SEARCH · Search NASA

Results for “Kernel methods”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

SourceResearch connectionFrom yearThrough year

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 577 records · Page 32

All sources NASA NTRS DOE OSTI

DOE OSTI2026

Enabling the Broader Use of MOOSE for Nuclear Energy and Other Simulation

This Final Scientific and Technical Report summarizes work performed under the Phase IIA SBIR project “Enabling the Broader Use of MOOSE for Nuclear Energy and Other Simulation” (DE-SC0020906) from August 2023 through August 2025. The objective of the Phase IIA effort was to mature and harden capabilities developed during Phase II, with the goal of enabling practical interoperability between Coreform’s isogeometric analysis (IGA) technologies and the Multiphysics Object-Oriented Simulation Environment (MOOSE), while improving robustness, performance, and scalability for complex, nuclear-relevant geometries. Over the course of Phase IIA, the project established and validated an extraction-based interoperability pathway between Coreform tools and MOOSE. A combined mesh and matrix format was defined collaboratively with MOOSE developers and integrated into the solver, enabling standard MOOSE workflows to operate on data exported from Coreform’s IGA and Flex Representation Method (FRM) pipelines. Early demonstrations validated architectural compatibility using linear solid mechanics problems, while later efforts focused on benchmark testing and external use. By the end of the project period, engineers at BWXT were able to independently set up and execute a simulation using the Coreform–MOOSE workflow and provide direct feedback that informed further refinement. In parallel, substantial effort was devoted to improving the robustness of trimmed U-spline construction for complex CAD geometries. A growing test suite of nuclear-relevant models was compiled through collaboration with multiple stakeholders and used to drive extensive bug fixing and reliability improvements. These efforts resulted in improved robustness and performance, including the addition of fallback capabilities that enhance reliability when the underlying commercial CAD kernel fails. Performance-oriented work progressed later in the project, with the development and demonstration of methods to decompose complex geometries into structured subregions and updated data representations to support more efficient solver processing. Additionally, extensive enhancements to threadsafe parallel data structures and trimming operations established a foundation for scalable processing of large assemblies. Collaboration with Sandia National Laboratories on the SGM geometric modeling kernel advanced to a functioning interface test case, positioning the workflow for future kernel integration. Overall, the Phase IIA effort successfully transitioned the project from architectural proof-of-concept to externally exercised, solver-integrated capability, while clarifying remaining technical challenges related to standardization, performance optimization, and kernel integration.

42 ENGINEERING↗

DOE OSTI2015

Acceleration of GPU-based Krylov solvers via data transfer reduction

Krylov subspace iterative solvers are often the method of choice when solving large sparse linear systems. At the same time, hardware accelerators such as graphics processing units continue to offer significant floating point performance gains for matrix and vector computations through easy-to-use libraries of computational kernels. However, as these libraries are usually composed of a well optimized but limited set of linear algebra operations, applications that use them often fail to reduce certain data communications, and hence fail to leverage the full potential of the accelerator. In this study, we target the acceleration of Krylov subspace iterative methods for graphics processing units, and in particular the Biconjugate Gradient Stabilized solver that significant improvement can be achieved by reformulating the method to reduce data-communications through application-specific kernels instead of using the generic BLAS kernels, e.g. as provided by NVIDIA’s cuBLAS library, and by designing a graphics processing unit specific sparse matrix-vector product kernel that is able to more efficiently use the graphics processing unit’s computing power. Furthermore, we derive a model estimating the performance improvement, and use experimental data to validate the expected runtime savings. Finally, considering that the derived implementation achieves significantly higher performance, we assert that similar optimizations addressing algorithm structure, as well as sparse matrix-vector, are crucial for the subsequent development of high-performance graphics processing units accelerated Krylov subspace iterative methods.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2011

