Portail des publications scientifiques IMT Mines Alès
Not a member yet
5198 research outputs found
Sort by
SkeltyMLOps: Orchestrating Collaborative MLOps Activities
No full textInternational audienceWith the increasing adoption of machine learning (ML), the need to manage, in a tailored manner, the complete lifecycle of this new type of software arises. MLOps (Machine Learning Operations) extends DevOps to address the specific challenges of development and lifelong management of ML-powered software. Despite the increasing attention to this domain, practitioners still lack an accessible, modular architecture to support MLOps projects. To address this lack, this paper proposes SkeltyMLOps, a reference architecture designed to promote collaboration between the diverse actors involved in MLOps processes. The architecture is derived from a thorough literature review, from which we extracted and clustered, using a Large Language Model (LLM), a comprehensive list of MLOps actors and activities. These clusters were then used to guide the architectural decomposition and component design of SkeltyMLOps. The originality of our proposed reference architecture is (i) that it clearly aligns its components with the responsibilities of MLOps actors, covering the identified dimensions of MLOps processes, and (ii) mediates collaboration by orchestrating interactions between the different types of actors. This architecture provides a foundation for building collaborative ML-powered software
Comprehensive Review of Open-Source Fundus Image Databases for Diabetic Retinopathy Diagnosis
No full textInternational audienceDatabases play a crucial role in training, validating, and comparing AI models for detecting retinal diseases, as well as in clinical research, technology development, and healthcare professional training. Diabetic retinopathy (DR), a common diabetes complication, is a leading cause of vision impairment and blindness worldwide. Early detection and management are essential to prevent irreversible vision loss. Fundus photography, known for being economical and non-contact, is a widely applicable gold standard method that offers a convenient way to diagnose and grade DR. This paper presents a comprehensive review of 22 open-source fundus retinal image databases commonly used in DR research, highlighting their main characteristics and key features. Most of these datasets were released between 2000 and 2022. These databases are analyzed through an in-depth examination of their images, enabling objective comparison using color space distances and Principal Component Analysis (PCA) based on 16 key statistical features. Finally, this review aims to support informed decision-making for researchers and practitioners involved in DR diagnosis and management, ultimately improving patient outcomes
Local explanations of aggregation results by fuzzy linguistic rules: application to avalanche risk
No full textInternational audienceExplainable Artificial Intelligence (XAI) is relatively new in offering the possibility for intelligent systems to give robust motivations for their decisions and behaviours. Since XAI is human-centred, it has strong connections with fuzzy systems. In this paper, we consider the local explanation of the output of a set of fuzzy linguistic rules from the contributions of the inputs to the output. The proposed contributions are defined from the average of the gradients on the line linking the start point to the end point. This approach ensures that the variation in the output is equal to the sum of the contributions of each input variable to the output. The considered application is based on avalanche expert knowledge expressed by a set of fuzzy rules that combines different physical variables to build an ordinal fuzzy scale of avalanche vigilance ("Relaxed", "Suspicious", "Alert", "Gamble"). The level of vigilance to be applied locally, along a mountain ski route, is explained in a way that is close to expert reasoning in order to decide modifications of the route
Prediction of Critical Force Based on Time-Domain Near Infrared Spectroscopy and Electromyography With Explainable Machine Learning
No full textInternational audienc
Toward a Model- and Pattern- Based Method for the Engineering and the Maintenance of Digital Twin Systems
No full textInternational audienc
Effect of Recycled Glass Fibre Reinforced Polymer on the Mechanical Strength of Cement Mortar
No full textInternational audienceGlass fibre-reinforced polymers, thanks to their high performance combined with low weight, are widelyused in many sector. However, at the end of their life cycle, these materials are difficult to recycle, due tothe dfficulty of separating the glass fibre from the matrix (thermoset resin). The incorporation of recycledglass fibre reinforced polymer (rGFRP) in cementitious materials is an interesting recycling and valorisationmethod. This involves cutting glass composites into small pieces and using them as aggregates. However,add rGFRP particles as aggregates in cementitious materials generally leads to reduction in workability andan improvement in certain mechanical performances. The aim of this study is to examine the effect ofdifferent size fractions (0.063 mm, 0.16 mm, 0.63 mm, 1.25 mm and 2 mm) of rGFRP on the mechanicalbehavior of mortar. In this way 10% by volume of the sand is replaced by rGFRP particles. The density ofrGFRP particles is lower than that of sand, ranging from 1.6 to 2.7 g/cm3. Six mortar formulationsincorporating rGFRP particles were carried out, varying the particle size in the different mixes. Eachformulation is designed by a letter from A to F. Formulations A, C and F consist of a single rGFRP particlesize of 0.063 mm, 0.63 mm and 2 mm respectively. For forlumations B, D and E, a mixture of differentparticle sizes is performed, consisting of 1.25 – 0.16 mm, 1.25 – 0.063 and 