Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
Investigating Characteristics, Biases and Evolution of Fact-Checked Claims on the Web
No full textInternational audienceGiven the recent proliferation of fake news online, fact-checking has emerged as a critical defence against misinformation. Several fact-checking organisations are currently employed in the initiative to assess the truthfulness of online claims. Verified claims serve as foundational data for various cross-domain research, including fields of social science and natural language processing, where they are used to study misinformation and several downstream tasks such as automated fact-verification. However, these fact-checking websites inherently harbour biases, posing challenges for academic endeavours aiming to discern truth from misinformation. In this study, we aim to explore the evolving landscape of online claims verified by multiple fact-checking organisations and analyse the underlying biases of individual fact-checking websites. Leveraging ClaimsKG, the largest available corpus of fact-checked claims, we analyse the temporal evolution of claims, focusing on topics, veracity levels, and entities to offer insights into the complex dimensions of online information. We utilise data and dimensions available from ClaimsKG for our analysis and for dimensions such as topics which are not present in ClaimsKG, we create a topic taxonomy and implement a transformer-based model, for multi-label classification of claims. We also observe how similar claims are co-occurant amongst different websites. Our work serves as a standardised framework for categorising claims sourced from diverse fact-checking organisations, laying the foundation for coherent and interpretable fact-checking datasets. The analysis conducted in this work sheds light on the dynamic landscape of online claims verified by several fact-checking organisations and dives into biases and distributions of several fact-checking websites
Robotic Grasping of Unknown Objects Based on Deep Learning-Based Feature Detection
No full textInternational audienceIn recent years, the integration of deep learning into robotic grasping algorithms has led to significant advancements in this field. However, one of the challenges faced by many existing deep learning-based grasping algorithms is their reliance on extensive training data, which makes them less effective when encountering unknown objects not present in the training dataset. This paper presents a simple and effective grasping algorithm that addresses this challenge through the utilization of a deep learning-based object detector, focusing on oriented detection of key features shared among most objects, namely straight edges and corners. By integrating these features with information obtained through image segmentation, the proposed algorithm can logically deduce a grasping pose without being limited by the size of the training dataset. Experimental results on actual robotic grasping of unknown objects over 400 trials show that the proposed method can achieve a higher grasp success rate of 98.25% compared to existing methods
Perception de la formation en chirurgie endoscopique endonasale par les internes d’oto-rhino-laryngologie français : analyse STROBE
No full textInternational audienceObjectivesTo analyze the perception of endoscopic endonasal surgery training by French otolaryngology residents.Material and methodsA multicenter retrospective observational study was conducted from March to April 2023. Otolaryngology residents from 7 French regions filled out a 27-item questionnaire on their training in endoscopic endonasal surgery.ResultsOut of 283 residents contacted, 126 (45%) filled out the questionnaire. Seventy-four (59%) had already partially or completely performed the surgeries specified in their diploma course. The level of mastery of the main steps of endonasal surgery and the level of autonomy were higher in the consolidation stage group than in the basic and advanced stages. Seventy residents (56%) felt they had gaps in their level of training. To improve training, 94 (75%) wished for more dissection sessions, surgical skills assessments each semester and simulation sessions. Eighty-nine (71%) felt they needed to find their own teaching aids and other methods to complete their training. One hundred and thirteen (90%) felt that the lack of funding available for congresses and training courses was detrimental.ConclusionThis study highlighted the overall satisfaction of residents with their training in endoscopic endonasal surgery. They expressed a desire for more dissection, simulation and evaluation
Comparing activation typicality and sparsity in a deep CNN to predict facial beauty
No full textDOI preprint : 10.21203/rs.3.rs-4435236/v1International audienceProcessing fluency, which describes the subjective sensation of ease with which information is processed by the sensory systems and the brain, has become one of the most popular explanations of aesthetic appreciation and beauty. Two metrics have recently been proposed to model fluency: the sparsity of neuronal activation, characterizing the extent to which neurons in the brain are unequally activated by a stimulus, and the statistical typicality of activations, describing how well the encoding of a stimulus matches a reference representation of stimuli of the category to which it belongs. Using Convolutional Neural Networks (CNNs) as a model for the human visual system, this study compares the ability of these metrics to explain variation in facial attractiveness. Our findings show that the sparsity of neuronal activations is a more robust predictor of facial beauty than statistical typicality. Refining the reference representation to a single ethnicity or gender does not increase the explanatory power of statistical typicality. However, statistical typicality and sparsity predict facial beauty based on different layers of the CNNs, suggesting that they describe different neural mechanisms underlying fluency
