Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “neural operator learning”

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 847 records · Page 47

Passband Signal Detection at the Edge

Algorithms for radio frequency (RF) spectrum awareness need to be compatible with edge hardware to be practical for many applications. We developed a signal detection and classification model for the ZCU111 RF System-on-a-Chip (RFSoC) that operates on the fast Fourier transform of passband RF data. The system can detect and classify multiple signals of interest and display the predictions in real-time. The model consists of a modified ConvNeXt backbone and YOLOv3 head to operate on the Deep Learning Processing Unit on the RFSoC. We gathered datasets for training and testing by using a software defined radio to transmit example signals of Wi-Fi 802.11 b/g, Wi-Fi 802.11 n, FM Radio, LTE and LTE-M. By leveraging multiple inputs on the RFSoC frontend, the datasets span up to 4 GHz of bandwidth. The models showed high performance in classification accuracy, center frequency error, bandwidth error, and detection accuracy for both single and multi-signal datasets.

42 ENGINEERING↗

Early Information Parameter-Set Analysis for Satellite Close Approaches using Machine Learning

Understanding orbital mechanics is essential in space flight and navigation applications, and leveraging modern force models for flight path projection remains an important aspect in space mission design and operation. However, force models do not capture all the dynamics or perturbations in the space environment and thus are subject to errors in predicting the state vectors. The further out the predicted miss distance between spacecraft is from the time of closest approach (TCA), the larger the propagated errors in the predicted miss distance at TCA is. The dependency on these force models for spacecraft flight state prediction calls for a more reliable method that can quantify, or even reduce, these propagated errors. With recent advances in the field artificial intelligence, specifically in machine and deep learning algorithms, a model that implements these approaches can improve on the modern force model approach. The goal for this work is to provide an early-information decision-making threshold, in order to prioritize risk assessment implementation, given the ongoing increase of space objects. In analyzing the relationship of several parameters from conjunction data messages(CDMs) and solar information, early information becomes viable in miss distance prediction with unsupervised learning techniques, which learn the parameters that are linked together with miss distance and probability of collision (Pc) variables. Another approach implemented for identifying relationships within CDMs is supervised learning, in which a shallow neural network binary classifier learns to distinguish events with Pc values¡108. These parameters detected in the unsupervised process are then applied to a regression neural network, which predicts the miss distance at TCA for a specific event within a given uncertainty bound. For the regression neural network, a Long Short Term Memory (LSTM) neural network is implemented, which yields memory about each time step in an event. Using an LSTM network, the model learns to predict miss distance within 0.2km of the value measured at TCA. Although there is a limited amount of "close miss" data to train a network, the network learns to associate parameters, like large energy dissipation rates with the secondary object, with an elevated Pc

Brianna I. Robertson↗

Co-optimizing water-alternating-carbon dioxide injection projects using a machine learning assisted computational framework

In this report, a robust machine-learning-based computational framework that couples multi-layer neural network (MLNN) proxies and a multi-objective particle swarm optimizer (MOPSO) to design water-alternating-carbon dioxide injection (CO 2 -WAG) projects is presented. The proposed optimization protocol considers various objectives, including oil recovery and CO 2 storage volume. Expert MLNN systems are trained and employed as surrogate models of the high-fidelity compositional simulator in the optimization workflow. When multiple objective functions are considered, two approaches are employed to treat the objectives: the weighted sum method and the Pareto-front-based scheme. A field-scale implementation focusing on tertiary recovery in the Morrow B formation at Farnsworth Unit (FWU) is presented. The developed Pareto-optimal solutions indicate the maximal available oil production can be 1.64 × 10 7 barrels and maximal carbon storage can achieve 2.35 × 10 7 tons. Trade-offs factor is defined to divide the constructed Pareto front into 4 sections with the trade-off factors’ value ranges from 0.35 to 49.9. This work also compares the optimum solution found by the aggregative objective function and the solution repository covered by the Pareto front that considers the physical and operational constraints and reduces uncertainties involved by the multi-objective optimization process. Our comparison indicates multiple solutions exist to satisfy the objective criteria of the WAG design, and these results cannot be found using the traditional weighted sum method. The Pareto front solution can provide more options for project designers, but decisions regarding necessary trade-offs must be made using the solution repository to balance the project economics and CO 2 storage amount.