Numerical treatment of the long-range Coulomb potential with Berggren bases

The Schroedinger equation incorporating the long-range Coulomb potential takes the form of a Fredholm equation whose kernel is singular on its diagonal when represented by a basis bearing a continuum of states, such as in a Fourier-Bessel transform. Several methods have been devised to tackle this difficulty, from simply removing the infinite range of the Coulomb potential with a screening or cut function to using discretizing schemes which take advantage of the integrable character of Coulomb kernel singularities. However, they have never been tested in the context of Berggren bases, which allow many-body nuclear wave functions to be expanded with halo or resonant properties within a shell-model framework. It is thus the object of this paper to test different discretization schemes of the Coulomb potential kernel in the framework of complex-energy nuclear physics. For that, the Berggren basis expansion of proton states pertaining to the sd shell arising in the A{approx}20 region, being typically resonant, will be effected. Apart from standard frameworks involving a cut function or analytical integration of singularities, a new method will be presented which replaces diagonal singularities by finite off-diagonal terms. It will be shown that this methodology surpasses in precision the two former techniques.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

DOE OSTI2015

Strain softening mechanism at meso scale during micro-compression in an ultrafine-grained pure copper

Strain softening behavior has been found at meso scale using micro-compression testing in an ultrafine-grained (UFG) pure copper by comparison with the typical strain hardening in conventional coarse-grained (CG) material. Microstructural observations show that grain size remains nearly the same including the fraction of high-angle grain boundaries during micro-compression in UFG pure copper. The Kernel average misorientation(KAM) distribution measured by electron backscatter diffraction (EBSD), as a statistical method, is applied to qualitatively evaluate dislocation density in the interior of the grains. It is suggested that the deformation mechanisms are dominated by grain boundary sliding and grain rotation accompanied by dislocation slip in UFG pure copper, which demonstrates that the strain softening behavior is primarily caused by dislocation annihilation during micro-compression.

36 MATERIALS SCIENCE↗

DOE OSTI2016

Thermal Density Functional Theory: Time-Dependent Linear Response and Approximate Functionals from the Fluctuation-Dissipation Theorem

We present that the van Leeuwen proof of linear-response time-dependent density functional theory (TDDFT) is generalized to thermal ensembles. This allows generalization to finite temperatures of the Gross-Kohn relation, the exchange-correlation kernel of TDDFT, and fluctuation dissipation theorem for DFT. Finally, this produces a natural method for generating new thermal exchange-correlation approximations.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

DOE OSTI2016

SU-F-T-672: A Novel Kernel-Based Dose Engine for KeV Photon Beams

Purpose: Mimicking state-of-the-art patient radiotherapy with high precision irradiators for small animals allows advanced dose-effect studies and radiobiological investigations. One example is the implementation of pre-clinical IMRT-like irradiations, which requires the development of inverse planning for keV photon beams. As a first step, we present a novel kernel-based dose calculation engine for keV x-rays with explicit consideration of energy and material dependencies. Methods: We follow a superposition-convolution approach adapted to keV x-rays, based on previously published work on micro-beam therapy. In small animal radiotherapy, we assume local energy deposition at the photon interaction point, since the electron ranges in tissue are of the same order of magnitude as the voxel size. This allows us to use photon-only kernel sets generated by MC simulations, which are pre-calculated for six energy windows and ten base materials. We validate our stand-alone dose engine against Geant4 MC simulations for various beam configurations in water, slab phantoms with bone and lung inserts, and on a mouse CT with (0.275mm)3 voxels. Results: We observe good agreement for all cases. For field sizes of 1mm{sup 2} to 1cm{sup 2} in water, the depth dose curves agree within 1% (mean), with the largest deviations in the first voxel (4%) and at depths>5cm (<2.5%). The out-of-field doses at 1cm depth agree within 8% (mean) for all but the smallest field size. In slab geometries, the mean agreement was within 3%, with maximum deviations of 8% at water-bone interfaces. The γ-index (1mm/1%) passing rate for a single-field mouse irradiation is 71%. Conclusion: The presented dose engine yields an accurate representation of keV-photon doses suitable for inverse treatment planning for IMRT. It has the potential to become a significantly faster yet sufficiently accurate alternative to full MC simulations. Further investigations will focus on energy sampling as well as calculation times. Research at ICR is also supported by Cancer Research UK under Programme C33589/A19727 and NHS funding to the NIHR Biomedical Research Centre at RMH and ICR. MFF is supported by Cancer Research UK under Programme C33589/A19908.

