Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “algorithm selection”

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.

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

At least 1,297 records · Page 72

Deep Learning Based Frequency Stability Assessment in Power Grid with High Renewables

Frequency stability assessment is one critical aspect of power system security assessment. Traditional N-1 screening method is based on the simulations of a few typical daily and seasonal operation scenarios. However, the increasing integration of inverter-based renewables and the retirement of conventional synchronous generators result in decreasing system inertia and growing complexity of system operating conditions. Selecting a few typical operation scenarios cannot cover all operating conditions, and the time-domain simulation of all operation conditions requires tremendous time. This paper proposes a more efficient frequency stability assessment method based on deep learning. The affinity propagation clustering algorithm is used to divide the dataset into different clusters, so the selected dataset for training can cover the diversified operating conditions as much as possible. Also, feature normalization is applied to both the training dataset and testing dataset in order to remove any unnecessary bias. Especially, trained model based on full dataset normalization has bounded error in the prediction. The case study on the reduced 240-bus WECC system demonstrates that the proposed method can predict accurate frequency nadir with limited training dataset. The deep learning model using the revised feature normalization can predict more accurate frequency nadir than that using the traditional feature normalization and has very small maximum prediction error.

affinity propagation↗

Training SVMs without offset

We develop, analyze, and test a training algorithm for support vector machine cla.'>sifiers without offset. Key features of this algorithm are a new stopping criterion and a set of working set selection strategies that, although inexpensive, do not lead to substantially more iterations than the optimal working set selection strategy. For these working set strategies, we establish convergence rates that coincide with the best known rates for SYMs with offset. We further conduct various experiments that investigate both the run time behavior and the performed iterations of the new training algorithm. It turns out, that the new algorithm needs less iterations and run-time than standard training algorithms for SYMs with offset.

97 MATHEMATICS AND COMPUTING↗

A multi criteria analog model for assessing the vulnerability of rural catchments to road spills of hazardous substances

Road spills of hazardous substances are common in developing countries due to increasing industrialization and traffic accidents, and represent a serious threat to soils and water in catchments. There is abundant literature on equations describing the wash-off of pollutants from roads during a storm event and there are a number of watershed models incorporating those equations in storm water quality algorithms that route runoff and pollution yields through a drainage system towards the catchment outlet. However, methods describing catchment vulnerability to contamination by road spills based solely on biophysical parameters are scarce. These methods could be particularly attractive to managers because they can operate with a limited amount of easily collectable data, while still being able to provide important insights on the areas more prone to contamination within the studied watershed. The purpose of this paper was then to contribute with a new vulnerability model. To accomplish the goal, a selection of medium properties appearing in wash-off equations and routing algorithms were assembled and processed in a parametric framework based on multi criteria analysis to define the watershed vulnerability. However, parameters had to be adapted because wash-off equations and water quality models have been developed to operate primarily in the urban environment while the vulnerability model is meant to run in rural watersheds. The selected parameters were hillside slope, ground roughness (depending on land use), soil permeability (depending on soil type), distance to water courses and stream density. The vulnerability model is a spatially distributed algorithm that was prepared to run under the IDRISI Selva software, a GIS platform capable of handling spatial and alphanumeric data and execute the necessary terrain model, hydrographic and thematic analyses. For illustrative purposes, the vulnerability model was applied to the legally protected Environmental Protection Area (APA), located in the Uberaba region, state of Minas Gerais, Brazil. In this region, the risk of accidents causing chemical spills is preoccupying because large quantities of dangerous materials are transported in two important distribution highways while the APA is fundamental for the protection of water resources, the riverine ecosystems and remnants of native vegetation. In some tested scenarios, model results show 60% of vulnerable areas within the studied area. The most sensitive parameter to vulnerability is soil type. To prevent soils from contamination, specific measures were proposed involving minimization of land use conflicts that would presumably raise the soil's organic matter and in the sequel restore the soil's structural functions. Additionally, the present study proposed the preservation and reinforcement of riparian forests as one measure to protect the quality of surface water. - Highlights: • A multi criteria analog model was developed to assess rural catchment vulnerability along roads. • Model parameters were defined by analogy with urban wash-off equations and routing algorithms. • The model mixes up various biophysical and socio-economic parameters. • Model application was based on a scenario analysis. • The study is focused on the Environmental Protection Area of Uberaba River, Brazil.