10 SYNTHETIC FUELS↗

MaLTESE: Large-Scale Simulation-Driven Machine Learning for Transient Driving Cycles

Optimal engine operation during a transient driving cycle is the key to achieving greater fuel economy, engine efficiency, and reduced emissions. In order to achieve continuously optimal engine operation, engine calibration methods use a combination of static correlations obtained from dynamometer tests for steady-state operating points and road and/or track performance data. As the parameter space of control variables, design variable constraints, and objective functions increases, the cost and duration for optimal calibration become prohibitively large. In order to reduce the number of dynamometer tests required for calibrating modern engines, a large-scale simulation-driven machine learning approach is presented in this work. A parallel, fast, robust, physics-based reduced-order engine simulator is used to obtain performance and emission characteristics of engines over a wide range of control parameters under various transient driving conditions (drive cycles). We scale the simulation up to 3,906 nodes of the Theta supercomputer at the Argonne Leadership Computing Facility to generate data required to train a machine learning model. The trained model is then used to predict various engine parameters of interest, and the results are compared with those predicted by the engine simulator. Our results show that a deepneural-network-based surrogate model achieves high accuracy: Pearson product-moment correlation values larger than 0.99 and mean absolute percentage error within 1.07% for various engine parameters such as exhaust temperature, exhaust pressure, nitric oxide, and engine torque. Once trained, the deep-neural-network-based surrogate model is fast for inference: it requires about 16 mu s for predicting the engine performance and emissions for a single design configuration compared with about 0.5 s per configuration with the engine simulator. Moreover, we demonstrate that transfer learning and retraining can be leveraged to incrementally retrain the surrogate model to cope with new configurations that fall outside the training data space.

deep learning↗

Performance of Advanced Signal Processing and Pattern Recognition Algorithms Using Raw Data from Ultrasonic Guided Waves and Fiber Optics Transducers

The Light Water Reactor Sustainability Program was initiated to evaluate technologies that could be used to perform online monitoring of piping and other secondary system structural components in commercial nuclear power plants. These online monitoring systems have the potential to identify when a more detailed inspection is needed using real-time measurements, rather than at a pre-determined inspection interval. This transition to condition-based, risk-informed automated maintenance will contribute to a significant reduction of operations and maintenance costs that account for most nuclear power generation costs. This report describes the current state of research related to ultrasonic guided wave testing and its application to detecting defects in commercial nuclear power plants. The report analyzes the applicability of the guided wave technology to secondary piping systems, as well as studying the potential for expanding the range of guided wave technology to include bent piping and other piping components. The ultrasonic guided waves can inspect long stretches of straight piping; however, more complex geometries such as elbows, welds, and tees are causing spurious reflections and coherent noise, which significantly decreases the sensitivity of the technique. To deal with these limitations, high-definition fiber optic sensors are applied to complex piping geometries, and advanced machine learning algorithms are used to detect deviations from healthy states. This report also analyzes the performance of advanced signal processing and machine learning-based pattern recognition algorithms in detecting defects in secondary structures. It is demonstrated on guided wave data collected at nuclear power plants that the independent component analysis can separate different coherent noise components and segregate them from useful signals. It also demonstrates that advanced machine learning techniques, such as neural networks, support vector machines, and deep learning networks, can detect minor defects present in inspected structures. Recommendation about the applicability of advanced machine learning techniques to online piping monitoring are also given.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

