On the Interpretability and Explainability of Prototype-Based Methods and Reinforcement Learning
Bibliographic record
Abstract
With the ever-growing use of AI to solve real-world problems, the need for transparency and trust in these methods has given rise to Interpretable and Explainable AI.While a growing body of research is trying to address these issues, some areas of the field can still be further improved.In the field of inherently interpretable AI methods, the embedding spaces used for the current prototype-based neural network methods have issues with prototype-query similarity.In the same area of prototype-based neural networks, another issue is the inadequacies of current schemes for evaluating the interpretability of part-prototype networks.In addition, in the field of reinforcement learning, post-hoc explainability methods are focused more on policy distillation rather than on local feature-based explanations.In this work, an inherently interpretable method is proposed to create more interpretable prototypes in a prototype-based classification scheme.In addition, a robust human-centric evaluation framework is proposed for part-prototype networks.Another method is also proposed to create posthoc explainability in reinforcement learning by utilizing state features.Experiments show the superiority and validity of these methods compared to the previous state of the art.In the case of the proposed evaluation metric, this work also includes a comprehensive comparison of the interpretability of existing part-prototype-based neural network methods.
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 | 0.014 | 0.077 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.005 |
| Scholarly communication | 0.003 | 0.007 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.002 | 0.004 |
| Insufficient payload (model declined to judge) | 0.003 | 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".