Sorbonne Paris North University (French: Université Sorbonne Paris Nord) is a public university based in Paris, France. It is one of the thirteen universities that succeeded the University of Paris in 1968. It is a multidisciplinary university located in north of Paris, in the municipalities of Villetaneuse, Saint-Denis, La Plaine Saint-Denis, Bobigny and Argenteuil.Successively named “Université Paris XIII”, “Université Paris-Nord”, “Université Paris 13 Paris Nord”, then “Université Paris 13”, it has been known by several names during the last half century. Most recently it was renamed "Université Sorbonne Paris North" on January 1, 2020.The University Sorbonne Paris Nord is a major teaching and research center located north of Paris. It has five campuses, spread over the two departments of Seine-Saint-Denis and Val d'Oise: Villetaneuse, Bobigny, Saint-Denis, the Plaine Saint-Denis and Argenteuil.The university has more than 25,000 students in initial or continuing training, in many fields: Health, Medicine and Human Biology - Letters, Languages, Humanities and Societies - Law, Political and social sciences - Communication sciences - Economics and management.
The purpose of this study is to determine the mechanical characteristics for inclusion composite materials by several homogenization methods. Firstly, it involves modeling the (RVE) using the semi-analytical homogenization method, then formulating and implementing a periodic homogenization technique, where we proposed introducing additional degrees of freedom supporting the components of macroscopic deformations applying at composite "Concrete". It is necessary to introduce behavior laws and damage models for composite materials. For this, we develop damage expressions based on material behavior models in order to analyze local fields and introduce failure criteria. Simulation by the semi-analytical Mori-Tanaka model gives good results for a two-phase material when the heterogeneity concentration is less than 30%. And as for modeling by the proposed periodic homogenization method, it gives us very satisfactory results regarding the prediction of the effective mechanical characteristics of composite materials. The percentage of inclusions as well as their shape have an influence on the latter. A parametric study was carried out to determine the influence of variations in the mechanical properties of mortar on those of effective "concrete" composite. The damage models developed allowed us to reproduce the softening behavior of the composite materials, which is progressively reduced as microcracks develop.
This research work involves modeling reinforced concrete and steel fibers structural elements, and investigating the safety domain of these structural elements. The main aim is to simulate their behavior until failure, and to estimate their failure probability. Firstly, the Hasofer-Lind model coupled with non-linear computation using the response surface method (RSM) is applied to estimate the reliability index, the scenarios and the failure probability. Furthermore, the direct Monte-Carlo simulation method is applied to assess the failure probability of the systems studied. We validate our formulation and modeling, by comparing the results of our calculations with experimental results and those obtained by other authors. The tests were carried out on RC beam, RC frame and SFRC pile. This comparison shows the efficiency and performance of each model used in this study. Among other things, after studying all possible cases of distribution laws for random load-resistance variables, it was observed that a log-normal law gives a very low failure probability, thus guaranteeing a certain level of safety. A parametric study is carried out for the SFRC pile to determine the influence of variation in modulus of elasticity and concrete strain in compression.
Transformers have become the dominant architecture across a wide range of domains, largely due to the effectiveness of multi-head attention in capturing diverse representation subspaces. However, standard multi-head attention activates all heads uniformly for every input, regardless of task requirements or input complexity. In many scenarios, particularly for coarse-grained tasks such as text classification, the relevant information is often global and does not require the full diversity of attention heads. As a consequence, using a fixed number of heads can introduce unnecessary computational cost or lead to suboptimal performance when the allocation does not match the input. To address this limitation, we introduce BudgetFormer, a Transformer architecture equipped with an adaptive multi-head attention mechanism that dynamically allocates computational resources. Our approach learns, for each input, both a head budget corresponding to the number of attention heads required, and a relevance distribution that selects the most informative heads. We also propose a training strategy based on an exploration and exploitation trade-off, allowing the model to discover effective head configurations before converging to efficient usage patterns. Experiments on text classification tasks of varying complexity show that our method reduces inference cost in terms of FLOPs and memory, while also achieving performance that can surpass standard full multi-head attention. These results highlight the potential of adaptive head allocation as a principled approach to improving both efficiency and effectiveness in Transformer models.
Lower limbs can play a major compensating role for sagittal malalignment; however, little is known about the different types of compensation. This study aimed to identify different patterns of lower limb compensation and to determine which parameters may affect the recruitment of knee flexion versus hip extension. This study included adult spinal deformity (ASD) patients with full-body X-rays in erect position from a multicentric prospective database. All parameters were measured at baseline: demographics, clinical scores and radiographic parameters: pelvic parameters, pelvic incidence-lumbar lordosis (PI-LL) mismatch, T1 pelvic angle (TPA), sacro-femoral angle (SFA), knee flexion angle (KA), ankle dorsi-flexion angle (AA), pelvic shift (PSh), hip and knee osteoarthritis (OA) grade. A K-means cluster analysis was conducted to identify patterns of lower limb compensation based on SFA and KA. The optimal number of clusters was determined using the silhouette score. The different parameters were then compared across clusters. 871 ASD patients were included, of whom 66.9
