This work proposes a novel method that can potentially provide actionable insight to the user when a neural network makes a less than favorable decision. The paper is interesting in that it provides stable and hence potentially actionable insight that can help the target user change an undesired outcome in the future. The work focuses on asymmetric insight in the sense that insight or suggestions are provided only when the classification is for a certain class. So it is mainly applicable to specific kind of binary classification problems where being classified into one class is more undesirable and requires justification. Some hand wavy arguments are provided in the supplement for extension to multiple classes (one vs all), however it would be good to see experiments on those in practice as it is not at all obvious how the solution extends when you have more than one undesirable class.

The Interval Algebra (IA) and a subset of the Region Connection Calculus (RCC), namely RCC-8, are the dominant Artificial Intelligence approaches for representing and reasoning about qualitative temporal and topological relations respectively. Such qualitative information can be formulated as a Qualitative Constraint Network (QCN). In this paper, we focus on the minimal labeling problem (MLP) and we propose an algorithm to efficiently derive all the feasible base relations of a QCN. Our algorithm considers chordal QCNs and a new form of partial consistency. Further, the proposed algorithm uses tractable subclasses of relations having a specific patchwork property for which closure under weak composition implies the consistency of the input QCN. Experimentations with QCNs of IA and RCC-8 show the importance and efficiency of this new approach.

Spatial and temporal reasoning is a crucial task for certain Artificial Intelligence applications. In this context, and since two decades, various formalisms representing the information through qualitative constraint networks (QCN) have been proposed. Given a QCN, the main two problems that are facing researchers are: deciding whether this QCN is consistent or not, and, the minimal labeling problem. In this paper, we propose an efficient algorithm aiming at solving the minimal labeling problem. This algorithm is based on subclasses of relations for which the property of closure under weak composition implies the minimality of the QCN.