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↗

Graph neural networks for CO 2 solubility predictions in Deep Eutectic Solvents

Deep Eutectic Solvents (DESs) are a promising class of solvents for CO 2 capture. DESs are complex mixtures that can be designed to optimize CO solubility and overall capture process efficiency. However, the vast design landscape of DES mixtures makes experimental investigation prohibitive; as such, there is a need for computational models that can quickly and efficiently navigate the design space and inform data collection efforts. In this work, we propose Graph Neural Network (GNN) models for predicting CO 2 solubility for DESs; the GNN leverages a mixture graph representation that captures the molecular structure of the DES components as well as their intermolecular interactions. Here, we compare the GNN framework against alternative architectures (neural networks, graph convolution networks, and random forests) and data representations (molecular fingerprints, sigma profiles, and graphs). We show that the proposed approach offers superior predictive performance; specifically, we show that solubility can be predicted reliably directly from molecular structure (without the need of using sigma profiles as proposed in previous studies). This result is important, as obtaining sigma profiles requires expensive density functional theory computations. We also explored the ability of GNNs to predict solubility for new DES mixtures and operating conditions. We found that the model extrapolates across temperature reliably. However, we also found deficiencies in the ability of the model to predict solubility for DES mixtures, pressures, and molar ratio not included in the training sets; we show that this is due to an inherent lack of chemical diversity in datasets available in the literature. The proposed computational capabilities can thus help navigate the design space of DES and inform data collection efforts. Our models, data, and benchmarks are shared as Python code implemented in Jupyter notebooks.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Out-of-distribution detection with non-parametric density estimation for models predicting processing history of uranium ore concentrates

The rapid advancement in machine learning (ML) and computer vision (CV) coincides with the growth of interest in deploying these ML/CV models in numerous fields from medicine to social science. Similar to those areas, we have witnessed a great number of works in materials science employing ML/CV models – neural networks in particular – in their studies in recent years. These models have proven to obtain accurate performance in various tasks. However, these models struggle to attain a similar performance when encountering test samples coming from a distribution that is different from the training set. More importantly, they fail without providing any warning to the users. Therefore, we propose a framework for detecting out-of-distribution (OOD) samples to alert users when a human intervention might be necessary in this work. Specifically, we explore the use of a non-parametric density estimation method to detect OOD samples. Here, we assess OOD detection capability of the proposed framework on ML models developed for categorizing precipitation routes of U 3 O 8 when encountering OOD datasets that contain samples (1) undergone different imaging acquisition process, (2) undergone different material synthesis process, and (3) different materials than ID set. Through those experiments, we achieve an average area under the receiver operating characteristic (AUROC) of at least 91% on average in detecting OOD samples. With minimal overhead cost and superior performance, the proposed framework enables a reliable and safe system when deploying in real-world scenarios.

Convolutional neural networks↗

Data-driven models of nonautonomous systems

Nonautonomous dynamical systems are characterized by time-dependent inputs, which complicates the discovery of predictive models describing the spatiotemporal evolution of the state variables of quantities of interest from their temporal snapshots. When dynamic mode decomposition (DMD) is used to infer a linear model, this difficulty manifests itself in the need to approximate the time-dependent Koopman operators. Our approach is to approximate the original nonautonomous system with a modified system derived via a local parameterization of the time-dependent inputs. The modified system comprises a sequence of local parametric systems, which are subsequently approximated by a parametric surrogate model using the DRIPS (dimension reduction and interpolation in parameter space) framework. The offline step of DRIPS relies on DMD to build a linear surrogate model, endowed with reduced-order bases for the observables mapped from training data. The online step interpolates on suitable manifolds to construct a sequence of iterative parametric surrogate models; the target/test parameter points on these manifolds are specified by a local parameterization of the test time-dependent inputs. Here, we use numerical experimentation to demonstrate the robustness of our method and compare its performance with that of deep neural networks.

97 MATHEMATICS AND COMPUTING↗

Identifiability and predictability of integer- and fractional-order epidemiological models using physics-informed neural networks

Here we analyze a plurality of epidemiological models through the lens of physics-informed neural networks (PINNs) that enable us to identify time-dependent parameters and data-driven fractional differential operators. In particular, we consider several variations of the classical susceptible-infectious-removed (SIR) model by introducing more compartments and fractional-order and time-delay models. We report the results for the spread of COVID-19 in New York City, Rhode Island and Michigan states and Italy, by simultaneously inferring the unknown parameters and the unobserved dynamics. For integer-order and time-delay models, we fit the available data by identifying time-dependent parameters, which are represented by neural networks. In contrast, for fractional differential models, we fit the data by determining different time-dependent derivative orders for each compartment, which we represent by neural networks. We investigate the structural and practical identifiability of these unknown functions for different datasets, and quantify the uncertainty associated with neural networks and with control measures in forecasting the pandemic.

