Search NASA⌕ Search

SEARCH · Search NASA

Results for “Deep Convolutional Neural Network”

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 271 records · Page 15
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

Exploration of Quantum Machine Learning and AI Accelerators for Fusion Science

The dawn of the noisy intermediate-scale quantum (NISQ) era sparked rapid development in variational quantum algorithms. These algorithms, utilizing parameterized quantum circuit optimized by classical computers with feedback, are practical under the constraints of the current hardware and can potentially show quantum advantage. In recent years, these variational circuit are applied to neural networks, hoping to boost the triumphant success of deep learning. We simulate various of quantum-classical hybrid neural networks applied to scientific tasks, and seek to understand their capabilities and limitations. Specifically, we devise quantum convolutional layers and apply them to deep neural networks used for predicting plasma disruption in fusion reactors.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2022

Automated Stellar Spectra Classification with Ensemble Convolutional Neural Network

Large sky survey telescopes have produced a tremendous amount of astronomical data, including spectra. Machine learning methods must be employed to automatically process the spectral data obtained by these telescopes. Classification of stellar spectra by applying deep learning is an important research direction for the automatic classification of high-dimensional celestial spectra. In this paper, a robust ensemble convolutional neural network (ECNN) was designed and applied to improve the classification accuracy of massive stellar spectra from the Sloan digital sky survey. We designed six classifiers which consist six different convolutional neural networks (CNN), respectively, to recognize the spectra in DR16. Then, according the cross-entropy testing error of the spectra at different signal-to-noise ratios, we integrate the results of different classifiers in an ensemble learning way to improve the effect of classification. The experimental result proved that our one-dimensional ECNN strategy could achieve 95.0% accuracy in the classification task of the stellar spectra, a level of accuracy that exceeds that of the classical principal component analysis and support vector machine model.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2022

TULIP: An RNA-seq-based Primary Tumor Type Prediction Tool Using Convolutional Neural Networks

Background: With cancer as one of the leading causes of death worldwide, accurate primary tumor type prediction is critical in identifying genetic factors that can inhibit or slow tumor progression. There have been efforts to categorize primary tumor types with gene expression data using machine learning, and more recently with deep learning, in the last several years. Methods In this paper, we developed four 1-dimensional (1D) Convolutional Neural Network (CNN) models to classify RNA-seq count data as one of 17 highly represented primary tumor types or 32 primary tumor types regardless of imbalanced representation. Additionally, we adapted the models to take as input either all Ensembl genes (60,483) or protein coding genes only (19,758). Unlike previous work, we avoided selection bias by not filtering genes based on expression values. RNA-seq count data expressed as FPKM-UQ of 9,025 and 10,940 samples from The Cancer Genome Atlas (TCGA) were downloaded from the Genomic Data Commons (GDC) corresponding to 17 and 32 primary tumor types respectively for training and validating the models. Results: All 4 1D-CNN models had an overall accuracy of 94.7% to 97.6% on the test dataset. Further evaluation indicates that the models with protein coding genes only as features performed with better accuracy compared to the models with all Ensembl genes for both 17 and 32 primary tumor types. For all models, the accuracy by primary tumor type was above 80% for most primary tumor types. Conclusions: We packaged all 4 models as a Python-based deep learning classification tool called TULIP (TUmor CLassIfication Predictor) for performing quality control on primary tumor samples and characterizing cancer samples of unknown tumor type. Further optimization of the models is needed to improve the accuracy of certain primary tumor types.

Jones, Sara↗
DOE OSTI2023

Dilated causal convolutional neural networks for forecasting zone airflow to estimate short-term energy consumption

Here this paper investigates the use of dilated causal convolutional neural networks for fine- grained temporal forecasting of building zone states. Specifically, we build and evaluate models using a small set of exogenous features (e.g., external temperature) to autoregressively predict zone airflow setpoints every minute for a 24-hour prediction window. We carefully explore the trade-off between generality and specificity in these models, training and evaluating them based on zone, zone type, month, season, and combinations thereof. When evaluated for a commercial office building in Eastern Washington with 16 zones served by variable air volume air handling units, we find that the highest performance comes from a zone-specific, season-agnostic approach; with it, we obtain an R 2 of 0.704 (averaged over zones) and an average normalized root mean square error (nRMSE) of 0.111. In contrast, the most general model (trained across all zones and seasons) yields an R 2 of only 0.416 and a nRMSE of 0.168, while a baseline zone-specific reduced order model obtains 0.443 R 2 and 0.159 nRMSE. We also report on factors affecting airflow forecasting performance, on the ability of models trained on a specific zone to generalize to other zones, and on the capability of those models trained on a specific month to generalize to other months.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2022

