HAL-MINES ParisTech
Not a member yet
27348 research outputs found
Sort by
Frugal AI: Introduction, Concepts, Development and Open Questions
No full textThis document aims to provide an overview and synopsis of frugal AI, with a particular focus on its role in promoting cost-effective and sustainable innovation in the context of limited resources. It discusses the environmental impact of AI technologies and the importance of optimising AI systems for efficiency and accessibility. It explains the interface between AI, sustainability and innovation. In fourteen sections, it also makes interested readers aware of various research topics related to frugal AI, raises open questions for further exploration, and provides pointers and references
Assessment of CAMS Radiation Service over France in different sky conditions
No full textInternational audienceThe assessment of solar irradiance variability is relevant for evaluating irradiance-based models, resource assessment and forecasting applications in the solar energy field. One well-established irradiance-based model database for solar project development is the Copernicus Atmosphere Monitoring Service (CAMS) through its CAMS Radiation Service (CRS) that offers historical all-sky solar irradiance estimates. In this work, the accuracy of the CRS GHI product over France is evaluated under different irradiance variability conditions by applying a sky condition classification method based on 1-minute Global Horizontal Irradiance (GHI) observations. A dense network of GHI measurements over France with more than 230 ground stations in the year 2015 is used as a case study.The classification method is based on a visual interpretation of GHI measurement patterns for the Baseline Surface Radiation Network (BSRN) station of Carpentras during the years 2012 and 2013, which forms a reference database. This reference database is composed of 280 manually classified hours in minute resolution for GHI into eight different classes (from clear sky to variable and overcast sky conditions). Ten variability indices (VIs) are applied in the classification scheme including the clear sky index (kc); the average, maximum and standard deviation of the absolute values for the first derivative of kc; the VIs proposed by Stein et al. (2012) and Coimbra et al. (2013); VIs based on envelopes curves obtained according to the local maxima and minima time-series; and three VIs that counts GHI values overpassing the clear sky irradiance in 3%, 5% and 10%. The classification model consists of three main steps: a discrimination filter, a probability classification approach and a median distance-based approach. The discrimination filter is a counting step that checks if the VIs are inside the Carpentras reference database domain for a particular class. The class with the most VIs will be the selected class. If the maximum number of VIs counted is the same for two or more classes, then a probability classification approach makes the class decision. This probability approach uses Kernel density estimation to calculate the neighborhood probability of a specific VI to be part of one of the eight classes. The class with the higher mean probability over all classes will be selected. Finally, for all the cases outside the domain of the reference database, the median distance-based approach with normalized VIs is applied as presented by Schroedter-Homscheidt et al. (2018). The evaluation of the CRS GHI over France is shown in Figure 1. The highest values of the Root Mean Square Deviation (RMSD) are found in class 6, which is mostly dominated by broken clouds. Also classes 4, 7 and 8 present large RMSD. The identification of this broken cloud conditions cluster is useful for further developments of the CRS algorithm in these challenging situations. Figure 1 – CRS RMSD in different sky conditions over France (hourly resolution)
Modeling of hydrogeochemical processes influencing uranium migration in anthropized arid environments with application to the Teloua aquifer
No full textInternational audience Sandstone-hosted uranium is mined in the Sahel regions of Niger. The Teloua aquifer is located beneath the oreprocessing facilities of one such former mine, COMINAK. The pores of the sandstone bedrock are partially filled by tosudite, a clay with sorption capacities. The local groundwater presents a strong oxidizing signature and very low water recharge. This study aims to determine the geochemical baseline of anthropogenic activity for uranium under such extreme conditions. The major and trace elements of both the contaminated and the pristine local groundwaters were sampled and analyzed to develop geochemical and reactive transport models. Kd distribution coefficients were calculated a posteriori from the mechanistic simulations. The entire water chemistry, with large variations in calcium, carbonate and sulfate concentrations, had to be taken into