The Experts below are selected from a list of 16101 Experts worldwide ranked by ideXlab platform
Geoffrey J. Mclachlan - One of the best experts on this subject based on the ideXlab platform.
-
a universal approximation theorem for mixture of experts models
arXiv: Machine Learning, 2016Co-Authors: Hien D Nguyen, Luke R Lloydjones, Geoffrey J. MclachlanAbstract:The mixture of experts (MoE) model is a popular neural network architecture for nonlinear regression and classification. The class of MoE mean Functions is known to be uniformly convergent to any unknown Target Function, assuming that the Target Function is from Sobolev space that is sufficiently differentiable and that the domain of estimation is a compact unit hypercube. We provide an alternative result, which shows that the class of MoE mean Functions is dense in the class of all continuous Functions over arbitrary compact domains of estimation. Our result can be viewed as a universal approximation theorem for MoE models.
-
A Universal Approximation Theorem for Mixture-of-Experts Models
Neural Computation, 2016Co-Authors: Hien D Nguyen, Luke R. Lloyd-jones, Geoffrey J. MclachlanAbstract:The mixture-of-experts (MoE) model is a popular neural network architecture for nonlinear regression and classification. The class of MoE mean Functions is known to be uniformly convergent to any unknown Target Function, assuming that the Target Function is from a Sobolev space that is sufficiently differentiable and that the domain of estimation is a compact unit hypercube. We provide an alternative result, which shows that the class of MoE mean Functions is dense in the class of all continuous Functions over arbitrary compact domains of estimation. Our result can be viewed as a universal approximation theorem for MoE models. The theorem we present allows MoE users to be confident in applying such models for estimation when data arise from nonlinear and nondifferentiable generative processes.
Hien D Nguyen - One of the best experts on this subject based on the ideXlab platform.
-
a universal approximation theorem for mixture of experts models
arXiv: Machine Learning, 2016Co-Authors: Hien D Nguyen, Luke R Lloydjones, Geoffrey J. MclachlanAbstract:The mixture of experts (MoE) model is a popular neural network architecture for nonlinear regression and classification. The class of MoE mean Functions is known to be uniformly convergent to any unknown Target Function, assuming that the Target Function is from Sobolev space that is sufficiently differentiable and that the domain of estimation is a compact unit hypercube. We provide an alternative result, which shows that the class of MoE mean Functions is dense in the class of all continuous Functions over arbitrary compact domains of estimation. Our result can be viewed as a universal approximation theorem for MoE models.
-
A Universal Approximation Theorem for Mixture-of-Experts Models
Neural Computation, 2016Co-Authors: Hien D Nguyen, Luke R. Lloyd-jones, Geoffrey J. MclachlanAbstract:The mixture-of-experts (MoE) model is a popular neural network architecture for nonlinear regression and classification. The class of MoE mean Functions is known to be uniformly convergent to any unknown Target Function, assuming that the Target Function is from a Sobolev space that is sufficiently differentiable and that the domain of estimation is a compact unit hypercube. We provide an alternative result, which shows that the class of MoE mean Functions is dense in the class of all continuous Functions over arbitrary compact domains of estimation. Our result can be viewed as a universal approximation theorem for MoE models. The theorem we present allows MoE users to be confident in applying such models for estimation when data arise from nonlinear and nondifferentiable generative processes.
Paul D Adams - One of the best experts on this subject based on the ideXlab platform.
-
use of knowledge based restraints in phenix refine to improve macromolecular refinement at low resolution
Acta Crystallographica Section D-biological Crystallography, 2012Co-Authors: Jeffrey J Headd, Nathaniel Echols, Pavel V Afonine, Ralf W Grossekunstleve, Vincent B Chen, Nigel W Moriarty, David C Richardson, Jane S Richardson, Paul D AdamsAbstract:Traditional methods for macromolecular refinement often have limited success at low resolution (3.0–3.5 A or worse), producing models that score poorly on crystallographic and geometric validation criteria. To improve low-resolution refinement, knowledge from macromolecular chemistry and homology was used to add three new coordinate-restraint Functions to the refinement program phenix.refine. Firstly, a `reference-model' method uses an identical or homologous higher resolution model to add restraints on torsion angles to the geometric Target Function. Secondly, automatic restraints for common secondary-structure elements in proteins and nucleic acids were implemented that can help to preserve the secondary-structure geometry, which is often distorted at low resolution. Lastly, we have implemented Ramachandran-based restraints on the backbone torsion angles. In this method, a φ,ψ term is added to the geometric Target Function to minimize a modified Ramachandran landscape that smoothly combines favorable peaks identified from nonredundant high-quality data with unfavorable peaks calculated using a clash-based pseudo-energy Function. All three methods show improved MolProbity validation statistics, typically complemented by a lowered Rfree and a decreased gap between Rwork and Rfree.
Emilio Marengo - One of the best experts on this subject based on the ideXlab platform.