Artificial neural network approach for multiphase segmentation of battery electrode nano-CT images

The segmentation of tomographic images of the battery electrode is a crucial processing step, which will have an additional impact on the results of material characterization and electrochemical simulation. However, manually labeling X-ray CT images (XCT) is time-consuming, and these XCT images are generally difficult to segment with histographical methods. We propose a deep learning approach with an asymmetrical depth encode-decoder convolutional neural network (CNN) for real-world battery material datasets. This network achieves high accuracy while requiring small amounts of labeled data and predicts a volume of billions voxel within few minutes. While applying supervised machine learning for segmenting real-world data, the ground truth is often absent. The results of segmentation are usually qualitatively justified by visual judgement. We try to unravel this fuzzy definition of segmentation quality by identifying the uncertainty due to the human bias diluted in the training data. Further CNN trainings using synthetic data show quantitative impact of such uncertainty on the determination of material’s properties. Nano-XCT datasets of various battery materials have been successfully segmented by training this neural network from scratch. We will also show that applying the transfer learning, which consists of reusing a well-trained network, can improve the accuracy of a similar dataset.

25 ENERGY STORAGE↗
DOE OSTI2020

Multimodal imaging and machine learning to enhance microscope images of shale

A machine learning based image processing workflow is presented to enhance shale source rock microscopic images obtained using diverse imaging platforms. Images were acquired from a 30 μm diameter cylindrical Vaca Muerta shale sample using both nondestructive Transmission X-Ray Microscopy (TXM, alternately referred to as nano computed tomography) and destructive Focused Ion Beam-Scanning Electron Microscopy (FIB-SEM). Output cross-sectional images from each modality were aligned using a combination of manual and automated registration techniques to create a registered image dataset. We then apply this dataset for two image processing tasks: prediction of image cross sections with SEM-like resolution from nondestructive TXM data and repair of charged region artifacts (localized accumulation of electrons) within SEM images. The image processing algorithms for both tasks use deep learning models, specifically image-to-image Convolutional Neural Networks (CNNs) and conditional Generative Adversarial Networks (cGANs). In the image enhancement tasks, we are able to achieve significant qualitative and quantitative improvement in TXM images. Here, the best model reaches an average Peak Signal to Noise Ratio (PSNR) of 15.8 dB. Conditioning on TXM data is also shown to reduce artifacts from SEM charging, achieving an average PSNR of 25.8 dB. Furthermore, our results suggest that properly trained and validated networks are capable of significant enhancement of images obtained using nondestructive techniques, thereby improving interpretation of two- and three-dimensional images while preserving samples for future use.

58 GEOSCIENCES↗
DOE OSTI2021

Iterative self-organizing SCEne-LEvel sampling (ISOSCELES) for large-scale building extraction

Convolutional neural networks (CNN) provide state-of-the-art performance in many computer vision tasks, including those related to remote-sensing image analysis. Successfully training a CNN to generalize well to unseen data, however, requires training on samples that represent the full distribution of variation of both the target classes and their surrounding contexts. With remote sensing data, acquiring a sufficiently representative training set is a challenge due to both the inherent multi-modal variability of satellite or aerial imagery and the general high cost of labeling data. To address this challenge, we have developed ISOSCELES, an Iterative Self-Organizing SCEne LEvel Sampling method for hierarchical sampling of large image sets. Using affinity propagation, ISOSCELES automates the selection of highly representative training images. Compared to random sampling or using available reference data, the distribution of the training is principally data driven, reducing the chance of oversampling uninformative areas or undersampling informative ones. In comparison to manual sample selection by an analyst, ISOSCELES exploits descriptive features, spectral and/or textural, and eliminates human bias in sample selection. Using a hierarchical sampling approach, ISOSCELES can obtain a training set that reflects both between-scene variability, such as in viewing angle and time of day, and within-scene variability at the level of individual training samples. We verify the method by demonstrating its superiority to stratified random sampling in the challenging task of adapting a pre-trained model to a new image and spatial domain for country-scale building extraction. Using a pair of hand-labeled training sets comprising 1,987 sample image chips, a total of 496,000,000 individually labeled pixels, we show, across three distinct model architectures, an increase in accuracy, as measured by F1-score, of 2.2–4.2%.

