Search

International audience; Cet article a pour objectif de rendre compte d’une première expérience professionnelle d’une jeune linguiste en formation dans une entreprise de l’Intelligence Artificielle (IA). Il n’a pas la prétention d’établir l’état de l’art de nouvelles technologies numériques, encore moins d’apporter des propositions théoriques. Il s’agit plutôt de l’exposé d’un retour d’expérience d’une doctorante en linguistique ayant passé un peu plus de quatre mois dans un milieu professionnel où les métiers de langue sont reconnus et recherchés.Dans cet article, je vais d’abord essayer de présenter le secteur économique ainsi que la place et la nature de l’activité de l’entreprise Snips. Ensuite, je parlerai de l’environnement de travail dans lequel j’ai passé l’expérience en question. Enfin, je vais exposer précisément ce qu’on attendd’un linguiste dans ce secteur.

Considering the linear motion of particleboards in the production line, the detection of surface defects in particleboards is a major challenge. In this paper, a method based on moving target tracking is proposed for the detection of surface defects in particleboards. To achieve this, the kernel correlation filter (KCF) target tracking algorithm was modified with the median flow algorithm and used to capture the moving targets of surface defects. The defect images were extracted by a Sobel operator, and the defect number, the defect area, and the degree of damage were calculated. The level of surface defect in particleboards was evaluated by fuzzy pattern recognition. Experiments were then carried out to prove the effectiveness and accuracy of the proposed method.

A compound control strategy is investigated for Medium Density Fiberboard (MDF) continuous hot pressing electrohydraulic servo system (EHSS) with uncertainties and input saturation. Firstly, a hyperbolic tangent function is applied to approximate saturation nonlinearity in the system. And thus the mathematical model is continuous and differentiable. Subsequently, the slab thickness tracking controller is constructed by using a dynamic surface control (DSC) method, which introduces first-order low-pass filters to calculate derivatives of virtual control input in each step. Compared with the conventional backstepping controller, complexity of the design procedure is alleviated obviously. Moreover, a composite disturbance of uncertainties and input saturation is estimated by a nonlinear disturbance observer for compensation of the control law. Finally, an appropriate Lyapunov function is chosen to prove that all signals of the closed-loop system are semiglobally uniformly ultimately bounded and the tracking error converges to zero asymptotically. Numerical simulation results are also exhibited to authenticate and validate the benefits of the proposed control scheme.

Fault-tolerance is one of important issues in wireless sensor network (WSN) since it is critical in real deployed environments to realize network stability and reduce demand times. In this paper, we propose primary-backup technique by creating a backup path for every sensor on a primary path of data transmission. Especially, we take high reliability path as selected primary or backup path method. The experimental results show that the algorithm not only has the low packet delivery ratio characters, but also ensures the reliability of topology paths and extends the network life-cycle efficiently.

In order to realize high-precision position tracking control for (Medium Density Fiberboard, MDF) continuous hot press hydraulic position servo system in the presence of interior parameter perturbations and external load disturbance, an adaptive backstepping sliding mode control strategy with fuzzy disturbance observer (FDO) is presented. Firstly, a modified FDO is designed to estimate and compensate the interior parameter perturbations and external load disturbance of the system. Secondly, the backstepping sliding mode controller is designed and adaptive algorithm is used to online estimate the uncertainty of the system. Thirdly, the convergence of tracking error and the asymptotic stability of system are strictly proved according to Lyapunov stability theory. The results of simulation experiments indicate that the controller has strong robustness against the interior parameter perturbations and external load disturbance. Moreover, the controller has good performance including fast response, high tracking accuracy and small steady-state error.

With multiprocessor systems, redundant scheduling is a technique that trades processing power for increased reliability through redundancy. One novel approach, called primary-backup task scheduling, is a large number of used in hard real-time multiprocessor systems to guarantee the deadlines of tasks faults. In this paper, we analysis and compare the scheduling algorithm based on primary-backup replication in current distributed system. Finally, the important features that fault-tolerant Task Scheduling in Multiprocessor Systems based on primary-backup are summarized, and the future research strategies and tends are given.

In WSNs, energy conservation is the primary goal, while throughput and fault tolerance are other important factor. According to wireless sensor network nodes energy consumption, hardware failure and communication factors, in this paper, an evolving network method based on complex network theory of small-world Kleinberg model is proposed, which uses long links and short links in partial view information to build inter-cluster topology. Theoretical analysis and simulations show that the network lifetime is longer 27% than DECDC in the WSN topology which uses this method when the node failure probability is 0.2. And it has a good balance of energy consumption and lower cost. Thus the topology which uses the method to build has good fault-tolerant and robustness.