Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
Using Saliency for Semantic Image Abstractions in Robotic Painting
No full textInternational audienceWe present an adaptive, semantics‐based abstraction approach that balances aesthetic quality and structural coherence within the practical constraints of robotic painting. We apply panoptic segmentation with color‐based over‐segmentation to partition images into meaningful regions aligned with semantic objects, while providing flexible abstraction levels. Automatic parameter selection for region merging is enabled by semantic saliency maps, derived from Out‐of‐Distribution segmentation techniques in combination with machine learning methods for feature detection. This preserves the boundaries of salient objects while simplifying less prominent regions. A graph‐based community detection step further refines the abstraction by grouping regions according to local connectivity and semantic coherence. The runtime of our method outperforms optimization‐based image vectorization methods, enabling the efficient generation of multiple abstraction levels that can serve as hierarchical layers for robotic painting. We demonstrate the quality of our method by showing abstraction results, robotic paintings with the e‐David robot, and a comparison to other abstraction methods
Chronic Ulcers Healing Prediction through Machine Learning Approaches: Preliminary Results on Diabetic Foot Ulcers Case Study
No full textInternational audienceBackground: Chronic diabetic foot ulcers are a global health challenge, affecting approximately 18.6 million individuals each year. The timely and accurate prediction of wound healing paths is crucial for improving treatment outcomes and reducing complications. Methods: In this study, we apply predictive modeling to the case study of diabetic foot ulcers, analyzing and comparing multiple models based on Deep Neural Networks (DNNs) and Machine Learning (ML) algorithms to enhance wound prognosis and clinical decision making. Our approach leverages a dataset of 1766 diabetic foot wounds, each monitored for at least three visits, incorporating key clinical wound features such as WBP scores, wound area, depth, and tissue status. Results: Among the 12 models evaluated, the highest accuracy (80%) was achieved using a three-layer LSTM recurrent DNN trained on wound instances with four visits. The model performance was assessed through AUC (0.85), recall (0.80), precision (0.79), and F1-score (0.80). Our findings indicate that the wound depth and area at the first visit followed by the wound area and granulated tissue percentage at the second visit are the most influential factors in predicting the wound status. Conclusions: As future developments, we started building a weakly supervised semantic segmentation model that classifies wound tissues into necrosis, slough, and granulation, using tissue color proportions to further improve model performance. This research underscores the potential of predictive modeling in chronic wound management, specifically in the case of diabetic foot ulcers, offering a tool that can be seamlessly integrated into routine clinical practice
Semantic Enrichment of the Quantum Cascade Laser Properties in Text-A Knowledge Graph Generation Approach
No full textCo-located with the Extended Semantic Web Conference (ESWC 2025)International audienceA well structured collection of the various Quantum Cascade Laser (QCL) design and working properties data provides a platform to analyze and understand the relationships between these properties. This analysis can result in insights into how different design features impact laser performance properties. Most of these QCL properties are captured in scientific text. This poses challenges in generating structured QCL properties data for exploration properties due to the specific nature of this domain. There is therefore a need for efficient methodologies that can be utilized to extract QCL properties from text and generate a semantically enriched and interlinked platform where the properties can be analyzed. There is also the need to maintain provenance and reference information on which these properties are based. Semantic Web technologies such as Ontologies and Knowledge Graphs (KGs) have proven capability in providing interlinked data platforms for knowledge representation in various domains. In this paper, we propose an approach for generating a QCL properties Knowledge Graph (KG) from text. The approach is based on the QCL ontology and a Retrieval Augmented Generation (RAG) enabled information extraction pipeline based on GPT 4-Turbo language model. The properties of interest include: working temperature, laser design type, lasing frequency, laser optical power and the heterostructure. The experimental results demonstrate the feasibility and effectiveness of this approach for efficiently extracting QCL properties from unstructured text and generating a QCL properties Knowledge Graph, which has potential applications in semantic enrichment and analysis of QCL data
Verified Path Indexing
No full textInternational audienceThe indexing of syntactic terms is a key component for the efficient implementation of automated theorem provers. This paper presents the first verified implementation of a term indexing data structure, namely a formalization of path indexing in the proof assistant Isabelle/HOL. We define the data structure, maintenance operations, and retrieval operations, including retrieval of unifiable terms, instances, generalizations and variants. We prove that maintenance operations preserve the invariants of the structure, and that retrieval operations are sound and complete
Analog to Digital Memory Modeling for Test