42 ENGINEERING↗
DOE OSTI2025

Multi-modality deep learning for pulse prediction in homogeneous nonlinear systems via parametric conversion

In this Letter, we introduce FusionNet, a multi-modality deep learning framework designed to predict and analyze output pulses in high-power rare-earth-doped laser systems driving parametric conversion in homogeneous guided nonlinear media. FusionNet integrates temporal, spectral, and physical experimental conditions to model ultrafast nonlinear phenomena, including parametric nonlinear frequency conversion, self-phase modulation, and cross-phase modulation in homogeneous guided systems such as gas-filled hollow-core fibers. These systems bridge physical models with experimental data, advancing our understanding of light-guiding principles and nonlinear interactions while expediting the design and optimization of on-demand high-power, high-brightness systems. Our results demonstrate a 73% reduction in prediction error and an 83% improvement in computational efficiency compared to conventional neural networks. This work establishes a new paradigm for accelerating parametric simulations and optimizing experimental designs in high-power laser systems, with further implications for high-precision spectroscopy, quantum information science, and distributed entangled interconnects.

47 OTHER INSTRUMENTATION↗
DOE OSTI2026

A hybrid CNN-LSTM surrogate model for hyper-resolution spatiotemporal flood forecasting in Norfolk, Virginia

Study region: Norfolk, Virginia, United States Study focus: Accurate and timely flood forecasting is essential for enhancing resilience in coastal urban areas in the context of increasing frequency and intensity of rainfall, sea level rise and urbanization. This study presents a hybrid deep learning-based surrogate model that integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks to enable real-time spatiotemporal flood forecasting. The model leverages CNN to capture spatial features from inputs such as elevation and Topographic Wetness Index (TWI), while LSTM processes time-series inputs of rainfall and tide data to capture temporal features. New hydrologic insights for the region: The hybrid CNN-LSTM model was trained using the physics-based hydrodynamic model simulations obtained from the Two-dimensional Unsteady FLOW (TUFLOW) model for Norfolk, Virginia, and achieved high predictive accuracy across diverse flood-prone areas. The reduced computational time from four to six hours using TUFLOW to 3.2 min per event using CNN-LSTM enables rapid flood inundation mapping and early warning applications. The model effectively captured both spatial flood extents and their temporal evolution across different flooding scenarios, providing forecasts at a 2.5-m spatial resolution and 15-min temporal resolution and a one-hour-ahead prediction horizon. While challenges remain in terms of transferability to new regions and real-time data assimilation, this approach demonstrates strong potential for supporting operational flood risk management in coastal urban environments.

Coastal urban flooding↗
DOE OSTI2024

Predicting Missing Regions in Charged Particle Tracks Using a Sparse 3D Convolutional Neural Network

The 2x2 Demonstrator is a prototype of ND-LAr, the liquid argon time-projection chamber of the Deep Underground Neutrino Experiment’s Near Detector complex. Both the 2x2 Demonstrator and ND-LAr are modular detectors that will have pixelated charge readouts and inactive regions wherein there is no sensitivity to charge deposition and light signals that arise from charged particle interactions with liquid argon. In the 2x2, these inactive regions are located in between the active detector modules, which introduces the challenge of inferring what charge signals ought to look like in these regions. This study explores the use of a Sparse 3D Convolutional Neural Network (ConvNet) to infer missing regions in charged particle tracks. Hits corresponding to energy depositions are voxelized into a three-dimensional grid for each track. Voxels that fall into predefined inactive regions are removed to simulate the lack of detector output. The model is trained to infer the topology of the missing track voxels, with the ultimate goal of inferring the missing charge or energy values in these voxels as well. Results indicate that this approach shows promise in prediction of missing track regions with some accuracy.