0.16 mm respectively, and acomplete mixture of all particle sizes for formulation E. Mortars are manufactured in a accordance withstandard EN 196–1, in an automated mixer. After manufacture, the consistency is measured (spreaddiameter), then prismatic samples (4 × 4 × 16 cm3) are made, for subsequent measurement of density,mechanical behavior, water-accessible porosity and microscopic observation. The results obtained showacceptable workability for all formulations and a decrease in the density of mortars incorporating rGFRPdepending on their size, which could be justified by the lower density of rGFRP particles compared withsand particles. Compressive and flexion tests at 7 days and 7 months of curing show a logical inrease inperformance over time due to cement hydration. Figure 1a shows the results obtained in compression test,where the values at 7 months all reach 40 Mpa. Figure 1b shows the percentage variation in compressivestrength of samples with rGFRP. A reduction in comprssiv strength is observed for all samples althoughthis reduction is less significant for samples E and F (<10%). However, the samples A, which contains onlyfine particles, shows the greatest reduction in compressive strength, reaching 31%.Figure 2a shows the results for the flexural strength, with values of around 5 to 6 Mpa at 7 days and 10Mpa at 7 months curing of sample. Figure 2b shows the variation in flexural strength of samples containingrGFRP. A slight decrease in flexural strength is observed for samples B, C, D and E of less than 10% anda slight improvement of around 2.7% for sample F, which showed less brittle fracture during test. A studyof water-accessible porosity after 7 months curing shows a slight increase in porosity for the samples withrGFRP particles, which could account for the drop in density and compressive performance.Finally, microscopic observation of the samples after 7 months curing reveals the presence of rGFRPparticles in the matrix. Particle agglomeration is observed for the formulations with very fine particles(Sample A), and for the formulations containing 2 mm particles (Sample E and F), glass fibre filaments areobserved dispersed in the matrix, potentially improving resistance to crack propagation. However, the vastmajority of fibres are astill grouped in bundles. Thus, after 7 months of curing, the non-alkali resistant glassfibres remain present and show no degradation
Optic flow and cycling effort: Where to look to go faster
No full textInternational audienceOptic flow can significantly influence the perception and exertion of effort. In this study, we investigated the effects of exposure to proximal and distal areas of the optic flow field on exerted and perceived cycling effort. Thirty cyclists participated in two 20-min cycling trials within a virtual reality environment, with the goal of surpassing the power output achieved during a baseline trial. During these trials, they viewed the environment through a proximal or distal window, in counterbalanced order. We measured the cyclists' exerted effort on a bicycle trainer, and recorded their responses regarding perceived effort and psychological momentum experience. A one-way repeated measures ANCOVA with average baseline power as a covariate revealed a significant difference in exerted effort between the proximal and distal condition, with higher average exerted effort in the proximal condition. However, a significant interaction effect between condition and baseline power indicated that the beneficial effect of the proximal condition was mainly present for lower-level cyclists. We observed no significant differences in perceived effort or psychological momentum. These findings provide novel insights into the relation between optic flow and cycling effort, and call for new research on the mechanisms underlying this relation
Fabrication de tapes thermoplastiques en PET recyclé : une approche innovante exploitant la capillarité
No full textNational audienceLe Polyéthylène Teraphtalate (PET) est un des polymères le plus utilisé dans le monde dû à ses caractéristiquesmécaniques et ses hautes performances [1]. Cependant, son caractère persistant dans l’environnement présente denombreux problèmes en termes de pollution environnementale. C’est pourquoi aujourd’hui le PET est le polymère le plusrecyclé mais les méthodes de recyclage actuelles présentent de nombreuses limitations. L’énergie utilisée pour lerecyclage du PET est importante et on sait maintenant que les procédés actuels dégradent le matériau au fil des cycles derecyclage [2]. C’est pourquoi dans une démarche d’économie circulaire, une approche innovante pour recycler le PETsous forme de tapes thermoplastiques renforcés par des fibres est proposée. Dans cette étude, nous nous intéresserons àau procédé de fabrication des tapes thermoplastiques par dissolution de PET, en se focalisant sur l’imprégnation des fibreset les phénomènes capillaires et visqueux engendrés . Cet aspect du procédé est déterminant pour maximiserl’imprégnation et réduire les porosités induites dans le matériau final [3]. Cette étude amène une nouvelle voie defabrication de tapes thermoplastiques en combinant recyclabilité et économie circulaire avec la performanc
Enhancing associative classification on imbalanced data through ontology-based feature extraction and resampling
No full textInternational audienceAssociative classification models are valuable for discovering relationships within heterogeneous data systems, making them particularly useful for data integration tasks. However, they struggle with imbalanced and sparse data. This paper addresses the problem of imbalanced classification in building maintenance data by providing several updates based both on algorithms and preprocessing. Experiments conducted on real maintenance datasets demonstrate significant improvements in accuracy and precisio