54 ENVIRONMENTAL SCIENCES↗

Application of Reconfigurable Computing Technology to Multi-KiloHertz Micro-Laser Altimeter (MMLA) Data Processing

The Multi-KiloHertz Micro-Laser Altimeter (MMLA) is an aircraft based instrument developed by NASA Goddard Space Flight Center with several potential spaceflight applications. This presentation describes how reconfigurable computing technology was employed to perform MMLA signal extraction in real-time under realistic operating constraints. The MMLA is a "single-photon-counting" airborne laser altimeter that is used to measure land surface features such as topography and vegetation canopy height. This instrument has to date flown a number of times aboard the NASA P3 aircraft acquiring data at a number of sites in the Mid-Atlantic region. This instrument pulses a relatively low-powered laser at a very high rate (10 kHz) and then measures the time-of-flight of discrete returns from the target surface. The instrument then bins these measurements into a two-dimensional array (vertical height vs. horizontal ground track) and selects the most likely signal path through the array. Return data that does not correspond to the selected signal path are classified as noise returns and are then discarded. The MMLA signal extraction algorithm is very compute intensive in that a score must be computed for every possible path through the two dimensional array in order to select the most likely signal path. Given a typical array size with 50 x 6, up to 33 arrays must be processed per second. And for each of these arrays, roughly 12,000 individual paths must be scored. Furthermore, the number of paths increases exponentially with the horizontal size of the array, and linearly with the vertical size. Yet, increasing the horizontal and vertical sizes of the array offer science advantages such as improved range, resolution, and noise rejection. Due to the volume of return data and the compute intensive signal extraction algorithm, the existing PC-based MMLA data system has been unable to perform signal extraction in real-time unless the array is limited in size to one column, This limits the ability of the MMLA to operate in environments with sparse signal returns and a high number of noise return. However, under an IR&D project, an FPGA-based, reconfigurable computing data system has been developed that has been demonstrated to perform real-time signal extraction under realistic operating constraints. This reconfigurable data system is based on the commercially available Firebird Board from Annapolis Microsystems. This PCI board consists of a Xilinx Virtex 2000E FPGA along with 36 MB of SRAM arranged in five separately addressable banks. This board is housed in a rackmount PC with dual 850MHz Pentium processors running the Windows 2000 operating system. This data system performs all signal extraction in hardware on the Firebird, but also runs the existing "software based" signal extraction in tandem for comparison purposes. Using a relatively small amount of the Virtex XCV2000E resources, the reconfigurable data system has demonstrated to improve performance improvement over the existing software based data system by an order of magnitude. Performance could be further improved by employing parallelism. Ground testing and a preliminary engineering test flight aboard the NASA P3 has been performed, during which the reconfigurable data system has been demonstrated to match the results of the existing data system.

Powell, Wesley↗

Selection function of clusters in Dark Energy Survey year 3 data from cross-matching with South Pole Telescope detections

Context. Galaxy clusters selected based on overdensities of galaxies in photometric surveys provide the largest cluster samples. However, modeling the selection function of such samples is complicated by noncluster members projected along the line of sight (projection effects) and the potential detection of unvirialized objects (contamination). Aims. We empirically constrained the magnitude of these effects by cross-matching galaxy clusters selected in the Dark Energy Survey data with the redMaPPer algorithm with significant detections in three South Pole Telescope surveys (SZ, pol-ECS, pol-500d). Methods. For matched clusters, we augmented the redMaPPer catalog with the SPT detection significance. For unmatched objects we used the SPT detection threshold as an upper limit on the SZe signature. Using a Bayesian population model applied to the collected multiwavelength data, we explored various physically motivated models to describe the relationship between observed richness and halo mass. Results. Our analysis reveals a clear preference for models with an additional skewed scatter component associated with projection effects over a purely log-normal scatter model. We rule out significant contamination by unvirialized objects at the high-richness end of the sample. While dedicated simulations offer a well-fitting calibration of projection effects, our findings suggest the presence of redshift-dependent trends that these simulations may not have captured. Our findings highlight that modeling the selection function of optically detected clusters remains a complicated challenge that requires a combination of simulation and data-driven approaches.