Utaegbulam, Hilary↗
DOE OSTI2022

DISTEMA: distance map-based estimation of single protein model accuracy with attentive 2D convolutional neural network

Abstract Background Estimation of the accuracy (quality) of protein structural models is important for both prediction and use of protein structural models. Deep learning methods have been used to integrate protein structure features to predict the quality of protein models. Inter-residue distances are key information for predicting protein’s tertiary structures and therefore have good potentials to predict the quality of protein structural models. However, few methods have been developed to fully take advantage of predicted inter-residue distance maps to estimate the accuracy of a single protein structural model. Result We developed an attentive 2D convolutional neural network (CNN) with channel-wise attention to take only a raw difference map between the inter-residue distance map calculated from a single protein model and the distance map predicted from the protein sequence as input to predict the quality of the model. The network comprises multiple convolutional layers, batch normalization layers, dense layers, and Squeeze-and-Excitation blocks with attention to automatically extract features relevant to protein model quality from the raw input without using any expert-curated features. We evaluated DISTEMA’s capability of selecting the best models for CASP13 targets in terms of ranking loss of GDT-TS score. The ranking loss of DISTEMA is 0.079, lower than several state-of-the-art single-model quality assessment methods. Conclusion This work demonstrates that using raw inter-residue distance information with deep learning can predict the quality of protein structural models reasonably well. DISTEMA is freely at https://github.com/jianlin-cheng/DISTEMA

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2024

DASEventNet: AI‐Based Microseismic Detection on Distributed Acoustic Sensing Data From the Utah FORGE Well 16A (78)‐32 Hydraulic Stimulation

Abstract Distributed acoustic sensing (DAS) has emerged as a promising seismic technology for monitoring microearthquakes (MEQs) with high spatial resolution. Efficient algorithms are needed for processing large DAS data volumes. This study introduces a deep learning (DL) model based on a Residual Convolutional Neural Network (ResNet) for detecting MEQs using DAS data, named as DASEventNet. The test data were collected from the Utah FORGE 16A (78)‐32 hydraulic stimulation experiments conducted in April 2022. The DASEventNet model achieves a remarkable accuracy of 100% when discriminating MEQs from noise in the raw test set of 260 examples. Surprisingly, the model identified weak MEQ signatures that have been manually categorized as noise. The decision‐making process with the model is decoded by the classic activation map, which illuminates learning features of the DASEventNet model. These features provide clear illustrations of weak MEQs and varied noise types. Finally, we apply the trained model to the entire period (∼7 days) of continuous DAS recordings and find that it discovers >5,700 new MEQs, previously unregistered in the public Silixa DAS catalog. The DASEventNet model significantly outperforms the traditional seismic method Short‐Term Average/Long‐Term Average (STA/LTA), which detected only 1,307 MEQs. The DASEventNet detection threshold is M w −1.80 compared to the minimum magnitude of M w −1.14 detected by STA/LTA. The spatiotemporal distribution of the newly identified MEQs defines an extensive stimulation zone and more accurately characterizes fracture geometry. Our results highlight the potential of DL for long‐term, real‐time microseismic monitoring that can improve enhanced geothermal systems and other activities that include subsurface hydraulic fracturing.

15 GEOTHERMAL ENERGY↗
DOE OSTI2025

Deep-learning-enhanced assessment of wellbore barrier effectiveness in geologic storage systems with intermediate aquifers

For geologic systems where carbon dioxide (CO 2 ) is injected underground, existing wells represent potential pathways for fluid migration. Here, this study introduces a novel deep learning model to quantify the likelihood and potential magnitude of fluid migration through wellbores at sites with intermediate aquifers or thief zones between the injection units and underground drinking water sources. Synthetic datasets, generated using reservoir simulations, captured a wide range of subsurface conditions, well attributes, operational parameters, and fluid migration scenarios. Among the regression models developed to predict brine and CO 2 leakage rates and CO 2 saturations along leaky wellbores, convolutional neural network (CNN) outperformed both Light Gradient Boosting Machine and deep neural network. Additionally, a CNN-based classification model was created to predict whether brine and CO 2 would leak along a wellbore, further improving performance over regression alone. The best models were integrated into the National Risk Assessment Partnership Open-source Integrated Assessment Model for rapid, stochastic assessment of storage system containment and leakage risks. A case study demonstrated the model’s ability to simulate fluid migration through existing wells with multiple intermediate aquifers. This computationally efficient wellbore model offers value in support of site performance evaluation and risk-informed decision making by stakeholders.