Organotypic culture of post-mortem adult human brain explants exhibits synaptic plasticity
No full textInternational audienceBackground: Synaptic plasticity is an essential process encoding fine-tuned brain functions, but models to study this process in adult human systems are lacking.Objective/Hypothesis: We aim to test whether ex vivo organotypic culture of post-mortem adult brain explants (OPAB) retain synaptic plasticity.Methods: OPAB were seeded on 3D microelectrode arrays to measure local field potential (LFP). Paired stimulation of distant electrodes was performed over three days to investigate our capacity to modulate specific neuronal connections.Results: Long-term potentiation (LTP) or depression (LTD) did not occur within a single day. In contrast, after two and three days of training, OPABs showed a significant modulation of the paired electrodes’ response compared to the non-paired electrodes from the same array. This response was alleviated upon treatment with dopamine.Conclusion(s): Our work highlights that adult human brain explants retain synaptic plasticity, offering novel approaches to neural circuitry in animal-free models
Producing a Bidirectional ATPG Compliant Verilog-HDL Memory Model of SRAM
No full textInternational audienceMemory components are increasingly used in modern systems such as System-on-Chip (SoC). Moreover, they are becoming more complex as the technology node shrinks, which makes them more prone to manufacturing defects. A large part of memory testing is based on functional tests, using March algorithms to target Functional Fault Models (FFMs). New test methodologies are developed to anticipate the growing complexity of memory components, such as the Cell-Aware (CA) test methodology which has been recently introduced in the field of memory testing. However, in order to apply the CA methodology, which introduces a structural consideration of the circuit to be tested, an accurate digital model of the SRAM has to be designed. This paper proposes a methodology to produce a digital Verilog-HDL netlist of an SRAM, based on an initial analog model (SPICE netlist). The resulting digital model considers the bidirectional nature of the memory, and it is ATPG-compliant, allowing test patterns generation and fault simulation as well
FCAvizIR
No full textImplication is a core notion of Formal Concept Analysis and its extensions. It provides information about the regularities present in the data. When one considers a relational data set of real-size, implications are numerous and their formulation, which combines primitive and relational attributes computed using Relational Concept Analysis framework, is complex. For an expert wishing to answer a question based on such a corpus of implications, having a smart exploration strategy is crucial. FCAvizIR is a web platform which implements a visual approach for such exploration. Comprised of three interactive and coordinated views and a toolbox, FCAvizIR has been designed to explore corpora of implication rules following Schneiderman's famous mantra ``overview first, zoom and filter, then details on demand''. It enables metrics filtering, e.g. fixing a minimum and a maximum support value, and the multiple selection of relations and attributes in the premise and in the conclusion to identify the corresponding subset of implications presented as a list and Euler diagrams
Overview of GeoLifeCLEF 2024: Species Composition Prediction with High Spatial Resolution at Continental Scale using Remote Sensing
No full textSource Agritrop Cirad (https://agritrop.cirad.fr/613023/) * Autres projets (id;sigle;titre): 101060639;MAMBO;(EU) Modern Approaches to the Monitoring of BiOdiversity// 101060693;GUARDEN;(EU) safeGUARDing biodivErsity aNd critical ecosystem services across sectors and scales//International audienceUnderstanding the spatiotemporal distribution of species is a cornerstone of ecology and conservation. Pairing species observations with geographic and environmental predictors allows us to model the relationship between an environment and the species present at a given location. In light of that, we organize an annual competition, GeoLifeCLEF, which focuses on benchmarking and advancing state-of-the-art species distribution modeling using available bioclimatic and remote sensing data. The GeoLifeCLEF 2024 dataset spans across Europe and encompasses most of its flora. The species observation data comprises over 5 million Presence-Only (PO) occurrences and approximately 90 thousand Presence-Absence (PA) surveys. Those data are paired with various high-resolution rasters, including remote sensing imagery, land cover, and elevation, and are combined with coarse-resolution data such as climate, soil, and human footprint variables. In this paper, we present (i) an overview of the GeoLifeCLEF 2024 competition, (ii) a description of the provided data, (iii) an overview of approaches used by the participating teams, and (iv) the main results analysis
Exact antichain saturation numbers via a generalisation of a result of Lehman-Ron
No full textInternational audienceFor given positive integers k and n, a family F of subsets of {1, . . . , n} is kantichain saturated if it does not contain an antichain of size k, but adding any set to F creates an antichain of size k. We use sat * (n, k) to denote the smallest size of such a family. For all k and sufficiently large n, we determine the exact value of sat * (n, k). Our result implies that sat * (n, k) = n(k -1) -Θ(k log k), which confirms several conjectures on antichain saturation. Previously, exact values for sat * (n, k) were only known for k up to 6.We also prove a strengthening of a result of Lehman-Ron which may be of independent interest. We show that given m disjoint chains C 1 , . . . , C m in the Boolean lattice, we can create m disjoint skipless chains that cover the elements from ∪ m i=1 C i (where we call a chain skipless if any two consecutive elements differ in size by exactly one)