account to properly simulate the speciation and migration of U(VI) in the aquifer locally affected by the mining activities. U(VI) sorption significantly decreases during the propagation of the contaminant plume, due to the formation of Ca n UO 2 (CO 3 ) 3 (4-2n)-complexes that were clearly demonstrated by TRLFS acquisition. The sorption of UO 2 (CO 3 ) n(2-2n) can play a key role in the immobilization of U(VI). The mitigating factors for U(VI) are sorption on clay and the dispersion/ dilution of the contaminated source terms within the groundwater, in which the strong ternary complexes are less important. There should be an efficient immobilization of fixed anthropic uranium by natural attenuation once the contaminant source terms have become depleted.</div
Circumventing glioblastoma resistance to temozolomide through optimal drug combinations designed by systems pharmacology and machine learning
No full textInternational audienceBackground and Purpose Glioblastoma (GBM), the most frequent and aggressive brain tumour in adults, is associated with a dismal prognostic despite intensive treatment involving surgery, radiotherapy and temozolomide (TMZ)‐based chemotherapy. The initial or acquired resistance of GBM to TMZ appeals for precision medicine approaches to the design of novel efficient combination pharmacotherapies. Such investigation needs to account for the overexpression of the O6‐methylguanine‐DNA methyl‐transferase (MGMT) repair enzyme which is responsible for TMZ resistance in patients. Experimental Approach A comprehensive approach combining quantitative systems pharmacology (QSP) models and machine learning (ML) was undertaken to design TMZ‐based drug combinations circumventing the initial resistance to the alkylating agent. Key Results A QSP model representing TMZ cellular pharmacokinetics‐pharmacodynamics and dysregulated pathways in GBM was developed and validated using multi‐type time‐ and dose‐resolved datasets, available in control or MGMT‐overexpressing cells. In silico drug screening and subsequent experimental validation identified a strategy to re‐sensitise TMZ‐resistant cells consisting in combining TMZ with inhibitors of the base excision repair and of homologous recombination. Using ML, functional signatures of response to such optimal multi‐agent therapy were derived to assist decision‐making in patients. Conclusion and Implications We successfully demonstrated the relevance of combined QSP and ML to design efficient drug combinations re‐sensitising glioblastoma cells initially resistant to TMZ. The developed framework may further serve to identify personalised therapies and administration schedules by extending it to account for additional patient‐specific altered pathways and whole‐body features
« Fiabilité du diagnostic de performance énergétique : peut mieux faire, encore »
No full textInternational audienc
Speciation of uranium and radium during the treatment of acidic waters from legacy-mine (Le Cellier, France)
No full textInternational audienceThe knowledge of aqueous speciation of uranium and radium in mining context is important for the modelling strategies based on reactive transport. The use of thermochemical databases allows accessing to theoretical speciation when the water compositions are known. If the usual concentration of radium in natural or anthropized waters is too low to have access to speciation experimentally, this can be achieved for uranium(VI), e.g. using time-resolved laser-induced fluorescence spectroscopy (TRLFS). In this work, theoretical radium and uranium inorganic and organic speciation were calculated using the water compositions collected in the legacy-mine site of Le Cellier (Lozère, France) currently under monitoring after its closure and decommissioning, and database file extracted from the Prodata database for the PhreeqC and Orchestra codes. We also have measured TRLFS uranium spectra, which allows monitoring the uranium(VI) evolution from sulphate-like UO2(SO4)n2–2n complexes, at the beginning of the treatment, to characteristic CanUO2(CO3)3(4–2n)- complexes towards the end of the treatment, with an expected decrease of the total uranium concentration. Inorganic thermodynamic calculations are in excellent agreement with the spectroscopic attributions all along the treatment, whatever the speciation code used. The influence of natural organic matter, as part of the dissolved organic carbon outside of the legacy-mine perimeter, cannot be ruled out but is not straightforward to ascertain. The effluent seems only to disturb slightly the uranium speciation in the local stream from UO2CO3(aq) to CanUO2(CO3)3(4–2n)- due to calcium increase
Les risques en images