43 PARTICLE ACCELERATORS↗

DOE OSTI2015

Real time explosive hazard information sensing, processing, and communication for autonomous operation

Methods, computer readable media, and apparatuses provide robotic explosive hazard detection. A robot intelligence kernel (RIK) includes a dynamic autonomy structure with two or more autonomy levels between operator intervention and robot initiative A mine sensor and processing module (ESPM) operating separately from the RIK perceives environmental variables indicative of a mine using subsurface perceptors. The ESPM processes mine information to determine a likelihood of a presence of a mine. A robot can autonomously modify behavior responsive to an indication of a detected mine. The behavior is modified between detection of mines, detailed scanning and characterization of the mine, developing mine indication parameters, and resuming detection. Real time messages are passed between the RIK and the ESPM. A combination of ESPM bound messages and RIK bound messages cause the robot platform to switch between modes including a calibration mode, the mine detection mode, and the mine characterization mode.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2015

Real time explosive hazard information sensing, processing, and communication for autonomous operation

45 MILITARY TECHNOLOGY, WEAPONRY, AND NATIONAL DEF↗

DOE OSTI2010

GENUS STATISTICS USING THE DELAUNAY TESSELLATION FIELD ESTIMATION METHOD. I. TESTS WITH THE MILLENNIUM SIMULATION AND THE SDSS DR7

We study the topology of cosmic large-scale structure through the genus statistics, using galaxy catalogs generated from the Millennium Simulation and observational data from the latest Sloan Digital Sky Survey Data Release (SDSS DR7). We introduce a new method for constructing galaxy density fields and for measuring the genus statistics of its isodensity surfaces. It is based on a Delaunay tessellation field estimation (DTFE) technique that allows the definition of a piece-wise continuous density field and the exact computation of the topology of its polygonal isodensity contours, without introducing any free numerical parameter. Besides this new approach, we also employ the traditional approaches of smoothing the galaxy distribution with a Gaussian of fixed width, or by adaptively smoothing with a kernel that encloses a constant number of neighboring galaxies. Our results show that the Delaunay-based method extracts the largest amount of topological information. Unlike the traditional approach for genus statistics, it is able to discriminate between the different theoretical galaxy catalogs analyzed here, both in real space and in redshift space, even though they are based on the same underlying simulation model. In particular, the DTFE approach detects with high confidence a discrepancy of one of the semi-analytic models studied here compared with the SDSS data, while the other models are found to be consistent.

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2017

Monte Carlo Perturbation Theory Estimates of Sensitivities to System Dimensions

Here, Monte Carlo methods are developed using adjoint-based perturbation theory and the differential operator method to compute the sensitivities of the k-eigenvalue, linear functions of the flux (reaction rates), and bilinear functions of the forward and adjoint flux (kinetics parameters) to system dimensions for uniform expansions or contractions. The calculation of sensitivities to system dimensions requires computing scattering and fission sources at material interfaces using collisions occurring at the interface—which is a set of events with infinitesimal probability. Kernel density estimators are used to estimate the source at interfaces using collisions occurring near the interface. The methods for computing sensitivities of linear and bilinear ratios are derived using the differential operator method and adjoint-based perturbation theory and are shown to be equivalent to methods previously developed using a collision history–based approach. The methods for determining sensitivities to system dimensions are tested on a series of fast, intermediate, and thermal critical benchmarks as well as a pressurized water reactor benchmark problem with iterated fission probability used for adjoint-weighting. The estimators are shown to agree within 5% and 3σ of reference solutions obtained using direct perturbations with central differences for the majority of test problems.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DOE OSTI2013

DANCING IN THE DARK: NEW BROWN DWARF BINARIES FROM KERNEL PHASE INTERFEROMETRY

This paper revisits a sample of ultracool dwarfs in the solar neighborhood previously observed with the Hubble Space Telescope's NICMOS NIC1 instrument. We have applied a novel high angular resolution data analysis technique based on the extraction and fitting of kernel phases to archival data. This was found to deliver a dramatic improvement over earlier analysis methods, permitting a search for companions down to projected separations of {approx}1 AU on NIC1 snapshot images. We reveal five new close binary candidates and present revised astrometry on previously known binaries, all of which were recovered with the technique. The new candidate binaries have sufficiently close separation to determine dynamical masses in a short-term observing campaign. We also present four marginal detections of objects which may be very close binaries or high-contrast companions. Including only confident detections within 19 pc, we report a binary fraction of at least #Greek Lunate Epsilon Symbol#{sub b} = 17.2{sub -3.7}{sup +5.7}%. The results reported here provide new insights into the population of nearby ultracool binaries, while also offering an incisive case study of the benefits conferred by the kernel phase approach in the recovery of companions within a few resolution elements of the point-spread function core.

