Model Interpretation and Explainability towards Creating Transparency in Prediction Models

Dolk, Daniel R.; Kridel, Donald J.; Dineen, Jacob; Castillo, David G.

doi:10.24251/hicss.2020.120

Cited by 2 publications

(4 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…XAI offers a paradigm shift towards interpretability and explainability required in many fields utilizing ML and AI. We have introduced a taxonomy classifying two unique instances of XAI, 'dynamic' and 'static' cases [10], formulated harmony as a measure of the distance between explanations of these cases (Cosine and Jaccard Similarity), and employed a perturbation-based algorithm 1 to systematically quantify it on several models, metrics, and datasets.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Unified Explanations in Machine Learning Models: A Perturbation Approach

Dineen,

Kridel,

Dolk

et al. 2023

Proceedings of the Annual Hawaii International Conference on System Sciences

View full text Add to dashboard Cite

A high-velocity paradigm shift towards Explainable Artificial Intelligence (XAI) has emerged in recent years. Highly complex Machine Learning (ML) models have flourished in many tasks of intelligence, and the questions have started to shift away from traditional metrics of validity towards something deeper: What is this model telling me about my data, and how is it arriving at these conclusions? Inconsistencies between XAI and modeling techniques can have the undesirable effect of casting doubt upon the efficacy of these explainability approaches. To address these problems, we propose a systematic, perturbation-based analysis against a popular, model-agnostic method in XAI, SHapley Additive exPlanations (Shap). We devise algorithms to generate relative feature importance in settings of dynamic inference amongst a suite of popular machine learning and deep learning methods, and metrics that allow us to quantify how well explanations generated under the static case hold. We propose a taxonomy for feature importance methodology, measure alignment, and observe quantifiable similarity amongst explanation models across several datasets.

show abstract

Section: Discussionmentioning

confidence: 99%

“…Previous work [10] explored the taxonomy noted above to analyze how well measures of feature importance hold up under 'what-if' perturbations. We extend this work, not in breadth, but depth, offering a framework to systematically quantify the similarity between the two.…”

Section: Related Workmentioning

confidence: 99%

Unified Explanations in Machine Learning Models: A Perturbation Approach

Dineen,

Kridel,

Dolk

et al. 2023

Proceedings of the Annual Hawaii International Conference on System Sciences

View full text Add to dashboard Cite

show abstract

“…Notably, some contrast exists between explainable and accountable AI, c.f. [5] [12]. As noted in [13], not all AI solutions need to be explainable; for example, 1) trusted and trained algorithms are only useable within specified operating conditions and 2) wellstudied conditions, and/or 3) no significant consequences for unacceptable results.…”

Section: Explainable Aimentioning

confidence: 99%

“…Thus, accountable AI aims to develop trusted AI agents with known bounds of performance (akin to how decisions made by service dogs are unexplainable, but the dogs can be trusted) [5]. When humaninterpretable understanding of black-box decision making is needed [12], XAI attempts to provide explainable interpretations for AI models throughout their operations Prior work in the area of explaining complex algorithms includes neural network rule extraction methods, see [14], which began in the 1980s and 1990s by creating tree-based representations of neural network decision processes. However, as conceptualized in Figure 2, XAI extends beyond rule extraction approaches and includes humancomputer interaction concerns as well as userexplanation concerns.…”

Section: Explainable Aimentioning

confidence: 99%

A Preliminary Look at Heuristic Analysis for Assessing Artificial Intelligence Explainability

Combs

Fendley

Bihl

2020

Wseas Transactions on Computer Research

View full text Add to dashboard Cite

Artificial Intelligence and Machine Learning (AI/ML) models are increasingly criticized for their “black-box” nature. Therefore, eXplainable AI (XAI) approaches to extract human-interpretable decision processes from algorithms have been explored. However, XAI research lacks understanding of algorithmic explainability from a human factors’ perspective. This paper presents a repeatable human factors heuristic analysis for XAI with a demonstration on four decision tree classifier algorithms.

show abstract

Model Interpretation and Explainability towards Creating Transparency in Prediction Models

Cited by 2 publications

References 8 publications

Unified Explanations in Machine Learning Models: A Perturbation Approach

Unified Explanations in Machine Learning Models: A Perturbation Approach

A Preliminary Look at Heuristic Analysis for Assessing Artificial Intelligence Explainability

Contact Info

Product

Resources

About