Biological Dynamics Enabling Training of Binary Recurrent Networks
We demonstrate a method for the training of binary activation recurrent neural networks
Chapman, George William↗
SEARCH · Engineering Papers
Results for “Recurrent 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.
Online evolutionary neural architecture search for multivariate non-stationary time series forecasting
Time series forecasting (TSF) is one of the most important tasks in data science. TSF models are usually pre-trained with historical data and then applied on future unseen datapoints. However, real-world time series data is usually non-stationary and models trained offline usually face problems from data drift. Models trained and designed in an offline fashion can not quickly adapt to changes quickly or be deployed in real-time. To address these issues, this work presents the Online NeuroEvolution-based Neural Architecture Search (ONE-NAS) algorithm, which is a novel neural architecture search method capable of automatically designing and dynamically training recurrent neural networks (RNNs) for online forecasting tasks. Without any pre-training, ONE-NAS utilizes populations of RNNs that are continuously updated with new network structures and weights in response to new multivariate input data. ONE-NAS is tested on real-world, large-scale multivariate wind turbine data as well as the univariate Dow Jones Industrial Average (DJIA) dataset. These results demonstrate that ONE-NAS outperforms traditional statistical time series forecasting methods, including online linear regression, fixed long short-term memory (LSTM) and gated recurrent unit (GRU) models trained online, as well as state-of-the-art, online ARIMA strategies. Additionally, results show that utilizing multiple populations of RNNs which are periodically repopulated provide significant performance improvements, allowing this online neural network architecture design and training to be successful.
97 MATHEMATICS AND COMPUTING↗
Trainable Gene Regulation Networks with Applications to Drosophila Pattern Formation
This chapter will very briefly introduce and review some computational experiments in using trainable gene regulation network models to simulate and understand selected episodes in the development of the fruit fly, Drosophila melanogaster. For details the reader is referred to the papers introduced below. It will then introduce a new gene regulation network model which can describe promoter-level substructure in gene regulation. As described in chapter 2, gene regulation may be thought of as a combination of cis-acting regulation by the extended promoter of a gene (including all regulatory sequences) by way of the transcription complex, and of trans-acting regulation by the transcription factor products of other genes. If we simplify the cis-action by using a phenomenological model which can be tuned to data, such as a unit or other small portion of an artificial neural network, then the full transacting interaction between multiple genes during development can be modelled as a larger network which can again be tuned or trained to data. The larger network will in general need to have recurrent (feedback) connections since at least some real gene regulation networks do. This is the basic modeling approach taken, which describes how a set of recurrent neural networks can be used as a modeling language for multiple developmental processes including gene regulation within a single cell, cell-cell communication, and cell division. Such network models have been called "gene circuits", "gene regulation networks", or "genetic regulatory networks", sometimes without distinguishing the models from the actual modeled systems.
Mjolsness, Eric↗
Forward variable selection enables fast and accurate dynamic system identification with Karhunen-Loève decomposed Gaussian processes
A promising approach for scalable Gaussian processes (GPs) is the Karhunen-Loève (KL) decomposition, in which the GP kernel is represented by a set of basis functions which are the eigenfunctions of the kernel operator. Such decomposed kernels have the potential to be very fast, and do not depend on the selection of a reduced set of inducing points. However KL decompositions lead to high dimensionality, and variable selection thus becomes paramount. This paper reports a new method of forward variable selection, enabled by the ordered nature of the basis functions in the KL expansion of the Bayesian Smoothing Spline ANOVA kernel (BSS-ANOVA), coupled with fast Gibbs sampling in a fully Bayesian approach. It quickly and effectively limits the number of terms, yielding a method with competitive accuracies, training and inference times for tabular datasets of low feature set dimensionality. Theoretical computational complexities are O ( N P 2 ) in training and O ( P ) per point in inference, where N is the number of instances and P the number of expansion terms. The inference speed and accuracy makes the method especially useful for dynamic systems identification, by modeling the dynamics in the tangent space as a static problem, then integrating the learned dynamics using a high-order scheme. The methods are demonstrated on two dynamic datasets: a ‘Susceptible, Infected, Recovered’ (SIR) toy problem, along with the experimental ‘Cascaded Tanks’ benchmark dataset. Comparisons on the static prediction of time derivatives are made with a random forest (RF), a residual neural network (ResNet), and the Orthogonal Additive Kernel (OAK) inducing points scalable GP, while for the timeseries prediction comparisons are made with LSTM and GRU recurrent neural networks (RNNs) along with the SINDy package.