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2016

SU-F-SPS-06: Implementation of a Back-Projection Algorithm for 2D in Vivo Dosimetry with An EPID System

Purpose: To implement a back-projection algorithm for 2D dose reconstructions for in vivo dosimetry in radiation therapy using an Electronic Portal Imaging Device (EPID) based on amorphous silicon. Methods: An EPID system was used to calculate dose-response function, pixel sensitivity map, exponential scatter kernels and beam hardenig correction for the back-projection algorithm. All measurements were done with a 6 MV beam. A 2D dose reconstruction for an irradiated water phantom (30×30×30 cm{sup 3}) was done to verify the algorithm implementation. Gamma index evaluation between the 2D reconstructed dose and the calculated with a treatment planning system (TPS) was done. Results: A linear fit was found for the dose-response function. The pixel sensitivity map has a radial symmetry and was calculated with a profile of the pixel sensitivity variation. The parameters for the scatter kernels were determined only for a 6 MV beam. The primary dose was estimated applying the scatter kernel within EPID and scatter kernel within the patient. The beam hardening coefficient is σBH= 3.788×10{sup −4} cm{sup 2} and the effective linear attenuation coefficient is µAC= 0.06084 cm{sup −1}. The 95% of points evaluated had γ values not longer than the unity, with gamma criteria of ΔD = 3% and Δd = 3 mm, and within the 50% isodose surface. Conclusion: The use of EPID systems proved to be a fast tool for in vivo dosimetry, but the implementation is more complex that the elaborated for pre-treatment dose verification, therefore, a simplest method must be investigated. The accuracy of this method should be improved modifying the algorithm in order to compare lower isodose curves.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2017

Deploy Nalu/Kokkos algorithmic infrastructure with performance benchmarking.

The former Nalu interior heterogeneous algorithm design, which was originally designed to manage matrix assembly operations over all elemental topology types, has been modified to operate over homogeneous collections of mesh entities. This newly templated kernel design allows for removal of workset variable resize operations that were formerly required at each loop over a Sierra ToolKit (STK) bucket (nominally, 512 entities in size). Extensive usage of the Standard Template Library (STL) std::vector has been removed in favor of intrinsic Kokkos memory views. In this milestone effort, the transition to Kokkos as the underlying infrastructure to support performance and portability on many-core architectures has been deployed for key matrix algorithmic kernels. A unit-test driven design effort has developed a homogeneous entity algorithm that employs a team-based thread parallelism construct. The STK Single Instruction Multiple Data (SIMD) infrastructure is used to interleave data for improved vectorization. The collective algorithm design, which allows for concurrent threading and SIMD management, has been deployed for the core low-Mach element- based algorithm. Several tests to ascertain SIMD performance on Intel KNL and Haswell architectures have been carried out. The performance test matrix includes evaluation of both low- and higher-order methods. The higher-order low-Mach methodology builds on polynomial promotion of the core low-order control volume nite element method (CVFEM). Performance testing of the Kokkos-view/SIMD design indicates low-order matrix assembly kernel speed-up ranging between two and four times depending on mesh loading and node count. Better speedups are observed for higher-order meshes (currently only P=2 has been tested) especially on KNL. The increased workload per element on higher-order meshes bene ts from the wide SIMD width on KNL machines. Combining multiple threads with SIMD on KNL achieves a 4.6x speedup over the baseline, with assembly timings faster than that observed on Haswell architecture. The computational workload of higher-order meshes, therefore, seems ideally suited for the many-core architecture and justi es further exploration of higher-order on NGP platforms. A Trilinos/Tpetra-based multi-threaded GMRES preconditioned by symmetric Gauss Seidel (SGS) represents the core solver infrastructure for the low-Mach advection/diffusion implicit solves. The threaded solver stack has been tested on small problems on NREL's Peregrine system using the newly developed and deployed Kokkos-view/SIMD kernels. fforts are underway to deploy the Tpetra-based solver stack on NERSC Cori system to benchmark its performance at scale on KNL machines.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2014