60 APPLIED LIFE SCIENCES↗

A smart alarm for particle accelerator beamline operations

We present the initial results of a proof-of-concept 'smart alarm' for the Continuous Electron Beam Accelerator Facility injector beamline at Jefferson Lab. To minimize machine downtime and improve operational efficiency, an autonomous alarm system able to identify and diagnose unusual machine states is needed. Our approach leverages a trained neural network capable of alerting operators (a) when an anomalous condition exists in the beamline and (b) identifying the element setting that is the root cause. The tool is based on an inverse model that maps beamline readings (diagnostic readbacks) to settings (beamline attributes operators can modify). The model takes as input readings from the machine and computes machine settings which are compared to control setpoints. Instances where predictions differ from setpoints by a user-defined threshold are flagged as anomalous. Given data corresponding to 354 anomalous injector configurations, the model can narrow the root cause of an anomalous condition to three potential candidates with 94.6% accuracy. Furthermore, compared to the current method of identifying anomalous conditions which raises an alarm when machine parameters drift outside their normal tolerances, the data-driven model can identify 83% more anomalous conditions.

43 PARTICLE ACCELERATORS↗

Explainable and Differentiable Reinforcement Learning for Multi-objective Optimization in Particle Accelerators

Operating particle accelerators involves optimizing multiple goals simultaneously, which can be challenging due to trade-offs among objectives. While evolutionary algorithms like the genetic algorithm (GA) have been used for various Multi-Objective Optimization (MOO) tasks, they are not inherently suited for complex control problems. This talk highlights two variations of Reinforcement Learning (RL) for concurrently optimizing heat load and trip rates at the Continuous Electron Beam Accelerator Facility (CEBAF). The problem involves strict constraints on individual states, actions, and overall energy requirements of the beam. First, this talk highlights how differentiability can be harnessed through a Deep Differentiable Reinforcement Learning (DDRL) approach to address MOO issues within particle accelerators. We examine the DDRL method alongside Model Free Reinforcement Learning (MFRL), GA, and Bayesian Optimization (BO). The performance of these methods is assessed by generating a Pareto-front for two objectives. Our findings indicate that DDRL excels in handling high-dimensional problems more effectively than MFRL, BO, and GA. Next, we will show integration of explainable physics-based constraints into RL algorithms to enhance trans- parency and trust in decision-making processes by enabling users to verify that agents adhere to established physical principles. This surrogate function can be modeled using neural networks or sparse dictionary mod- els. By examining the mathematical form of the learned constraint function, we are able to confirm the agent has learned to use the established physics of each environment provided but the surrogate model. In addi- tion, we find that the introduction of a mathematical functional dictionary based surrogate model enables our reinforcement learning algorithms to reliably converge for difficult high-dimensional accelerator controls environments.

Rajput, Kishansingh [Thomas Jefferson National Acc↗

Efficient data-driven models for prediction and optimization of geothermal power plant operations

Increasing the capacity of geothermal energy as a renewable resource calls for development and deployment of efficient control and optimization technologies for geothermal power plants. A data-driven prediction and optimization model is presented as a cost-effective and efficient alternative to physics-based approach. The model predicts power output and operational cost by propagating the influence of control and disturbance variables within an artificial neural network (ANN). Numerical experiments with simulated and field data from a real geothermal power plant are first used to demonstrate the prediction performance of the ANN model. The model is then adopted to maximize the net predicted power production by automatically adjusting the working fluid circulation rate. The optimization performance of the model in evaluated using a thermodynamic flowsheet simulation model. The workflow is applied to model and control the effect of ambient temperature on an air-cooled binary cycle power plant, which is complex and costly to perform using a physics-based predictive model. As a result, the performance of the method is demonstrated by applying it to both simulated and field datasets from a binary cycle geothermal power plant.