Hayes, Kyle↗
Spatio-Temporal Anomaly Detection with Graph Networks for Data Quality Monitoring of the Hadron Calorimeter
The Compact Muon Solenoid (CMS) experiment is a general-purpose detector for high-energy collision at the Large Hadron Collider (LHC) at CERN. It employs an online data quality monitoring (DQM) system to promptly spot and diagnose particle data acquisition problems to avoid data quality loss. In this study, we present a semi-supervised spatio-temporal anomaly detection (AD) monitoring system for the physics particle reading channels of the Hadron Calorimeter (HCAL) of the CMS using three-dimensional digi-occupancy map data of the DQM. We propose the GraphSTAD system, which employs convolutional and graph neural networks to learn local spatial characteristics induced by particles traversing the detector and the global behavior owing to shared backend circuit connections and housing boxes of the channels, respectively. Recurrent neural networks capture the temporal evolution of the extracted spatial features. We validate the accuracy of the proposed AD system in capturing diverse channel fault types using the LHC collision data sets. The GraphSTAD system achieves production-level accuracy and is being integrated into the CMS core production system for real-time monitoring of the HCAL. We provide a quantitative performance comparison with alternative benchmark models to demonstrate the promising leverage of the presented system.
43 PARTICLE ACCELERATORS↗
Neural Networks for Rapid Design and Analysis
Artificial neural networks have been employed for rapid and efficient dynamics and control analysis of flexible systems. Specifically, feedforward neural networks are designed to approximate nonlinear dynamic components over prescribed input ranges, and are used in simulations as a means to speed up the overall time response analysis process. To capture the recursive nature of dynamic components with artificial neural networks, recurrent networks, which use state feedback with the appropriate number of time delays, as inputs to the networks, are employed. Once properly trained, neural networks can give very good approximations to nonlinear dynamic components, and by their judicious use in simulations, allow the analyst the potential to speed up the analysis process considerably. To illustrate this potential speed up, an existing simulation model of a spacecraft reaction wheel system is executed, first conventionally, and then with an artificial neural network in place.
Sparks, Dean W., Jr.↗
TorchBraid: High-Performance Layer-Parallel Training of Deep Neural Networks with MPI and GPU Acceleration
TorchBraid is a high-performance implementation of layer-parallel training for deep neural networks (DNNs) supporting MPI-based parallelism and GPU acceleration. Layer-parallel training has been developed to overcome the serialization inherent in forward and backward propagation of DNNs that limits utilization of computational resources in the strong scaling limit. To achieve this, TorchBraid integrates the PyTorch neural network framework with the state-of-the-art XBraid time-parallel library. Furthermore, this article presents the use and performance of TorchBraid, in addition to solutions for overcoming the algorithmic challenges inherent in combining automatic differentiation with layer-parallel. Results are presented with and without GPU acceleration for the Tiny ImageNet and MNIST image classification data sets, as well as recurrent neural networks. Overall, TorchBraid enables fast training of DNNs, both in a strong and weak scaling context. In addition to the TorchBraid software, several new advances in applying layer-parallel algorithms are detailed. Integration of layer-parallel with data-parallel algorithms is presented for the first time, showing the computational advantages of the combination. Standard deep learning techniques, like batch-normalization, are developed for layer-parallel training. Finally, a new approach combining layer-parallel with spatial coarsening in order to accelerate training for 3D image classification shows roughly a 10× speedup over serial execution.
Layer-parallel↗
Data-driven based coordinated smart inverter control for distributed energy resources
Smart inverters (SI) for distributed energy resources (DER) are becoming popular since they have the ability to stabilize as well as restore the voltage and frequency of power systems. Aiming at establishing the mathematical models combined with SI control methods, multiple optimization methods are developed. However, the computational complexity of solving such a mathematical model with various uncertainties limits the real-time application of the SI control. To conquer this challenge, a data-driven-based SI control approach is developed to achieve coordinated control in the high penetration DER system. First, an optimization problem for maximizing the active power generation and minimizing the power loss is designed using the Volt/VAR control. To reduce the time consumption, the recurrent neural network (RNN) is proposed to model the relationship between the uncertainties and control actions during the offline site. The RNN with different sub-structures such as the long short-term memory cell and gated recurrent unit cell are included to enrich the diversity of features. In the last stage, different experiment comparisons, including multiple uncertainties maps and stateof- art machine learning methods, are conducted to verify the effectiveness of the proposed method based on the IEEE 123 bus power system. The results demonstrate that the proposed method can effectively achieve a rapid and coordinated control with a lower error rate.