TH-E-BRE-07: Development of Dose Calculation Error Predictors for a Widely Implemented Clinical Algorithm

Purpose: Several shortcomings of the current implementation of the analytic anisotropic algorithm (AAA) may lead to dose calculation errors in highly modulated treatments delivered to highly heterogeneous geometries. Here we introduce a set of dosimetric error predictors that can be applied to a clinical treatment plan and patient geometry in order to identify high risk plans. Once a problematic plan is identified, the treatment can be recalculated with more accurate algorithm in order to better assess its viability. Methods: Here we focus on three distinct sources dosimetric error in the AAA algorithm. First, due to a combination of discrepancies in smallfield beam modeling as well as volume averaging effects, dose calculated through small MLC apertures can be underestimated, while that behind small MLC blocks can overestimated. Second, due the rectilinear scaling of the Monte Carlo generated pencil beam kernel, energy is not properly transported through heterogeneities near, but not impeding, the central axis of the beamlet. And third, AAA overestimates dose in regions very low density (< 0.2 g/cm{sup 3}). We have developed an algorithm to detect the location and magnitude of each scenario within the patient geometry, namely the field-size index (FSI), the heterogeneous scatter index (HSI), and the lowdensity index (LDI) respectively. Results: Error indices successfully identify deviations between AAA and Monte Carlo dose distributions in simple phantom geometries. Algorithms are currently implemented in the MATLAB computing environment and are able to run on a typical RapidArc head and neck geometry in less than an hour. Conclusion: Because these error indices successfully identify each type of error in contrived cases, with sufficient benchmarking, this method can be developed into a clinical tool that may be able to help estimate AAA dose calculation errors and when it might be advisable to use Monte Carlo calculations.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2024

Advances in statistical methods for cancer surveillance research: an age-period-cohort perspective

Background: Analysis of Lexis diagrams (population-based cancer incidence and mortality rates indexed by age group and calendar period) requires specialized statistical methods. However, existing methods have limitations that can now be overcome using new approaches. Methods: We assembled a “toolbox” of novel methods to identify trends and patterns by age group, calendar period, and birth cohort. We evaluated operating characteristics across 152 cancer incidence Lexis diagrams compiled from United States (US) Surveillance, Epidemiology and End Results Program data for 21 leading cancers in men and women in four race and ethnicity groups (the “cancer incidence panel”). Results: Nonparametric singular values adaptive kernel filtration (SIFT) decreased the estimated root mean squared error by 90% across the cancer incidence panel. A novel method for semi-parametric age-period-cohort analysis (SAGE) provided optimally smoothed estimates of age-period-cohort (APC) estimable functions and stabilized estimates of lack-of-fit (LOF). SAGE identified statistically significant birth cohort effects across the entire cancer panel; LOF had little impact. As illustrated for colon cancer, newly developed methods for comparative age-period-cohort analysis can elucidate cancer heterogeneity that would otherwise be difficult or impossible to discern using standard methods. Conclusions: Cancer surveillance researchers can now identify fine-scale temporal signals with unprecedented accuracy and elucidate cancer heterogeneity with unprecedented specificity. Birth cohort effects are ubiquitous modulators of cancer incidence in the US. The novel methods described here can advance cancer surveillance research.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2013

Motion as a perturbation: Measurement-guided dose estimates to moving patient voxels during modulated arc deliveries