79 ASTRONOMY AND ASTROPHYSICS↗

Accelerating iterative ptychography with an integrated neural network

Electron ptychography is a powerful and versatile tool for high-resolution and dose-efficient imaging. Iterative reconstruction algorithms are powerful but also computationally expensive due to their relative complexity and the many hyperparameters that must be optimised. Gradient descent-based iterative ptychography is a popular method, but it may converge slowly when reconstructing low spatial frequencies. Here, in this work, we present a method for accelerating a gradient descent-based iterative reconstruction algorithm by training a neural network (NN) that is applied in the reconstruction loop. The NN works in Fourier space and selectively boosts low spatial frequencies, thus enabling faster convergence in a manner similar to accelerated gradient descent algorithms. We discuss the difficulties that arise when incorporating a NN into an iterative reconstruction algorithm and show how they can be overcome with iterative training. We apply our method to simulated and experimental data of gold nanoparticles on amorphous carbon and show that we can significantly speed up ptychographic reconstruction of the nanoparticles.

4DSTEM↗

ACTIVE LEARNING TO OVERCOME SAMPLE SELECTION BIAS: APPLICATION TO PHOTOMETRIC VARIABLE STAR CLASSIFICATION

Despite the great promise of machine-learning algorithms to classify and predict astrophysical parameters for the vast numbers of astrophysical sources and transients observed in large-scale surveys, the peculiarities of the training data often manifest as strongly biased predictions on the data of interest. Typically, training sets are derived from historical surveys of brighter, more nearby objects than those from more extensive, deeper surveys (testing data). This sample selection bias can cause catastrophic errors in predictions on the testing data because (1) standard assumptions for machine-learned model selection procedures break down and (2) dense regions of testing space might be completely devoid of training data. We explore possible remedies to sample selection bias, including importance weighting, co-training, and active learning (AL). We argue that AL-where the data whose inclusion in the training set would most improve predictions on the testing set are queried for manual follow-up-is an effective approach and is appropriate for many astronomical applications. For a variable star classification problem on a well-studied set of stars from Hipparcos and Optical Gravitational Lensing Experiment, AL is the optimal method in terms of error rate on the testing data, beating the off-the-shelf classifier by 3.4% and the other proposed methods by at least 3.0%. To aid with manual labeling of variable stars, we developed a Web interface which allows for easy light curve visualization and querying of external databases. Finally, we apply AL to classify variable stars in the All Sky Automated Survey, finding dramatic improvement in our agreement with the ASAS Catalog of Variable Stars, from 65.5% to 79.5%, and a significant increase in the classifier's average confidence for the testing set, from 14.6% to 42.9%, after a few AL iterations.

79 ASTRONOMY AND ASTROPHYSICS↗

Performance Demonstration of an Occupancy Sensor-enabled Integrated Solution for Commercial Buildings

Traditionally, a single-loop fixed-gain controller is applied to supply fan (SF) and cooling coil (CC) valve controls while a fixed-damper position control is applied to outdoor air (OA) damper control at air handling units (AHUs) in commercial buildings. With the increasing application of occupancy sensors, the information generated by occupancy sensors is applied to not only reduce the electricity loads from lighting and controllable plug loads, but also reset OA intake and minimum supply airflow setpoints. Meanwhile, these intermittent operation actions greatly elevate the dynamics of AHU systems, which may introduce unstable SF and CC valve operations and inaccurate OA flow control at AHUs and consequently degrade maximum energy efficiency gains. With virtual fan and valve flow meter technologies, two advanced controls, including cascade control and gain scheduling control, can be implemented on both the SF and CC valve, and an advanced control using a virtual OA flow meter can be implemented on OA damper integrated with occupancy sensors. The goal of this project is to demonstrate the savings, cost, and performance of an integrated solution that integrates the three advanced HVAC controls with occupancy sensors to allow accurate and stable AHU operations in real buildings. The project objectives are to: 1) develop and validate an advance SF control algorithm; 2) develop and validate an advanced CC valve control algorithm; 3) validate an algorithm to implement a virtual OA flow meter; and 4) demonstrate the savings, cost, and performance of the integrated solution in real buildings. The technical approaches are to: 1) select a test system at the University of Oklahoma; 2) develop and implement the algorithms of advanced SF and CC valve controls and validate the performance; 3) develop and implement the advanced OA control using a virtual OA flow meter and validate the performance; 4) demonstrate the savings, cost, and performance of the proposed integrated solution with and without three advanced HVAC controls; and 5) disseminate the project results through publications and presentations. For the SF control, both the gain scheduling and cascade controls can improve the fan energy performance by reducing the fan power during the transient period and the fan control performance at lower speeds by reducing fan speed variation. Moreover, the gain scheduling control provides a simple and low-cost solution and is recommended. The fan power savings can reach 30% in a transient period. For the CC valve control, the gain scheduling control can considerably reduce the supply air temperature oscillation range and frequency under both higher and lower load conditions and the control valve response is much more stable. As a result, the gain scheduling control is recommended. The projected pump energy consumption can be reduced by 68.5%. With the developed virtual OA flow meter, the OA can be accurately controlled at its setpoint, which is determined based on the actual number of occupants in the building provided by occupancy sensors. The RMSE of the proposed OA control is 15.9 L/s. The energy data shows that the fan power and CC cooling energy were significantly reduced. On the other hand, the energy savings majorly results from the occupancy sensors and the energy savings by the advanced HVAC controls is minimal because that the controllers in the test AHU were tuned with very slow response. An annual technical savings potential is estimated as 0.5 quads in the commercial sector.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Algorithm Determines Wind Speed and Direction from Venturi-Sensor Data

