Improving Explainable AI Interpretability: Mathematical Models for Evaluating Explanation Methods.
Bibliographic record
Abstract
<title>Abstract</title> AI has transformed various industries. Understanding and trusting AI decision-making processes is crucial as they become more integrated into our lives. Explainable AI (XAI) aims to provide transparency and interpretability to AI models, addressing concerns about accountability, fairness, and ethical AI. Lack of transparency in AI can lead to uncertainty, especially in critical domains where incorrect or biased decisions can have adverse outcomes. This paper aims to introduce Explainable Artificial Intelligence (XAI) and its significance in enhancing transparency, accountability, fairness, and trustworthiness in AI systems. The primary focus is on presenting mathematical expressions for quantitatively evaluating the accuracy and relevance of explanations offered by XAI methods, thereby enhancing the quality and dependability of these explanations. The paper conducts a literature review on XAI methods and their applications, specifically examining whether evaluation metrics are provided for assessing the explanations. The paper presents a mathematical formulation for an Intrusion Detection System (IDS) that utilizes autoencoders along with an explanation technique like SHAP, as a case study. We further present the application of the proposed evaluation metrics and mathematical formulas for quantitative assessment of the correctness of the explanations. Screenshots of the results have been presented for each of the quantitative mathematical formulas of each metric. The contributions to the mathematical derivation of the IDS case study is also profound wherein we adopt the cross-entropy loss function for derivation and mathematically provide solutions to address the overfitting problem with L1regularization and also express the threshold updation using Chebyshev’s formula. The results presented in the results and discussion section include the correctness evaluation of the mathematical formulations of the evaluation metrics for XAI, which is demonstrated using a case study (Autoencoder-based Intrusion Detection System with SHAPley explanations) demonstrating their applicability and transparency. The significance of XAI in promoting comprehension and confidence in AI systems is underscored by this paper. Through transparency and interpretability, XAI effectively tackles apprehensions related to accountability, fairness, and ethical AI. The mathematical assessment metrics put forth in this study provide a means to evaluate the accuracy and pertinence of explanations furnished by XAI techniques, thereby facilitating advancements and comparisons in AI research and development. The future generalized implementation of these metrics with real-time data across various domains will enhance the practicality and usefulness of XAI across diverse domains. This study was conducted on open-access data obtained from Canadian Institute for Cybersecurity and NSL KDD dataset.
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 distilled prediction
Teacher imitationNot 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.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.031 | 0.015 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.002 | 0.001 |
| Open science | 0.003 | 0.007 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.000 | 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; both teacher heads agree on what is shown here.
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".