What and How of Machine Learning Transparency: Building Bespoke Explainability Tools with Interoperable Algorithmic Components

Abstract

Explainability techniques for data-driven predictive models based on artificial intelligence and machine learning algorithms allow us to better understand the operation of such systems and help to hold them accountable (Sokol & Flach, 2021a). New transparency approaches are developed at breakneck speed, enabling us to peek inside these black boxes and interpret their decisions. Many of these techniques are introduced as monolithic tools, giving the impression of one-size-fits-all and end-to-end algorithms with limited customisability. Nevertheless, such approaches are often composed of multiple interchangeable modules that need to be tuned to the problem at hand to produce meaningful explanations (Sokol et al., 2019). This paper introduces a collection of hands-on training materials -slides, video recordings and Jupyter Notebooks -that provide guidance through the process of building and evaluating bespoke modular surrogate explainers for tabular data. These resources cover the three core building blocks of this technique: interpretable representation composition, data sampling and explanation generation (Sokol et al., 2019).

What Does Evaluation of Explainable Artificial Intelligence Actually Tell Us? A Case for Compositional and Contextual Validation of XAI Building Blocks

“…Misconstrued Explainer Structure. While the separation between explanation content, presentation format, provision mechanism and the act of explaining is gaining traction in XAI, explainability systems remain predominantly perceived as monolithic entities despite in fact being highly modular [72]. This misconception results in a widespread use of off-the-shelf explainers without much thought or effort put into customising these tools and building bespoke XAI systems on top of them, which practice promises to offer more trustworthy and better quality explanations [76].…”

“…This inspires an alternative, diagnostic conceptualisation of XAI, which focuses on providing users with rigorously tested and well-specified insights into a predictive model instead of attempting to solve the ill-defined "black box" problem [14]. In terms of evaluation, we should thus not only test explainers end-to-end but also validate their individual components independently and provide clear guidelines on how and when to operationalise these tools to guarantee correctness and trustworthiness of their outputs [56,72].…”

Preference-based and local post-hoc explanations for recommender systems

“…Further properties are discussed to create its own LIME explanation algorithm in [59]. Noticeably, [59] discusses one of the key restriction of LIME that consists in knowing by advance the relationship between the interpretable space and original space.…”

“…Further properties are discussed to create its own LIME explanation algorithm in [59]. Noticeably, [59] discusses one of the key restriction of LIME that consists in knowing by advance the relationship between the interpretable space and original space. [59] indicate that, whenever possible, bijective functions should be considered to limit errors when projecting from the interpretable space to the original one.…”

Virtual metrology scheme for predictive plasma uniformity diagnosis of plasma-enhanced atomic layer deposition process using optical emission spectroscopy

“…LIME is an algorithm that can explain predictions of all classifiers or regressors by local approximation using an interpretable model. It can enhance the reliability of prediction modeling by neural network analysis 6,7 .…”