An algorithm computes the velocity of wind from the readings of an instrument like the one described in another Tech Brief. To recapitulate: The sensor has no moving parts and is a compact, rugged means of measuring wind vectors having magnitudes of as much as 300 mph (134 m/s). The sensor includes a Venturi gap bounded by a curved upper and a curved lower surface that are axisymmetric with respect to a vertical axis and mirror-symmetric with respect to a horizontal midplane. One of the curved surfaces is instrumented with multiple ports for measuring dynamic pressures. The sensor also incorporates auxiliary sensors for measuring temperature, relative humidity, and static atmospheric pressure. The design and operation of the sensor are based on the concepts of (1) using Bernoulli's equation (which expresses the relationship among variations of speed, density, and pressure along a streamline) to calculate the speed of the wind from differences among the pressure readings at the various ports; and (2) calculating the direction of the wind from the angular positions of ports selected according to comparisons among their pressure readings. The present algorithm performs these calculations.

Zysko, Jan A.↗

Sensitivity of Global Sea-Air CO2 Flux to Gas Transfer Algorithms, Climatological Wind Speeds, and Variability of Sea Surface Temperature and Salinity

Sensitivity analyses of sea-air CO2 flux to gas transfer algorithms, climatological wind speeds, sea surface temperatures (SST) and salinity (SSS) were conducted for the global oceans and selected regional domains. Large uncertainties in the global sea-air flux estimates are identified due to different gas transfer algorithms, global climatological wind speeds, and seasonal SST and SSS data. The global sea-air flux ranges from -0.57 to -2.27 Gt/yr, depending on the combination of gas transfer algorithms and global climatological wind speeds used. Different combinations of SST and SSS global fields resulted in changes as large as 35% on the oceans global sea-air flux. An error as small as plus or minus 0.2 in SSS translates into a plus or minus 43% deviation on the mean global CO2 flux. This result emphasizes the need for highly accurate satellite SSS observations for the development of remote sensing sea-air flux algorithms.

McClain, Charles R.↗

Highlights from 10 Years of NASA/KNMI/FMI Collaboration on UV Remote Sensing from Space

The first joint meeting between NASA, KNMI and FMI scientists was held on 13 & 14 June, 2000, almost exactly 10 years ago. NASA had recently selected 14 US scientists to work on instrument calibration, science algorithms, and validation activities related to the Ozone Monitoring Instrument (OMI) that we being built by collaboration between the Netherlands and Finland for flight on NASA's EOS Aura satellite. The progress on this project had been remarkable for a space based instrument. Only two years before this meeting my colleague Ernest Hilsenrath and I had visited Netherlands at the invitation of Fokker Space to persuade KNMI management to collaborate with NASA on this mission. And only 4 years after the first science meeting was held OMI was lunched on the Aura spacecraft. Next month will be the 6 th anniversary of this launch and very successful operation of OMI. All this was possible because of the leadership from Dr. Hennie Kelder and KNMI management who in 1998 saw the opportunity for Netherlands in the mission and stepped up to the challenge by creating a young and talented team of scientists at KNMI under the leadership of Dr. Pieterenel Levelt. This vision has now put Netherlands as the leading country in the world in monitoring air quality from space. Recent selection of TROPOMI by ESA attests to the success of this vision. I will present some selected highlights of our very successful collaboration on this project over the past 10 years.

