This paper presents FactNET, a framework for human-centered granular fact-checking. Each claim to check is decomposed into sub claims that are fact-checked independently by using an LLM’s internal knowledge. The user interacts with the framework through an innovative interface, displaying the complex claim in the form of a graph, allowing the input of human knowledge on the topic, and evaluating the trust given in the generated evidence. A video demonstrating the system is available at TO_BE_PUBLISHED (in submission documents for review phase).

We address the problem of designing interpretable algorithms for image classification. Modern computer vision algorithms implement classification in two phases: feature extraction - the encoding - that relies on deep neural networks (DNN), followed by a task-oriented decision - the decoding - often also using a DNN. We propose to formulate this last phase as an argumentative DialoguE Between two agents relying on visual ATtributEs and Similarity to prototypes (DEBATES). DEBATES represents the combination of information provided by two encoders in a transparent and interpretable way. It relies on a dual process that combines similarity to prototypes and visual attributes, each extracted from an encoder. DEBATES makes explicit the agreements and conflicts between the two encoders managed by the two agents, reveals the causes of unintended behaviors, and helps identify potential corrective actions to improve performance. The approach is demonstrated on two problems of fine-grained image classification.

International audience

In this paper, we explore the min-max inference mechanism of any rule-based system of n if-then possibilistic rules. We establish an additive formula for the output possibility distribution obtained by the inference. From this result, we deduce the corresponding possibility and necessity measures. Moreover, we give necessary and sufficient conditions for the normalization of the output possibility distribution. As application of our results, we tackle the case of a cascade of two if-then possibilistic rules sets and establish an input-output relation between the two min-max equation systems. Finally, we associate to the cascade construction an explicit min-max neural network.

Semantic image annotation is a field of paramount importance in which deep learning excels. However, some application domains, like security or medicine, may need an explanation of this annotation. Explainable Artificial Intelligence is an answer to this need. In this work, an explanation is a sentence in natural language that is dedicated to human users to provide them clues about the process that leads to the decision: the labels assignment to image parts. We focus on semantic image annotation with fuzzy logic that has proven to be a useful framework that captures both image segmentation imprecision and the vagueness of human spatial knowledge and vocabulary. In this paper, we present an algorithm for textual explanation generation of the semantic annotation of image regions.

We introduce a way of reasoning about preferences represented as pairwise comparative statements, based on a very simple yet appealing principle: cancelling out common values across statements. We formalize and streamline this procedure with argument schemes. As a result, any conclusion drawn by means of this approach comes along with a justification. It turns out that the statements which can be inferred through this process form a proper preference relation. More precisely, it corresponds to a necessary preference relation under the assumption of additive utilities. We show the inference task can be performed in polynomial time in this setting, but that finding a minimal length explanation is NP-complete.

We consider decision situations in which a set of points of view (voters, criteria) are to sort a set of candidates to ordered categories (Good/Bad). Candidates are judged good, when approved by a sufficient set of points of view; this corresponds to NonCompensatory Sorting. To be accountable, such approval sorting should provide guarantees about the decision process and decisions concerning specific candidates. We formalize accountability using a feasibility problem expressed as a boolean satisfiability formulation. We illustrate different forms of accountability when a committee decides with approval sorting and study the information that should be disclosed by the committee.

We address the problem of multicriteria ordinalsorting through the lens of accountability, i.e. theability of a human decision-maker to own a recommendationmade by the system. We put forward anumber of model features that would favor the capabilityto support the recommendation with a convincingexplanation. To account for that, we designa recommender system implementing and formalizingsuch features. This system outputs explanationsdefined under the form of specific argumentschemes tailored to represent the specific rules ofthe model. At the end, we discuss possible andpromising argumentative perspectives.

Usually, in argumentation, the proof-standards that are used are fixed a priori by the procedure. However (multicriteria) decision-aiding is a context where it may be modified dynamically during the process, depending on the responses of the decision-maker. The expert indeed needs to adapt and refine its choice of an appropriate method of aggregating arguments pros and cons, so that it fits the preference model inferred from the interaction. In this short paper we introduce how this aspect can be handled in an argumentation-based decision-aiding framework. The first contribution of the paper is conceptual: the notion of a concept lattice based on simple properties and allowing to navigate among the different proof-standards is put forward. We then show how this can be integrated within the Carneades model.