-
iterative optimization of an esi it mass spectrometer using regular simplex and a multivariate Target Function representing the s n ratio
Journal of the American Society for Mass Spectrometry, 2011Co-Authors: Elisa Robotti, F Gosetti, E Mazzucco, Davide Zampieri, Emilio MarengoAbstract:Standard automatic tuning for mass spectrometry usually exploits a one-variable-at-a-time approach. This method suffers from important drawbacks: the Target Function selected for optimization improves the signal of a single channel or a pool of channels without considering noise; the interactions between the parameters are not evaluated. The optimization of the experimental settings of an ESI IT mass spectrometer is carried out here by a multivariate procedure exploiting a Target Function representing the S/N ratio calculated by principal component analysis and a regular simplex optimization algorithm. A preliminary feasibility study was performed since the Target Function must be sensitive to the changes in the experimental conditions applied during the iterative tuning and be free from drifts. The feasibility study was carried out to evaluate: the presence of memory effects; the size of the variations in the S/N ratio; the number of scans needed to generate a reliable S/N ratio; the concentration of the multi-standard mixture to use during tuning; the experimental duration required to achieve S/N stability when the experimental settings are modified. The feasibility study led to the identification of the best protocol to accomplish the tuning, while simplex optimization allowed the S/N ratio to be improved by about 70% with respect to the default conditions suggested by the manufacturer.
-
Evaluation of Signal and Noise and Identification of a Suitable Target Function in the Tuning of an ESI Ion Trap Mass Spectrometer by Multivariate Pattern Recognition Tools
Journal of the American Society for Mass Spectrometry, 2009Co-Authors: Emilio Marengo, Elisa Robotti, Fabio Gosetti, Orfeo Zerbinati, Maria Carla GennaroAbstract:When mass spectrometry is not combined to separation techniques, the evaluation of signal and noise in a complex mass spectrum is not trivial. The tuning of the spectrometer based only on the increase of the signal of a selected number of m/z values does not ensure the achievement of the best experimental conditions: signal could improve and noise could increase as well. The scope of this work is the development of a Function separating signal and noise (for evaluating the S/N) from complex mass spectra for potential use as Target Function for the automatic tuning of the instrument. Two different methods were applied: the first is based on the separation of a pool of m/z values attributable to the signal from the m/z values due to the noise, while the second is based on the application of principal component analysis to separate the signal (present in the significant components) from the noise (present in the residuals). The comparison of the two methods was carried out by the evaluation of the stability of the signal and the Target Functions obtained, and the evaluation of the variation of the Target Functions as a Function of concentration.
Wolfgang Boehmer - One of the best experts on this subject based on the ideXlab platform.
-
CIT - Toward a Target Function of an Information Security Management System
2010 10th IEEE International Conference on Computer and Information Technology, 2010Co-Authors: Wolfgang BoehmerAbstract:The limits of traditional (static) policies are well-known in many areas of computer science and information security, and are extensively discussed in the literature. Although some flexibility has been achieved with the introduction of dynamic policies, these efforts have only addressed a fraction of the requirements necessary to secure today's enterprises. Currently, no feedback mechanisms are in place to evaluate the effectiveness or economic impacts of static or dynamic policy implementation. Here, we address the requirement for feedback and present a policy for the next generation. This is a policy that includes a dynamic feedback response to the effectiveness of changes. The structure of this new type of policy, called a ``management system'', is borrowed from discrete event system (DES) theory and Functions as a control loop. A management system consists of four elements (control system, sensor, controller, and actuator) that are involved in a control law. Two types of management system can be defined. A simple management system (1$^\textrm{st}$ order management system) responds to and regulates only perturbations. An advanced management system (2$^\textrm{nd}$ order management system) has an overarching Target Function that influences the controller. This Target Function is usually economically oriented. Finally, we compare our new type of policy with two management systems that follows the Plan-Do-Check-Act (PDCA cycle) model. We investigate the two PDCA cycle standards ISO/IEC 27001 (Information Security Management System, ISMS) and BS 25999 (Business Continuity Management System, BCMS). We also show that the new type of policy can be applied to management systems based on a PDCA cycle.
-
Toward a Target Function of an Information Security Management System
2010 10th IEEE International Conference on Computer and Information Technology, 2010Co-Authors: Wolfgang BoehmerAbstract:The limits of traditional (static) policies are wellknown in many areas of computer science and information security, and are extensively discussed in the literature. Although some flexibility has been achieved with the introduction of dynamic policies, these efforts have only addressed a fraction of the requirements necessary to secure today's enterprises. Currently, no feedback mechanisms are in place to evaluate the effectiveness or economic impacts of static or dynamic policy implementation. Here, we address the requirement for feedback and present a policy for the next generation. This is a policy that includes a dynamic feedback response to the effectiveness of changes. The structure of this new type of policy, called a "management system", is borrowed from discrete event system (DES) theory and Functions as a control loop. A management system consists of four elements (control system, sensor, controller, and actuator) that are involved in a control law. Two types of management system can be defined. A simple management system (1st order management system) responds to and regulates only perturbations. An advanced management system (2nd order management system) has an overarching Target Function that influences the controller. This Target Function is usually economically oriented. Finally, we compare our new type of policy with two management systems that follows the Plan-Do-Check-Act (PDCA cycle) model. We investigate the two PDCA cycle standards ISO/IEC 27001 (Information Security Management System, ISMS) and BS 25999 (Business Continuity Management System, BCMS). We also show that the new type of policy can be applied to management systems based on a PDCA cycle.