Bhartia, Pawan K.↗

Evolutionary NAS in Light of Model Stability for Accurate Continual Learning

Continual learning, the capability to learn new knowledge from streaming data without forgetting the previous knowledge, is a critical requirement for dynamic learning systems, especially for emerging edge devices such as self-driving cars and drones. However, continual learning is still facing the catastrophic forgetting problem. Previous work illustrate that model performance on continual learning is not only related to the learning algorithms but also strongly dependent on the inherited model, i.e., the model where continual learning starts. The better stability of the inherited model, the less catastrophic forgetting and thus, the inherited model should be elaborately selected. Inspired by this finding, we develop an evolutionary neural architecture search (ENAS) algorithm that emphasizes the Stability of the inherited model, namely ENAS-S. ENAS-S aims to find optimal architectures for accurate continual learning on edge devices. On CIFAR-10 and CIFAR-100, we present that ENAS-S achieves competitive architectures with lower catastrophic forgetting and smaller model size when learning from a data stream, as compared with handcrafted DNNs.

Du, Xiaocong↗

Reconstructing Generalized Logical Networks of Transcriptional Regulation in Mouse Brain from Temporal Gene Expression Data

The problem of reconstructing generalized logical networks to account for temporal dependencies among genes and environmental stimuli from high-throughput transcriptomic data is addressed. A network reconstruction algorithm was developed that uses the statistical significance as a criterion for network selection to avoid false-positive interactions arising from pure chance. Using temporal gene expression data collected from the brains of alcohol-treated mice in an analysis of the molecular response to alcohol, this algorithm identified genes from a major neuronal pathway as putative components of the alcohol response mechanism. Three of these genes have known associations with alcohol in the literature. Several other potentially relevant genes, highlighted and agreeing with independent results from literature mining, may play a role in the response to alcohol. Additional, previously-unknown gene interactions were discovered that, subject to biological verification, may offer new clues in the search for the elusive molecular mechanisms of alcoholism.

59 BASIC BIOLOGICAL SCIENCES↗

Method for exploiting bias in factor analysis using constrained alternating least squares algorithms

Bias plays an important role in factor analysis and is often implicitly made use of, for example, to constrain solutions to factors that conform to physical reality. However, when components are collinear, a large range of solutions may exist that satisfy the basic constraints and fit the data equally well. In such cases, the introduction of mathematical bias through the application of constraints may select solutions that are less than optimal. The biased alternating least squares algorithm of the present invention can offset mathematical bias introduced by constraints in the standard alternating least squares analysis to achieve factor solutions that are most consistent with physical reality. In addition, these methods can be used to explicitly exploit bias to provide alternative views and provide additional insights into spectral data sets.

Keenan, Michael R↗

Constraints on Neutrino Oscillation Parameters from Neutrinos and Antineutrinos with Machine Learning