15 GEOTHERMAL ENERGY↗

A New Modeling Framework for Geothermal Operational Optimization with Machine Learning (GOOML)

Geothermal power plants are excellent resources for providing low carbon electricity generation with high reliability. However, many geothermal power plants could realize significant improvements in operational efficiency from the application of improved modeling software. Increased integration of digital twins into geothermal operations will not only enable engineers to better understand the complex interplay of components in larger systems but will also enable enhanced exploration of the operational space with the recent advances in artificial intelligence (AI) and machine learning (ML) tools. Such innovations in geothermal operational analysis have been deterred by several challenges, most notably, the challenge in applying idealized thermodynamic models to imperfect as-built systems with constant degradation of nominal performance. This paper presents GOOML: a new framework for Geothermal Operational Optimization with Machine Learning. By taking a hybrid data-driven thermodynamics approach, GOOML is able to accurately model the real-world performance characteristics of as-built geothermal systems. Further, GOOML can be readily integrated into the larger AI and ML ecosystem for true state-of-the-art optimization. This modeling framework has already been applied to several geothermal power plants and has provided reasonably accurate results in all cases. Therefore, we expect that the GOOML framework can be applied to any geothermal power plant around the world.

15 GEOTHERMAL ENERGY↗

Universal Polarization Transformations: Spatial Programming of Polarization Scattering Matrices Using a Deep Learning‐Designed Diffractive Polarization Transformer

Abstract Controlled synthesis of optical fields having nonuniform polarization distributions presents a challenging task. Here, a universal polarization transformer is demonstrated that can synthesize a large set of arbitrarily‐selected, complex‐valued polarization scattering matrices between the polarization states at different positions within its input and output field‐of‐views (FOVs). This framework comprises 2D arrays of linear polarizers positioned between isotropic diffractive layers, each containing tens of thousands of diffractive features with optimizable transmission coefficients. After its deep learning‐based training, this diffractive polarization transformer can successfully implement N i N o = 10 000 different spatially‐encoded polarization scattering matrices with negligible error, where N i and N o represent the number of pixels in the input and output FOVs, respectively. This universal polarization transformation framework is experimentally validated in the terahertz spectrum by fabricating wire‐grid polarizers and integrating them with 3D‐printed diffractive layers to form a physical polarization transformer. Through this set‐up, an all‐optical polarization permutation operation of spatially‐varying polarization fields is demonstrated, and distinct spatially‐encoded polarization scattering matrices are simultaneously implemented between the input and output FOVs of a compact diffractive processor. This framework opens up new avenues for developing novel devices for universal polarization control and may find applications in, e.g., remote sensing, medical imaging, security, material inspection, and machine vision.

Optical neural networks↗

FluxSat: Long-term Earth Science Data Record (ESDR) for Terrestrial Gross Primary Production (GPP) based on satellite data calibrated with eddy covariance data

Gross primary production (GPP), the amount of carbon dioxide (CO 2 ) assimilated by plants through photosynthesis, is one of the most variable and uncertain components of the global carbon cycle. Global GPP has been estimated with a number of process-based models, data-driven, and hybrid approaches. Dynamic global vegetation models (DGVMs), driven by observed environmental changes, are used for global carbon budget assessments and long-term (climate) prediction. Benchmarking these and other models globally with data-driven GPP estimates is critical for understanding the land sink and ensuring accurate forecasts of the carbon cycle. In addition, global data-driven GPP estimates are crucial for studies of interannual variability, including trends that are linked to mechanisms with large uncertainties, such as the indirect CO 2 fertilization effect related to greening. In response to a community need for a GPP data set that well captures spatio-temporal variability, we developed FluxSat, a data-driven approach that optimizes the use of satellite reflectance data from the NASA MODerate-resolution Imaging Spectroradiometer (MODIS) on the Terra and Aqua satellites, calibrated using ground-based eddy covariance (EC) data. We are enhancing (spatially, higher resolution) and extending FluxSat (in time, with additional sensors) to create a high quality long term GPP Earth System Data Record (ESDR) for use in model benchmarking, carbon cycle modeling, and studies of trends and interannual variability. Our team’s objectives are to: 1. Update and document the current MODIS FluxSat GPP (daily, 0.05o and 0.5o resolutions) products with latest available MODIS and EC data sets; 2. Extend FluxSat GPP record forward in time with the Visible Infrared Imaging Radiometer Suite (VIIRS) on operational weather satellites going forward; 3. Extend FluxSat GPP record backward in time using the Advanced Very High Resolution Radiometer (AVHRR) on weather satellites dating back to 1981; 4. Provide higher spatial resolution MODIS and VIIRS GPP (0.0083o). 5. Thoroughly evaluate all FluxSat products with independent data; and 6. Create a homogenized long-term GPP record spanning 40+ years. We will discuss plans for this long-term data set that is supported through the NASA Making Earth System Data Records for Use in Research Environments (MEaSUREs) program.

