The Experts below are selected from a list of 20043 Experts worldwide ranked by ideXlab platform

Isabelle Bloch - One of the best experts on this subject based on the ideXlab platform.

  • tradi tracking deep Neural Network Weight distributions
    European Conference on Computer Vision, 2020
    Co-Authors: Gianni Franchi, Andrei Bursuc, Emanuel Aldea, Séverine Dubuisson, Isabelle Bloch
    Abstract:

    During training, the Weights of a Deep Neural Network (DNN) are optimized from a random initialization towards a nearly optimum value minimizing a loss function. Only this final state of the Weights is typically kept for testing, while the wealth of information on the geometry of the Weight space, accumulated over the descent towards the minimum is discarded. In this work we propose to make use of this knowledge and leverage it for computing the distributions of the Weights of the DNN. This can be further used for estimating the epistemic uncertainty of the DNN by aggregating predictions from an ensemble of Networks sampled from these distributions. To this end we introduce a method for tracking the trajectory of the Weights during optimization, that does neither require any change in the architecture, nor in the training procedure. We evaluate our method, TRADI, on standard classification and regression benchmarks, and on out-of-distribution detection for classification and semantic segmentation. We achieve competitive results, while preserving computational efficiency in comparison to ensemble approaches.

  • TRADI: Tracking deep Neural Network Weight distributions
    arXiv: Learning, 2019
    Co-Authors: Gianni Franchi, Andrei Bursuc, Emanuel Aldea, Séverine Dubuisson, Isabelle Bloch
    Abstract:

    During training, the Weights of a Deep Neural Network (DNN) are optimized from a random initialization towards a nearly optimum value minimizing a loss function. Only this final state of the Weights is typically kept for testing, while the wealth of information on the geometry of the Weight space, accumulated over the descent towards the minimum is discarded. In this work we propose to make use of this knowledge and leverage it for computing the distributions of the Weights of the DNN. This can be further used for estimating the epistemic uncertainty of the DNN by sampling an ensemble of Networks from these distributions. To this end we introduce a method for tracking the trajectory of the Weights during optimization, that does not require any changes in the architecture nor on the training procedure. We evaluate our method on standard classification and regression benchmarks, and on out-of-distribution detection for classification and semantic segmentation. We achieve competitive results, while preserving computational efficiency in comparison to other popular approaches.

Hongfei Ding - One of the best experts on this subject based on the ideXlab platform.

  • adaptive sliding mode control of dynamic system using rbf Neural Network
    Nonlinear Dynamics, 2012
    Co-Authors: Hongfei Ding
    Abstract:

    This paper presents a robust adaptive sliding mode control strategy using radial basis function (RBF) Neural Network (NN) for a class of time varying system in the presence of model uncertainties and external disturbance. Adaptive RBF Neural Network controller that can learn the unknown upper bound of model uncertainties and external disturbances is incorporated into the adaptive sliding mode control system in the same Lyapunov framework. The proposed adaptive sliding mode controller can on line update the estimates of system dynamics. The asymptotical stability of the closed-loop system, the convergence of the Neural Network Weight-updating process, and the boundedness of the Neural Network Weight estimation errors can be strictly guaranteed. Numerical simulation for a MEMS triaxial angular velocity sensor is investigated to verify the effectiveness of the proposed adaptive RBF sliding mode control scheme.

Juntao Fei - One of the best experts on this subject based on the ideXlab platform.

  • Robust adaptive nonsingular terminal sliding mode control of MEMS gyroscope using fuzzy-Neural-Network compensator
    International Journal of Machine Learning and Cybernetics, 2016
    Co-Authors: Wei Yan, Shixi Hou, Weifeng Yan, Juntao Fei
    Abstract:

    To attenuate the effect of time-varying parameters, quadrature errors, and external disturbances and realize finite-time control, a robust adaptive nonsingular terminal sliding mode (NTSM) tracking control scheme using fuzzy-Neural-Network (FNN) compensator is presented for micro-electro-mechanical systems (MEMS) vibratory gyroscopes in this paper. By introducing a nonsingular terminal sliding mode manifold, a novel terminal sliding mode controller is designed for MEMS gyroscopes, while ensuring the control system could reach the sliding surface and converge to equilibrium point in a finite period of time from any initial state. In the presence of unknown model uncertainties and external disturbances, an adaptive fuzzy-Neural-Network controller is employed to compensate such system nonlinearities and improve the tracking performance. Online fuzzy-Neural-Network Weight tuning algorithms are derived in the sense of Lyapunov stability theorem to guarantee the Network convergence as well as stable control performance. Numerical simulations for a MEMS gyroscope are provided to justify the claims of the proposed adaptive fuzzy-Neural-Network control scheme and demonstrate the satisfactory tracking performance and robustness.