No full textInternational audienceMemory testing is crucial as memories play an everincreasing important role in modern computing systems, to which a memory malfunction can lead to a system failure. Memory testing is commonly addressed by a functional testing approach that consist in verifying the manufactured memory function. Functional testing focuses on identifying memory functional failure mechanisms, which are modeled by Functional Fault Models (FFM), and for which dedicated test algorithms are developed to ensure their detection. However, as technology shrinks, fault mechanisms in memories become more complex, as well as their detection conditions. To anticipate any limitation, memory structural testing is investigated. Structural testing proposes to study the defect before the fault, as one or several manufactured defects or imperfections may be responsible for a fault. A structural test methodology for memory has been recently published and proposes to adapt the Cell-Aware test methodology from the digital domain to analog memories. As the resulting Structural Fault Models (SFM) for analog memory is compatible with digital test environment, this work proposes a digital SRAM modeling methodology, compatible with digital simulation and test environments, leveraging Fault Simulator for test algorithm coverage analysis, and Automatic Test Pattern Generator for dedicated and optimized defect-specific test generation
Gypscie-KG: Building a Logic-Based Approach for Knowledge Graph Data Integration View in ML Systems
No full textInternational audienceThis paper presents Gypscie-KG, a system that integrates heterogeneous machine learning (ML) data into a knowledge graph (KG) using logicbased rules to enable semantic queries and reasoning. It also explores the use of ChatGPT to support relational-to-graph schema mapping, demonstrating AI's potential in declarative data integration for complex ML systems
Advent of code, jour 17
No full textNational audienceJe vous avais présenté l'énigme du 7e jour de l’Advent of Code (mini calendrier de l’avent du code) dans un précédent article, je continue sur ma lancée en vous présentant l'énigme du 17e jour. Il est question cette fois-ci d'émuler un ordinateur 3 bits. L'occasion de voir l'assembleur sous un jour nouveau (le 17e, donc)
Efficient Federated Learning with Heterogeneous Data and Adaptive Dropout
No full textInternational audienceFederated Learning (FL) is a promising distributed machine learning approach that enables collaborative training of a global model using multiple edge devices. The data distributed among the edge devices is highly heterogeneous. Thus, FL faces the challenge of data distribution and heterogeneity, where non-Independent and Identically Distributed (non-IID) data across edge devices may yield in significant accuracy drop. Furthermore, the limited computation and communication capabilities of edge devices increase the likelihood of stragglers, thus leading to slow model convergence. In this paper, we propose the FedDHAD FL framework, which comes with two novel methods: Dynamic Heterogeneous model aggregation (FedDH) and Adaptive Dropout (FedAD). FedDH dynamically adjusts the weights of each local model within the model aggregation process based on the non-IID degree of heterogeneous data to deal with the statistical data heterogeneity. FedAD performs neuron-adaptive operations in response to heterogeneous devices to improve accuracy while achieving superb efficiency. The combination of these two methods makes FedDHAD significantly outperform state-of-the-art solutions in terms of accuracy (up to 6.7% higher), efficiency (up to 2.02 times faster), and computation cost (up to 15.0% smaller)
Using reflectance spectra and Pl@ntNet to identify herbarium specimens: a case study with <i>Lithocarpus</i>
No full textInternational audienceThe digitisation of plant collections is bringing large quantities of information into accessible electronic databases. However, in recent decades, traditional taxonomic work in collections has declined, meaning that more specimens are only determined to family or genus, particularly when lacking key identification structures. If unaddressed, large-scale digitisation risks widening the gap between well-studied species and those lacking data.Hyperspectral reflectance and computer vision are two emerging approaches for identifying species, but these have yet to be cross-compared for herbarium-based taxonomy. Using Lithocarpus species as a case study, we compared classification accuracy obtained from leaf reflectance spectra with computer vision (implemented via Pl@ntNet), a RGB (red, green, blue) image-based approach known to work well on specimens presenting reproductive structures. In the spectral approach, we assessed how much data are needed to optimise classification accuracy, how many species could be discriminated between, and whether close relatives were more frequently confounded.We found that Lithocarpus herbarium specimens were accurately identified to species from relatively small spectral datasets. Despite not incorporating reproductive structures, this was only 14% less accurate than [email protected] suggest these rapid, nondestructive leaf reflectance measurements, paired with computer vision, could fill identification gaps in collections, particularly for specimens lacking reproductive features.</div
4-tangrams are 4-avoidable
No full textInternational audienceA tangram is a word in which every letter occurs an even number of times. Thus it can be cut into parts that can be arranged into two identical words. The of a tangram is the minimum number of required cuts in this process. Tangrams with cut number one corresponds to squares. For , let denote the minimum size of an alphabet over which an infinite word avoids tangrams with cut number at most~. The existence of infinite ternary square-free words shows that . We show that , answering a question from Dębski, Grytczuk, Pawlik, Przybyło, and Śleszyńska-Nowak