A Model-Agnostic Approach to Quantifying the Informativeness of Explanation Methods for Time Series Classification

Nguyen, Thu Trang; Nguyen, Thach Le; Ifrim, Georgiana

doi:10.1007/978-3-030-65742-0_6

Cited by 27 publications

(23 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our experiments (Section 4), we also test two schemes that change the test set while leaving the training set unchanged, namely, Remove and Evaluate (ROAE) and Keep and Evaluate (KAE) [23,21]. Table 1 summarizes the comparison between the automated evaluation schemes.…”

Section: Automated Evaluation Schemes For Xai Methodsmentioning

confidence: 99%

“…test set training set ROAR [12] top KAR [12] bottom ROAE [23,21] top -KAE [23,21] bottom explanations directly. Similarly, Can et al [5] asked crowd workers to rate the saliency maps of Grad-CAM on the visual characteristics of venues.…”

Section: Xai Methodsmentioning

confidence: 99%

“…While various XAI methods have been proposed, their evaluation strategy has not been established well, and there is an urgent demand for quantitative measures to answer the question "Given several XAI methods of a black-box prediction model, which one yields the best interpretations?" Several automated evaluation schemes have been proposed [12,23,21]; they usually delete or replace pixels that are said important by an XAI method, and check the deterioration in the prediction performance. However, as pointed out by a recent research [20], high interpretability for machines does not imply the same for humans.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Crowdsourcing Evaluation of Saliency-based XAI Methods

Lu¹,

Tolmachev²,

Yamamoto³

et al. 2021

Preprint

View full text Add to dashboard Cite

Understanding the reasons behind the predictions made by deep neural networks is critical for gaining human trust in many important applications, which is reflected in the increasing demand for explainability in AI (XAI) in recent years. Saliency-based feature attribution methods, which highlight important parts of images that contribute to decisions by classifiers, are often used as XAI methods, especially in the field of computer vision. In order to compare various saliency-based XAI methods quantitatively, several approaches for automated evaluation schemes have been proposed; however, there is no guarantee that such automated evaluation metrics correctly evaluate explainability, and a high rating by an automated evaluation scheme does not necessarily mean a high explainability for humans. In this study, instead of the automated evaluation, we propose a new human-based evaluation scheme using crowdsourcing to evaluate XAI methods. Our method is inspired by a human computation game, "Peek-a-boom", and can efficiently compare different XAI methods by exploiting the power of crowds. We evaluate the saliency maps of various XAI methods on two datasets with automated and crowd-based evaluation schemes. Our experiments show that the result of our crowd-based evaluation scheme is different from those of automated evaluation schemes. In addition, we regard the crowd-based evaluation results as ground truths and provide a quantitative performance measure to compare different automated evaluation schemes. We also discuss the impact of crowd workers on the results and show that the varying ability of crowd workers does not significantly impact the results.

show abstract

Section: Automated Evaluation Schemes For Xai Methodsmentioning

confidence: 99%

Section: Xai Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Crowdsourcing Evaluation of Saliency-based XAI Methods

Lu¹,

Tolmachev²,

Yamamoto³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In this section, we provide an example of using MrSQM to train and test on a sample dataset. 6 In addition, we show how to obtain the saliency map of a time series for explanation purposes. A more detailed example including the sample dataset can be found in our github repository.…”

Section: Examplementioning

confidence: 99%

“…State-of-the-art deep learning or ensemble architectures are very accurate but often require a tremendous amount of computing resources (e.g., time, memory, space). Moreover, it is not trivial to obtain explanations from such complex models [3,6,7].…”

Section: Introductionmentioning

confidence: 99%

A short tutorial for time series classification and explanation with MrSQM

Nguyen

Ifrim

2022

Software Impacts

Self Cite

View full text Add to dashboard Cite

This paper presents MrSQM, a Python tool for the task of time series classification and explanation. Time series classification is a critical problem not only in scientific research but also in many real-life applications. However, state-of-the-art time series classifiers including deep learning and ensemble architectures are often impractical due to their complexity. MrSQM can provide an alternative lightweight solution, just as accurate but faster, and explainable. The tool is written mainly in C++ but wrapped with Cython to provide a more accessible Python interface.

show abstract