Gianni Franchi - One of the best experts on this subject based on the ideXlab platform.

  • tradi tracking deep Neural Network Weight distributions
    European Conference on Computer Vision, 2020
    Co-Authors: Gianni Franchi, Andrei Bursuc, Emanuel Aldea, Séverine Dubuisson, Isabelle Bloch
    Abstract:

    During training, the Weights of a Deep Neural Network (DNN) are optimized from a random initialization towards a nearly optimum value minimizing a loss function. Only this final state of the Weights is typically kept for testing, while the wealth of information on the geometry of the Weight space, accumulated over the descent towards the minimum is discarded. In this work we propose to make use of this knowledge and leverage it for computing the distributions of the Weights of the DNN. This can be further used for estimating the epistemic uncertainty of the DNN by aggregating predictions from an ensemble of Networks sampled from these distributions. To this end we introduce a method for tracking the trajectory of the Weights during optimization, that does neither require any change in the architecture, nor in the training procedure. We evaluate our method, TRADI, on standard classification and regression benchmarks, and on out-of-distribution detection for classification and semantic segmentation. We achieve competitive results, while preserving computational efficiency in comparison to ensemble approaches.

  • TRADI: Tracking deep Neural Network Weight distributions
    arXiv: Learning, 2019
    Co-Authors: Gianni Franchi, Andrei Bursuc, Emanuel Aldea, Séverine Dubuisson, Isabelle Bloch
    Abstract:

    During training, the Weights of a Deep Neural Network (DNN) are optimized from a random initialization towards a nearly optimum value minimizing a loss function. Only this final state of the Weights is typically kept for testing, while the wealth of information on the geometry of the Weight space, accumulated over the descent towards the minimum is discarded. In this work we propose to make use of this knowledge and leverage it for computing the distributions of the Weights of the DNN. This can be further used for estimating the epistemic uncertainty of the DNN by sampling an ensemble of Networks from these distributions. To this end we introduce a method for tracking the trajectory of the Weights during optimization, that does not require any changes in the architecture nor on the training procedure. We evaluate our method on standard classification and regression benchmarks, and on out-of-distribution detection for classification and semantic segmentation. We achieve competitive results, while preserving computational efficiency in comparison to other popular approaches.

Fang Wang - One of the best experts on this subject based on the ideXlab platform.

  • distributed adaptive coordination control for uncertain nonlinear multi agent systems with dead zone input
    Journal of The Franklin Institute-engineering and Applied Mathematics, 2016
    Co-Authors: Fang Wang, Zhi Liu, Yun Zhang, Bing Chen
    Abstract:

    Abstract In this paper, a distributed design scheme is developed for consensus tracking control of multi-agent system with nonlinear input under a Weighted directed graph topology. Each agent is modeled by a strict-feedback nonlinear system with unknown nonlinear dynamics and unknown external disturbances. The time-varying leader node only gives commands to a small portion of the followers. By using backstepping technique and Neural Networks method, adaptive distributed controllers for each follower node are constructed, which only require relative state information between themselves and their neighbors. The proposed controllers and adaptive laws guarantee that the tracking errors between all followers and the leader convergence to a small neighborhood of the origin. Moreover, by employing the maximum norm of the unknown Neural Network Weight vectors as the estimated parameter, the algorithm proposed in this paper contains only N(N represents the number of the followers) adaptive parameters that need to be updated online. The number of online learning parameters is independent of the number of the Neural Networks׳ nodes, which reduces the computation burden significantly. Finally, a numerical example demonstrates the effectiveness of the proposed approach.