CO2 leakage↗
DOE OSTI2020

Application of deep neural networks to improve diagnostic accuracy of rheumatoid arthritis using diffuse optical tomography

A set of deep neural network models for rheumatoid arthritis (RA) classification using a highway network, a convolutional neural network and a residual network is proposed based on the data of diffuse optical tomography (DOT) utilising near-infrared light, which ensures early diagnosis of pathophysiological changes resulting from inflammation. A numerical model of the finger is used to generate images to overcome the inherent problem of insufficient clinical DOT images available. The proposed deep neural network models are applied to automatically classify simulated DOT images of inflamed and non-inflamed joints and transfer learning is also used to improve the performance of the classification. The results demonstrate that all three deep neural network methods improve the diagnostic accuracy as compared to the widely applied support vector machine (SVM), especially for high inter-subject variability databases. In cases of distinct modelled severity of disease, residual network achieved the highest accuracy (> 99 %), and both of highway and convolutional neural networks reached 99 %, respectively. However, as the severity of the modelled disease is reduced, this accuracy is reduced to 75.2 % for residual networks. The results indicate that transfer learning can improve the performance of deep neural network methods on RA classification from DOT data and highlight their potential as a computer aided tool in DOT diagnostic systems. (paper)

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2023

Identifying common stored product insects using automated deep learning methods

Monitoring stored product insect pests is a common practice for post-harvest management of stored grain and grain-based commodities, which helps ensure product quality from harvest to final consumer. Current methods of sampling and monitoring can be time-consuming, labor-intensive, expensive and require expertise in insect identification. Therefore, this study aims to develop an image-based automated identification system for common stored product insect species using deep-learning methods. Top-down images of the common stored product adult insect species of Rhyzopertha dominica, Cryptolestes ferrugineus, Tribolium castaneum, Sitophilus oryzae, and Oryzaephilus surinamensis were acquired and analyzed. Deep learning-based, state-of-the-art Convolutional Neural Networks (CNN) models (ResNet-50, MobileNet-v2, DarkNet-53, and EfficientNet-b0) were fine-tuned with a transfer learning approach to classify the insect species. All models were able to correctly identify the insect species with at least 96% accuracy and with few misclassifications. One issue with trained CNNs is that they do not explain the reasoning for the classification and are often called a “black box”. Therefore, visualization methods called Gradient-weighted Class Activation Mapping (Grad-CAM) were implemented to explore the black box network. The Grad-CAM uses heat maps to highlight the major image features that the network focused on to make insect species predictions. The Grad-CAM verifies the network's prediction and also helps improve network performance. This study contributes to the overall goal of developing a camera-based system for monitoring stored grain insects. As a result, the developed system would empower warehouse, flour mills, and other food facilities with a tool to quickly and accurately identify insect species in stored product environments and could be implemented as part of a close to real-time monitoring system.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2022