No full textLes risques en imagesEntre juin 2019 et mai 2020, le Centre de recherche sur les Risques et les Crises (CRC) de l’École Nationale Supérieure des Mines de Paris publia sur son site internet une série intitulée « Les risques en images ». L’objectif de ce projet pluridisciplinaire et collaboratif était de décrypter, chaque mois, une peinture, une affiche, une photographie, une gravure, un graphique ou un modèle représentant un risque, un accident ou une situation de crise. La direction éditoriale des Risques en images fut assurée par Aurélien Portelli, chargé de recherche au CRC. Le document suivant regroupe les douze numéros de la série
Assessing the Engineering Design Costs to Meet Environmental Regulations: The Case of Packaging
No full textInternational audienceThis novel contributions reveal how environmental regulations drive engineering design costs, focusing on the emblematic case of packaging. Using a regulatory database and simulation-based modeling, we evaluate functional expansion as a key driver of cost escalation, identifying its volume effect (rising costs from added environmental functions) and scope effect (increased interdependencies among ecosystem actors). The findings offer a simulated cost envelope to support engineering design teams in their forecasts, but also underscore the hurdles of sustainably managing these regulatory-driven costs in the packaging product system, by benchmarking cost trajectories against sustainability metrics, such as carbon pricing
Integrated optimization and machine learning: an application to predictive maintenance
No full textInternational audienceThis paper examines how optimization and machine learning can be combined to improve predictive maintenance in wind farms. Optimization integrated with machine learning creates powerful methods to solve complex problems by improving solution search strategies and enhancing decision-making processes, in contrast with the usual sequential approach. This synergy enables algorithms to learn from data and guide optimization procedures more effectively. In this proposal, we present an application of this combination for predictive maintenance, where a MIP schedules tasks to minimize downtime while maximizing system reliability and machine learning models, including random forrest and isolation forrest, anticipate equipment failures. The model we propose mobilizes ensemble learning with a loop such that the performance of the corresponding tasks planning allows to adjust the threshold for aggregating the results of the different prediction models. The predictions are adjusted through iterations by the results of the optimization. The optimization then leads to a return to the ensemble learning, becoming similar to a layer of a learning algorithm. Experimentations were carried out on industrial data
Developing a Digital Decision-Support Tool for Personalized treatment of Metastatic Thyroid Cancer: Developing a Digital Decision-Support Tool for Personalized treatment of Metastatic Thyroid Cancer
No full textInternational audienceEach year, a high number of new cases of thyroid cancer are identified. Although this disease has a good prognosis, it is often immediately metastatic, which requires removal of the thyroid body, followed by iratherapy (administration of iodine 131) in order to eradicate the various metastatic sites present.Until now, iodine 131 administration protocols (activities of iodine 131 to be administered, number of iratherapy sessions and interval between two consecutive sessions) are conducted empirically and a large inter-individual variability in responses to treatment has been observed as well as toxicities induced in the more or less long term. In order to optimize the effectiveness-toxicity balance, it is essential to provide clinicians with a decision support tool allowing them to perform in silico simulations of the evolution of thyroglobulin concentrations depending on the chosen diode 131 administration protocol by the therapist.An in silico simulator will be presented built from a mathematical model managed by a set of parameters whose values are specific to each individual, in particular the parameter (Td) of the doubling time of tumor cells under treatment is identified as a parameter key allowing discrimination from the first weeks of treatment, responding patients and patients refractory to iodine 131. In addition, in the case of a given responding patient, the simulator then makes it possible to propose to the clinician a diagram effective administration while using the minimum amount of iodine 131, so as to minimize the probability of the appearance of possible iatrogenic pathologies induced by excessive irradiation.The precision of this personalized therapeutic planning simulator is conditioned by a good estimation of the individual parameters of each patient, in particular the Td parameter. In this sense, the use of Artificial Intelligence could help to refine the estimation of these parameters.Additionally, this work provides proof of concept that the development of powerful digital tools could help enhance the precision of personalized medicine