gross Primary Production↗

Mesh-based super-resolution of fluid flows with multiscale graph neural networks

A graph neural network (GNN) approach is introduced in this work which enables mesh-based three-dimensional super-resolution of fluid flows. In this framework, the GNN is designed to operate not on the full mesh-based field at once, but on localized meshes of elements (or cells) directly. To facilitate mesh-based GNN representations in a manner similar to spectral (or finite) element discretizations, a baseline GNN layer (termed a message passing layer, which updates local node properties) is modified to account for synchronization of coincident graph nodes, rendering compatibility with commonly used element-based mesh connectivities. Furthermore, the architecture is multiscale in nature, and is comprised of a combination of coarse-scale and fine-scale message passing layer sequences (termed processors) separated by a graph unpooling layer. The coarse-scale processor embeds a query element (alongside a set number of neighboring coarse elements) into a single latent graph representation using coarse-scale synchronized message passing over the element neighborhood, and the fine-scale processor leverages additional message passing operations on this latent graph to correct for interpolation errors. Demonstration studies are performed using hexahedral mesh-based data from Taylor–Green Vortex and backward-facing step flow simulations at Reynolds numbers of 1600 and 3200. Through analysis of both global and local errors, the results ultimately show how the GNN is able to produce accurate super-resolved fields compared to targets in both coarse-scale and multiscale model configurations. Reconstruction errors for fixed architectures were found to increase in proportion to the Reynolds number. Geometry extrapolation studies on a separate cavity flow configuration show promising cross-mesh capabilities of the super-resolution strategy.

Backward-facing step↗

Cascade Error Projection: A Learning Algorithm for Hardware Implementation

In this paper, we workout a detailed mathematical analysis for a new learning algorithm termed Cascade Error Projection (CEP) and a general learning frame work. This frame work can be used to obtain the cascade correlation learning algorithm by choosing a particular set of parameters. Furthermore, CEP learning algorithm is operated only on one layer, whereas the other set of weights can be calculated deterministically. In association with the dynamical stepsize change concept to convert the weight update from infinite space into a finite space, the relation between the current stepsize and the previous energy level is also given and the estimation procedure for optimal stepsize is used for validation of our proposed technique. The weight values of zero are used for starting the learning for every layer, and a single hidden unit is applied instead of using a pool of candidate hidden units similar to cascade correlation scheme. Therefore, simplicity in hardware implementation is also obtained. Furthermore, this analysis allows us to select from other methods (such as the conjugate gradient descent or the Newton's second order) one of which will be a good candidate for the learning technique. The choice of learning technique depends on the constraints of the problem (e.g., speed, performance, and hardware implementation); one technique may be more suitable than others. Moreover, for a discrete weight space, the theoretical analysis presents the capability of learning with limited weight quantization. Finally, 5- to 8-bit parity and chaotic time series prediction problems are investigated; the simulation results demonstrate that 4-bit or more weight quantization is sufficient for learning neural network using CEP. In addition, it is demonstrated that this technique is able to compensate for less bit weight resolution by incorporating additional hidden units. However, generation result may suffer somewhat with lower bit weight quantization.

Duong, Tuan A.↗