NOvA is a two detector, long baseline neutrino oscillation experiment measuring the oscillations of muon neutrinos from the \numi neutrino beam over a baseline of \SI{810}{km}. The experiment uses four oscillation channels, $\numu \rightarrow \numu$, $\numubar \rightarrow \numubar$, $\numu \rightarrow \nue$, and $\numubar \rightarrow \nuebar$, with a peak neutrino energy of \SI{1.8}{GeV}. This dissertation describes the analysis of these channels using a dataset of $13.6\times10^{20}$ protons on target neutrino beam mode and $12.5\times10^{20}$ protons on target antineutrino beam mode. The analysis makes use of improvements in the treatment of systematic uncertainties and machine learning techniques to reconstruct neutrino interactions. A technique for decorrelating systematic errors using principle component analysis was utilized to reduce and optimize neutrino cross section and beam related uncertainties. The improved machine learning algorithms make use of convolutional ne ural net works for neutrino event classification, particle classification, and instance segmentation. The selection of neutrino signal events utilizing the neutrino event classifier shows an efficiency of 63\% for the selection of electron neutrinos in neutrino beam mode and 75\% for electron antineutrinos in antineutrino beam mode. Using this algorithm, 82 appearing electron neutrino candidates and 33 appearing electron antineutrino candidates were observed with expected backgrounds of 26.8 and 14.0 respectively. In addition, 211 surviving muon neutrino candidates and 105 muon antineutrino candidates were identified with a purity of more than 96\% using the same neutrino event classifier. Fitting these data to the three flavor neutrino oscillation model, using constraints on \thetaonetwo, \thetaonethree, and \dmsqonetwo from solar and reactor neutrino experiments, the oscillation parameters are measured to be $\sintwothree = 0.57^{+0.04}_{-0.03}$, $\dmsqthreetwo = \SI[parse-numbers= false]{+ 2.41\pm0.07 \times 10^{-3}}{eV^2}$, and $\dcp=0.82^{+0.27}_{-0.87}\pi$ with a preference for the normal neutrino mass hierarchy. Leading systematic uncertainties for these measurements come from detector calibration and neutrino interaction models.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

VST ATLAS galaxy cluster catalogue I: cluster detection and mass calibration

Taking advantage of ∼4700 deg 2 optical coverage of the Southern sky offered by the VST ATLAS survey, we construct a new catalogue of photometrically selected galaxy groups and clusters using the orca cluster detection algorithm. The catalogue contains ∼22 000 detections with N 200 > 10 and ∼9000 with N 200 > 20. We estimate the photometric redshifts of the clusters using machine learning and find the redshift distribution of the sample to extend to z ∼ 0.7, peaking at z ∼ 0.25. We calibrate the ATLAS cluster mass-richness scaling relation using masses from the MCXC, Planck, ACT DR5, and SDSS redMaPPer cluster samples. We estimate the ATLAS sample to be > 95 per cent complete and > 85 per cent pure at z < 0.35 and in the M 200m >1 x 10 14 h -1 M ⊙ mass range. At z < 0.35, we also find the ATLAS sample to be more complete than redMaPPer, recovering a ~ 40 per cent higher fraction of Abell clusters. This higher sample completeness places the amplitude of the z < 0.35 ATLAS cluster mass function closer to the predictions of a ΛCDM model with parameters based on the Planck CMB analyses, compared to the mass functions of the other cluster samples. However, strong tensions between the observed ATLAS mass functions and models remain. We shall present a detailed cosmological analysis of the ATLAS cluster mass functions in paper II. In the future, optical counterparts to X-ray-detected eROSITA clusters can be identified using the ATLAS sample. The catalogue is also well suited for auxiliary spectroscopic target selection in 4MOST. The ATLAS cluster catalogue is publicly available at http://astro.dur.ac.uk/cosmology/vstatlas/cluster_catalogue/.

79 ASTRONOMY AND ASTROPHYSICS↗

Moment preserving adaptive particle weights using octree velocity distributions for PIC simulations

The ratio of computational to physical particles is of primary concern to statistical particle based simulations such as DSMC and PIC. An adaptive computational particle weight algorithm is presented that conserves mass, momentum, and energy. This algorithm is then enhanced with an octree adaptive mesh in velocity space to mitigate artificial thermalization. The new octree merge is compared to a merge that randomly selects merge partners for a bi-Maxwellian velocity distribution. Results for crossing beams in a fixed potential well along with an electrostatic PIC version with and without MCC collisions based ionizing breakdown show the advantages of the merge algorithm to both fixed particle weights and randomly selected merge partners.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Optimizing deep learning hyper-parameters through an evolutionary algorithm. In: MLHPC '15 Proceedings of the Workshop on Machine Learning in High-Performance Computing Environments, Article No. 4

There has been a recent surge of success in utilizing Deep Learning (DL) in imaging and speech applications for its relatively automatic feature generation and, in particular for convolutional neural networks (CNNs), high accuracy classification abilities. While these models learn their parameters through data-driven methods, model selection (as architecture construction) through hyper-parameter choices remains a tedious and highly intuition driven task. To address this, Multi-node Evolutionary Neural Networks for Deep Learning (MENNDL) is proposed as a method for automating network selection on computational clusters through hyper-parameter optimization performed via genetic algorithms.

Young, Steven R.↗