Causal Interpretability for Machine Learning - Problems, Methods and Evaluation

Abstract

Machine learning models have had discernible achievements in a myriad of applications. However, most of these models are black-boxes, and it is obscure how the decisions are made by them. This makes the models unreliable and untrustworthy. To provide insights into the decision making processes of these models, a variety of traditional interpretable models have been proposed. Moreover, to generate more humanfriendly explanations, recent work on interpretability tries to answer questions related to causality such as "Why does this model makes such decisions?" or "Was it a specific feature that caused the decision made by the model?". In this work, models that aim to answer causal questions are referred to as causal interpretable models. The existing surveys have covered concepts and methodologies of traditional interpretability. In this work, we present a comprehensive survey on causal interpretable models from the aspects of the problems and methods. In addition, this survey provides in-depth insights into the existing evaluation metrics for measuring interpretability, which can help practitioners understand for what scenarios each evaluation metric is suitable.

A Review on Explainability in Multimodal Deep Neural Nets

“…Specifying the minimal desired changes required to flip the decision in favor of the user, mapping well with the human mental model leveraging the class-specific and discriminative features and enhancing the model trust, transparency, VOLUME XX, 2017 accountability, reliability, social acceptance, and usability [190]. Interactive machine learning with a human-centered AI approach paves the way towards multimodal causal learning [32]. Some efforts in this direction are highlighted.…”

“…[31] presents the categorization of different explainability methods applicable to the medical and healthcare sector. In [32], a survey on interpretability methods in machine learning from a causal perspective is presented. Graph neural nets show a key role in explainability from a causal perspective.…”

“…Deep learning-based explanatory systems are broadly classified into justification [17] or introspection-based systems [163], [173]. Humans understand the justificationbased explanations, but it lacks in deriving the causal aspect and interpretation [32]. The introspection-based systems focus on the network's internal behavior but are not well understood by humans.…”

Research Trend of Causal Machine Learning Method: A Literature Review

“…Casual modeling can analyze and find a causal relationship between an event's treatment and outcome [3]. Interpretability is the understanding that humans can recognize the symptoms of strategic analysis, can persistently estimate the modeling approach, part of the rationalization, as the ability of human beings to realize in an understandable way [4]. Meanwhile, machine learning can solve two aspects of a problem with its algorithmic capabilities based on complex problems and adaptability [5].…”

Causal discovery in machine learning: Theories and applications

“…Finally, in the work of Moraffah et al [65], a slightly different perspective of causal discovery is presented. In this paper, the authors discuss the relationship between causal discovery and the new topic of explainability and eXplainable Artificial Intelligence (XAI) [22].…”