Supervised machine learning models boast remarkable predictive capabilities. But can you trust your model? Will it work in deployment? What else can it tell you about the world? We want models to be not only good, but interpretable. And yet the task of interpretation appears underspecified. Papers provide diverse and sometimes non-overlapping motivations for interpretability, and offer myriad n...

This paper presents amethod for an automatic and complete design of fuzzy systems fromdata. Themain objective is to build fuzzy systems with a user-controllable trade-off between accuracy and interpretability. Whereas criteria for accuracy mostly follow straightforwardly from the application, definition of interpretability and its criteria are subject to controversial discussion. For this reaso...

This contribution is concerned with the interpretability of fuzzy rule-based systems. While this property is widely considered to be a crucial one in fuzzy rule-based modeling, a more detailed formal investigation of what "interpretability" actually means is not available. So far, interpretability has most often been associated with rather heuristic assumptions about shape and mutual overlappin...

This paper discusses interpretability in two main categories of fuzzy systems fuzzy rule-based classifiers and interpolative fuzzy systems. Our goal is to show that the aspect of high level interpretability is more relevant to fuzzy classifiers, whereas fuzzy systems employed in modeling and control benefit more from low-level interpretability. We also discuss the interpretabilityaccuracy trade...

We adopt a theory of relativisation based on the idea that relatives, like wh-constructions in the analysis of Chomsky (1998), require two sorts of features to construct their LF-interpretation. We argue that it is the variable interpetability of these features that gives rise to different syntactic patterns. We use this theory to provide an explanation for some curious syntactic facts found in...

This paper reviews recent studies in emerging directions of understanding neural-network representations and learning neural networks with interpretable/disentangled middle-layer representations. Although deep neural networks have exhibited superior performance in various tasks, the interpretability is always an Achilles' heel of deep neural networks. At present, deep neural networks obtain a h...

PA is Peano arithmetic. The formula InterppA(r,/l) is a formalization of the assertion that the theory PA + a interprets the theory PA + , O (the variables a and are intended to range over codes of sentences of PA). We extend Solovay's modal analysis of the formalized provability predicate of PA, PrpA(x), to the case of the formalized interpretability relation Interp,,(x, y). The relevant modal...

Vector representation of words improves performance in various NLP tasks, but the high-dimensional word vectors are very difficult to interpret. We apply several rotation algorithms to the vector representation of words to improve the interpretability. Unlike previous approaches that induce sparsity, the rotated vectors are interpretable while preserving the expressive performance of the origin...

In [8], P. Hájek and V. Švejdar determined normal forms for the system ILF, and showed that we can eliminate the modal operator ¤ from IL–formulas. The normal form for the closed fragment of the interpretability logic IL is an open problem (see [13]). We prove that in some cases we can eliminate the modal operator ¤. We give an example where it is impossible to eliminate ¤. AMS subject classifi...