Scalable deep learning for watershed model calibration

Watershed models such as the Soil and Water Assessment Tool (SWAT) consist of high-dimensional physical and empirical parameters. These parameters often need to be estimated/calibrated through inverse modeling to produce reliable predictions on hydrological fluxes and states. Existing parameter estimation methods can be time consuming, inefficient, and computationally expensive for high-dimensional problems. In this paper, we present an accurate and robust method to calibrate the SWAT model (i.e., 20 parameters) using scalable deep learning (DL). We developed inverse models based on convolutional neural networks (CNN) to assimilate observed streamflow data and estimate the SWAT model parameters. Scalable hyperparameter tuning is performed using high-performance computing resources to identify the top 50 optimal neural network architectures. We used ensemble SWAT simulations to train, validate, and test the CNN models. We estimated the parameters of the SWAT model using observed streamflow data and assessed the impact of measurement errors on SWAT model calibration. We tested and validated the proposed scalable DL methodology on the American River Watershed, located in the Pacific Northwest-based Yakima River basin. Our results show that the CNN-based calibration is better than two popular parameter estimation methods (i.e., the generalized likelihood uncertainty estimation [GLUE] and the dynamically dimensioned search [DDS], which is a global optimization algorithm). For the set of parameters that are sensitive to the observations, our proposed method yields narrower ranges than the GLUE method but broader ranges than values produced using the DDS method within the sampling range even under high relative observational errors. The SWAT model calibration performance using the CNNs, GLUE, and DDS methods are compared using R 2 and a set of efficiency metrics, including Nash-Sutcliffe, logarithmic Nash-Sutcliffe, Kling-Gupta, modified Kling-Gupta, and non-parametric Kling-Gupta scores, computed on the observed and simulated watershed responses. The best CNN-based calibrated set has scores of 0.71, 0.75, 0.85, 0.85, 0.86, and 0.91. The best DDS-based calibrated set has scores of 0.62, 0.69, 0.8, 0.77, 0.79, and 0.82. The best GLUE-based calibrated set has scores of 0.56, 0.58, 0.71, 0.7, 0.71, and 0.8. The scores above show that the CNN-based calibration leads to more accurate low and high streamflow predictions than the GLUE and DDS sets. Our research demonstrates that the proposed method has high potential to improve our current practice in calibrating large-scale integrated hydrologic models.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Predicting future well performance for environmental remediation design using deep learning

Here in this study, we developed a deep learning (DL) framework with a multi-channel three-dimensional convolutional neural network (MC3D-CNN) to predict well performance and thereby assist future environmental remediation design. Such prediction of extraction well performance at designated locations is critical for configuring pump-and-treat (P&T) well network design and operation, setting reasonable target closure dates for overall remedying, and estimating remedy costs. The framework is developed with operational and monitoring data routinely collected during P&T remedy operations, including well extraction and injection rates as well as in situ contaminant concentrations. Traditionally, the collected data were rarely used for purposes other than assessing past well performance and the accuracy of the conceptual site model. However, recent advances in data-driven computational approaches enable better use of the large datasets to inform future well performance, enhance site characterization, and improve remediation planning. In this study, we established a DL framework to integrate transient three-dimensional contaminant plumes and multiple aquifer properties (e.g., hydraulic conductivity and hydrostratigraphic maps) to identify characteristic patterns controlling and representing extraction well mass recovery, aiming at providing future mass recovery estimates for existing wells and candidate wells at any proposed locations. We evaluated our framework by using a realistic synthetic dataset generated from a well-calibrated flow and transport model used in the 200 West Area of the U.S. Department of Energy’s Hanford Site in southeastern Washington state. The multi-channel feature in our framework allows integration of various types and temporal densities of training datasets for DL model development. Overall, we found that the trained DL model achieved an accuracy of over 90% in ranking extraction well performance in validation datasets, and over 80% in predicting high-performance-ranking well locations. This data-informed approach provides a flexible tool to support adaptive site management, streamline decision-making, and potentially reduce remediation time and costs. Our DL framework can be used as a filtering tool to improve the current P&T network optimization design by reducing the number of candidate well locations.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

WavPool: A New Block for Deep Neural Networks

Modern deep neural networks comprise many operational layers, such as dense or convolutional layers, which are often collected into blocks. In this work, we introduce a new, wavelet-transform-based network architecture that we call the multi-resolution perceptron: by adding a pooling layer, we create a new network block, the WavPool. The first step of the multi-resolution perceptron is transforming the data into its multi-resolution decomposition form by convolving the input data with filters of fixed coefficients but increasing size. Following image processing techniques, we are able to make scale and spatial information simultaneously accessible to the network without increasing the size of the data vector. WavPool outperforms a similar multilayer perceptron while using fewer parameters, and outperforms a comparable convolutional neural network by ~ 10% on relative accuracy on CIFAR-10.

McDermott, Samuel D.↗