International Professional University of Technology in Nagoya Repository
Not a member yet
15131 research outputs found
Sort by
Prédiction multi-modale à l'aide d'apprentissage PRObabiliste de Mouvement Primitives
No full textInternational audienceThis paper proposes a method for multi-modal prediction of intention based on a probabilistic description of movement primitives and goals. We target dyadic interaction between a human and a robot in a collaborative scenario. The robot acquires multi-modal models of collaborative action primitives containing gaze cues from the human partner and kinetic information about the manipulation primitives of its arm. We show that if the partner guides the robot with the gaze cue, the robot recognizes the intended action primitive even in the case of ambiguous actions. Furthermore, this prior knowledge acquired by gaze greatly improves the prediction of the future intended trajectory during a physical interaction. Results with the humanoid iCub are presented and discussed.Dans ce papier, nous proposons une méthode de prédiction multi-modale de l'intention basé sur une description probabiliste de primitives de mouvements et de buts. On s'interesse ici à un scénario d'interaction collaborative entre un humain et un robot. Le robot modelise l'action collaborative de manière multi-modale, à l'aide de primitives contenant des informations visuelles (orientation du regard du partenaire) ainsi que des informations sur la dynamique de ses propre bras. Nous montrons dans cette étude que si le partenaire guide le robot en utilisant son regard, le robot reconnait l'action attendu par le partenaire et ce, même dans le cas où les mouvements sont ambigus. Nous montrons aussi qu'en guidant le début du mouvement du robot physiquement , le robot peut même afiner sa trajectoire pour respecter encore mieux la volonté de son partenaire. Finalement, en utilisant les deux modalités, le robot peut utiliser l'information visuelle comme un a-priori sur l'action a effectuer, ce qui permet d'améliorer la reconnaissance de la trajectoire attendue lors d'interaction physique
Statistical Prioritization for Software Product Line Testing: an Experience Report
No full textInternational audienc
Statistical Model Checking of LLVM Code
No full textWe present our work in providing Statistical Model Checking for programs in LLVM bitcode. As part of this work we develop a semantics for programs that separates the program itself from its environment. The program interact with the environment through function calls. The environment is furthermore allowed to perform actions that alter the state of the C-program-useful for mimicking an interrupt system. On top of this semantics we build a probabilistic semantics and present an algorithm for simulating traces under that semantics.. This paper also includes the development of the new tool component Lodin that provides a statistical model checking infrastructure for LLVM programs. The tool currently implement standard Monte Carlo algorithms and a simulator component to manually inspect the behaviour of programs. The simulator also proves useful in one of our other main contributions; namely producing the first tool capable of doing importance splitting on LLVM code. Importance splitting is implemented by integrating Lodin with the existing statistical model checking tool Plasma-Lab
La co-production de contenu par le numérique pour optimiser et renforcer l’apprentissage hybride : retour d’expériences de pédagogie co-active
No full textInternational audienc
A General and Accurate Formula for the Beamwidth of 1-D Leaky-Wave Antennas
No full textInternational audienceIn this paper, a general formula for the half-power beamwidth of one-dimensional leaky-wave antennas (1-D LWAs) is presented. With respect to previous beamwidth formulas found in the literature, the new formula allows for both arbitrary propagation wavenumber and length of the antenna. The pointing angle of the beam is allowed to be arbitrary, and may be near broadside or endfire, or any angle in between. The beamwidth is also allowed to be arbitrary, though a simpler approximate expression is obtained under the assumption of a narrow beam. Numerical results confirm the accuracy of the new beamwidth formula. Furthermore, it is shown that the previous existing formulas are approximate limiting cases of this more general result. This new formula will provide a very useful tool for obtaining more reliable results for the beamwidth of any type of 1-D LWA. © 2017 IEEE
Energy-efficient Joint Power Allocation in Uplink Massive MIMO Cognitive Radio Networks with Imperfect CSI
No full textInternational audienceIn this paper, a joint pilot and data power allocationproblem with max-min fair energy efficiency (EE) guaranteein the uplink massive multiple-input multiple-output (MIMO)cognitive radio networks (CRNs) is investigated. Given the fractionalobjective function, channel estimation errors, and interuserinterference, the joint allocation problem is formulated asa nonconvex and NP-hard problem. To tackle this, we transformthe original problem into its convex form by introducing auxiliaryvariables and variable substitution, then address it with the helpof the Lagrangian dual method. Since the optimization variablesare interrelated and interact on each other, it is difficult todirectly obtain the closed-form solution to this problem. To settlethis issue, we propose an alternately iterative algorithm to achievethe optimal power policy by a gradient-based adaption method,with its corresponding optimal Lagrangian multipliers obtainedby the subgradient method. Numerical results show that theproposed approach has the best minimum EE performance anddecent spectral efficiency (SE) performance. Besides, comparedto other schemes, significant saving in total transmit power andgood cognitive user (CU) fairness are achieved by the proposedalgorithm
Two distinct bifurcation routes for delayed optoelectronic oscillators
No full textInternational audienc
Reservation, a tool to reduce the balking effect and the probability of delay
No full textInternational audienceWe investigate a threshold reservation policy implemented within a single customer's class. From an explicit performance analysis, we prove that the potential of reservation is into the reduction of the balking effect together with a higher server's utilization. However, it also may result in a higher expected waiting time and more abandonment. We conclude that reservation can be efficiently implemented in large systems, under high workload situations, with a low waiting aversion and a low impatience
Toward Optimal Run Racing: Application to Deep Learning Calibration
No full textThis paper aims at one-shot learning of deep neural nets, where a highly parallel setting is considered to address the algorithm calibration problem - selecting the best neural architecture and learning hyper-parameter values depending on the dataset at hand. The notoriously expensive calibration problem is optimally reduced by detecting and early stopping non-optimal runs. The theoretical contribution regards the optimality guarantees within the multiple hypothesis testing framework. Experimentations on the Cifar10, PTB and Wiki benchmarks demonstrate the relevance of the approach with a principled and consistent improvement on the state of the art with no extra hyper-parameter