Purpose: To present a framework for measurement-guided VMAT dose reconstruction to moving patient voxels from a known motion kernel and the static phantom data, and to validate this perturbation-based approach with the proof-of-principle experiments. Methods: As described previously, the VMAT 3D dose to a static patient can be estimated by applying a phantom measurement-guided perturbation to the treatment planning system (TPS)-calculated dose grid. The fraction dose to any voxel in the presence of motion, assuming the motion kernel is known, can be derived in a similar fashion by applying a measurement-guided motion perturbation. The dose to the diodes in a helical phantom is recorded at 50 ms intervals and is transformed into a series of time-resolved high-density volumetric dose grids. A moving voxel is propagated through this 4D dose space and the fraction dose to that voxel in the phantom is accumulated. The ratio of this motion-perturbed, reconstructed dose to the TPS dose in the phantom serves as a perturbation factor, applied to the TPS fraction dose to the similarly situated voxel in the patient. This approach was validated by the ion chamber and film measurements on four phantoms of different shape and structure: homogeneous and inhomogeneous cylinders, a homogeneous cube, and an anthropomorphic thoracic phantom. A 2D motion stage was used to simulate the motion. The stage position was synchronized with the beam start time with the respiratory gating simulator. The motion patterns were designed such that the motion speed was in the upper range of the expected tumor motion (1-1.4 cm/s) and the range exceeded the normally observed limits (up to 5.7 cm). The conformal arc plans for X or Y motion (in the IEC 61217 coordinate system) consisted of manually created narrow (3 cm) rectangular strips moving in-phase (tracking) or phase-shifted by 90 Degree-Sign (crossing) with respect to the phantom motion. The XY motion was tested with the computer-derived VMAT MLC sequences. For all phantoms and plans, time-resolved (10 Hz) ion chamber dose was collected. In addition, coronal (XY) films were exposed in the cube phantom to a VMAT beam with two different starting phases, and compared to the reconstructed motion-perturbed dose planes. Results: For the X or Y motions with the moving strip and geometrical phantoms, the maximum difference between perturbation-reconstructed and ion chamber doses did not exceed 1.9%, and the average for any motion pattern/starting phase did not exceed 1.3%. For the VMAT plans on the cubic and thoracic phantoms, one point exhibited a 3.5% error, while the remaining five were all within 1.1%. Across all the measurements (N = 22), the average disagreement was 0.5 {+-} 1.3% (1 SD). The films exhibited {gamma}(3%/3 mm) passing rates {>=}90%. Conclusions: The dose to an arbitrary moving voxel in a patient can be estimated with acceptable accuracy for a VMAT delivery, by performing a single QA measurement with a cylindrical phantom and applying two consecutive perturbations to the TPS-calculated patient dose. The first one accounts for the differences between the planned and delivered static doses, while the second one corrects for the motion.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2017

Universal Method of Modeling Linear Stationary Physical Fields

The aim of this paper is to develop a method, described earlier, of solving quantum-mechanical problems into a universal numerical method of modeling fields of various physical nature. This method is based on reducing the initial equation of mathematical physics describing a given physical field to a simpler inhomogeneous equation with a known fundamental solution. This equation is then transformed into an inhomogeneous integral equation with a kernel expressed in terms of the known fundamental solution. The obtained integral equation with boundary conditions is solved numerically. To confirm the efficiency of the proposed numerical method, a two-dimensional and a three-dimensional boundary value problem with known solutions have been solved. Another important illustration of the efficiency of the proposed method is the solution of quantum-mechanical problems for one-dimensional and two-dimensional quantum oscillators. It is shown that the considered method allows one to find the energy eigenvalues and the eigenfunctions with acceptable accuracy.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

DOE OSTI2019

Scalable and efficient learning from crowds with Gaussian processes

Over the last few years, multiply-annotated data has become a very popular source of information. Online platforms such as Amazon Mechanical Turk have revolutionized the labelling process needed for any classification task, sharing the effort between a number of annotators (instead of the classical single expert). This crowdsourcing approach has introduced new challenging problems, such as handling disagreements on the annotated samples or combining the unknown expertise of the annotators. Probabilistic methods, such as Gaussian Processes (GP), have proven successful to model this new crowdsourcing scenario. However, GPs do not scale up well with the training set size, which makes them prohibitive for medium-to-large datasets (beyond 10K training instances). This constitutes a serious limitation for current real-world applications. In this work, we introduce two scalable and efficient GP-based crowdsourcing methods that allow for processing previously-prohibitive datasets. The first one is an efficient and fast approximation to GP with squared exponential (SE) kernel. The second allows for learning a more flexible kernel at the expense of a heavier training (but still scalable to large datasets). Since the latter is not a GP-SE approximation, it can be also considered as a whole new scalable and efficient crowdsourcing method, useful for any dataset size. Both methods use Fourier features and variational inference, can predict the class of new samples, and estimate the expertise of the involved annotators. A complete experimentation compares them with state-of-the-art probabilistic approaches in synthetic and real crowdsourcing datasets of different sizes. Finally, they stand out as the best performing approach for large scale problems. Moreover, the second method is competitive with the current state-of-the-art for small datasets.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