications can have serious consequences. The complexity of these models makes it challenging to target specific enhancements, such as rectifying misclassifications within a subgroup of the dataset or improving the classification rate of a particular label group. Achieving these improvements often necessitates extensive testing with the hope of attaining the desired results. This paper presents an innovative approach to proactively reduce misclassifications between specific pairs of labels in multiclass scenarios with already trained models and with minimal decrease in overall performance. The focus on already trained models is to reduce testing and retraining costs, enabling teams to efficiently and affordably audit their models and enhance robustness and fairness before deploying them into production. By modifying loss functions to assign penalties to undesirable misclassifications, our approach directs the optimization process to prioritize the reduction of specific critical errors. Our experimental results demonstrate the effectiveness of this strategy in reducing targeted misclassifications while maintaining or improving overall model performance metrics such as accuracy, precision, and F1 score. This study contributes to the field of Explainable Artificial Intelligence (XAI) by offering a cheap, fast and practical tool for embedding fairness and interpretability directly into the model training process, paving the way for more transparent and accountable AI systems.
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 | 0.017 | 0.060 |
| Meta-epidemiology (narrow) | 0.003 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.004 | 0.004 |
| Open science | 0.003 | 0.005 |
| Research integrity | 0.003 | 0.005 |
| 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".