Raphaël Fournier-S’niehotta

Victor David, Raphaël Fournier-S'niehotta, Nicolas Travers

We aim at improving reasoning on inconsistent and uncertain data. We focus on knowledge-graph data, extended with time intervals to specify their validity, as regularly found in historical sciences. We propose principles on semantics for efficient Maximum A-Posteriori inference on the new Temporal Markov Logic Networks (TMLN) which extend the Markov Logic Networks (MLN) by uncertain temporal facts and rules. We examine total and partial temporal (in)consistency relations between sets of temporal formulae. Then we propose a new Temporal Parametric Semantics, which may combine several sub-functions, allowing to use different assessment strategies. Finally, we expose the constraints that semantics must respect to satisfy our principles.

Marc Lafon, Yannis Karmim, Julio Silva-Rodríguez et al.

Reliable Uncertainty Quantification (UQ) and failure prediction remain open challenges for Vision-Language Models (VLMs). We introduce ViLU, a new Vision-Language Uncertainty quantification framework that contextualizes uncertainty estimates by leveraging all task-relevant textual representations. ViLU constructs an uncertainty-aware multi-modal representation by integrating the visual embedding, the predicted textual embedding, and an image-conditioned textual representation via cross-attention. Unlike traditional UQ methods based on loss prediction, ViLU trains an uncertainty predictor as a binary classifier to distinguish correct from incorrect predictions using a weighted binary cross-entropy loss, making it loss-agnostic. In particular, our proposed approach is well-suited for post-hoc settings, where only vision and text embeddings are available without direct access to the model itself. Extensive experiments on diverse datasets show the significant gains of our method compared to state-of-the-art failure prediction methods. We apply our method to standard classification datasets, such as ImageNet-1k, as well as large-scale image-caption datasets like CC12M and LAION-400M. Ablation studies highlight the critical role of our architecture and training in achieving effective uncertainty quantification. Our code is publicly available and can be found here: https://github.com/ykrmm/ViLU.

Chen Dang, Hicham Randrianarivo, Raphaël Fournier-S'Niehotta et al.

Web Image Context Extraction (WICE) consists in obtaining the textual information describing an image using the content of the surrounding webpage. A common preprocessing step before performing WICE is to render the content of the webpage. When done at a large scale (e.g., for search engine indexation), it may become very computationally costly (up to several seconds per page). To avoid this cost, we introduce a novel WICE approach that combines Graph Neural Networks (GNNs) and Natural Language Processing models. Our method relies on a graph model containing both node types and text as features. The model is fed through several blocks of GNNs to extract the textual context. Since no labeled WICE dataset with ground truth exists, we train and evaluate the GNNs on a proxy task that consists in finding the semantically closest text to the image caption. We then interpret importance weights to find the most relevant text nodes and define them as the image context. Thanks to GNNs, our model is able to encode both structural and semantic information from the webpage. We show that our approach gives promising results to help address the large-scale WICE problem using only HTML data.

Francesco Foscarin, Nicolas Audebert, Raphaël Fournier-S'Niehotta

We present PKSpell: a data-driven approach for the joint estimation of pitch spelling and key signatures from MIDI files. Both elements are fundamental for the production of a full-fledged musical score and facilitate many MIR tasks such as harmonic analysis, section identification, melodic similarity, and search in a digital music library. We design a deep recurrent neural network model that only requires information readily available in all kinds of MIDI files, including performances, or other symbolic encodings. We release a model trained on the ASAP dataset. Our system can be used with these pre-trained parameters and is easy to integrate into a MIR pipeline. We also propose a data augmentation procedure that helps retraining on small datasets. PKSpell achieves strong key signature estimation performance on a challenging dataset. Most importantly, this model establishes a new state-of-the-art performance on the MuseData pitch spelling dataset without retraining.

Pedro Ramaciotti Morales, Lionel Tabourier, Raphaël Fournier-S'niehotta

The Heterogeneous Information Network (HIN) formalism is very flexible and enables complex recommendations models. We evaluate the effect of different parts of a HIN on the accuracy and the diversity of recommendations, then investigate if these effects are only due to the semantic content encoded in the network. We use recently-proposed diversity measures which are based on the network structure and better suited to the HIN formalism. Finally, we randomly shuffle the edges of some parts of the HIN, to empty the network from its semantic content, while leaving its structure relatively unaffected. We show that the semantic content encoded in the network data has a limited importance for the performance of a recommender system and that structure is crucial.

Pedro Ramaciotti Morales, Robin Lamarche-Perrin, Raphael Fournier-S'niehotta et al.

Diversity is a concept relevant to numerous domains of research varying from ecology, to information theory, and to economics, to cite a few. It is a notion that is steadily gaining attention in the information retrieval, network analysis, and artificial neural networks communities. While the use of diversity measures in network-structured data counts a growing number of applications, no clear and comprehensive description is available for the different ways in which diversities can be measured. In this article, we develop a formal framework for the application of a large family of diversity measures to heterogeneous information networks (HINs), a flexible, widely-used network data formalism. This extends the application of diversity measures, from systems of classifications and apportionments, to more complex relations that can be better modeled by networks. In doing so, we not only provide an effective organization of multiple practices from different domains, but also unearth new observables in systems modeled by heterogeneous information networks. We illustrate the pertinence of our approach by developing different applications related to various domains concerned by both diversity and networks. In particular, we illustrate the usefulness of these new proposed observables in the domains of recommender systems and social media studies, among other fields.