Qiu, Wei↗
Neural-network learning of SPOD latent dynamics
Here, we aim to reconstruct the latent space dynamics of high dimensional, quasi-stationary systems using model order reduction via the spectral proper orthogonal decomposition (SPOD). The proposed method is based on three fundamental steps: in the first, once that the mean flow field has been subtracted from the realizations (also referred to as snapshots), we compress the data from a high-dimensional representation to a lower dimensional one by constructing the SPOD latent space; in the second, we build the time-dependent coefficients by projecting the snapshots containing the fluctuations onto the SPOD basis and we learn their evolution in time with the aid of recurrent neural networks; in the third, we reconstruct the high-dimensional data from the learnt lower -dimensional representation. The proposed method is demonstrated on two different test cases, namely, a compressible jet flow, and a geophysical problem known as the Madden-Julian Oscillation. An extensive comparison between SPOD and the equivalent POD-based counterpart is provided and differences between the two approaches are highlighted. The numerical results suggest that the proposed model is able to provide low rank predictions of complex statistically stationary data and to provide insights into the evolution of phenomena characterized by specific range of frequencies. The comparison between POD and SPOD surrogate strategies highlights the need for further work on the characterization of the interplay of error between data reduction techniques and neural network forecasts.
97 MATHEMATICS AND COMPUTING↗
Using Neural Networks for Low Energy Reconstruction and Neutron Identification in the MicroBooNE LArTPC
Identifying and reconstructing final-state neutrons from neutrino interactions in Liquid Argon Time Projection Chambers (LArTPCs) will enhance future oscillation measurements by recovering missing energy and improving neutrino interaction channel identification. However, neutrons are challenging to reconstruct as the majority leave only small, isolated charge signatures known as blips. Here we present initial efforts to identify neutrons in the MicroBooNE LArTPC with low energy protons from neutron-argon inelastic interactions that present as blips below the traditional tracking threshold in the TPC. Unlike for tracks, there is no algorithmic method to determine direction for blips since they span only a few wires. Therefore, we developed and trained a Recurrent Neural Network (RNN) to reconstruct the directionality of proton-induced blips, allowing us to separate signal from background by selecting blips that point back to the neutrino vertex. The model achieves a preliminary average angular resolution of 17 degrees when tested on a simulated sample of protons over 6 MeV in kinetic energy. This novel tool will enhance neutron detection in LArTPCs and expand a broad range of other low-energy physics searches such as for solar and supernova neutrinos.
Silva, Liani Isabel [Unlisted, US]↗
Evapotranspiration partitioning estimates from 8 methods from 47 NEON sites, 2019-2021
This dataset provides daily estimates of evapotranspiration (ET) and the transpiration-to-evapotranspiration ratio (T/ET) across 47 terrestrial National Ecological Observatory Network (NEON) sites spanning diverse environmental and biome conditions in the United States across three years of data (2019-2021). Daily ET is reported in both energy units (MJ m⁻² day⁻¹) and equivalent water depth (mm day⁻¹), assuming a constant latent heat of vaporization of 2.45 MJ/kg. The primary method uses a hybrid recurrent neural network–Penman–Monteith framework (RNN-PM), which integrates physically based surface energy balance constraints with data-driven learning to partition ET into transpiration and evaporation components. Model inputs include in situ meteorological observations (air temperature, vapor pressure deficit, wind speed, and radiation) combined with satellite-derived land surface temperature, leaf area index, and soil moisture. For benchmarking and uncertainty assessment, T/ET estimates from seven additional models are included: Priestley-Taylor Jet Propulsion Laboratory (PT-JPL), Penman-Monteith (P-M), Two-Source Energy Balance (TSEB), Support Vector Regression (SVR), and Categorical Boosting (CatBoost), among others—spanning empirical, machine-learning, and process-based approaches (see methods section or linked publication for detailed descriptions). Data Package Contents: The dataset a csv files containing daily ET and T/ET estimates for each site and model, along with associated metadata files these variables. Data can be accessed using common spreadsheet software (e.g., Microsoft Excel, LibreOffice) or programming environments such as R or Python. Together, these data support cross-site comparisons of ecosystem water use, evaluation of ET partitioning methods, and development of improved land–atmosphere exchange models.
EARTH SCIENCE > ATMOSPHERE↗
From Coexpression to Coregulation: An Approach to Inferring Transcriptional Regulation Among Gene Classes from Large-Scale Expression Data
We provide preliminary evidence that existing algorithms for inferring small-scale gene regulation networks from gene expression data can be adapted to large-scale gene expression data coming from hybridization microarrays. The essential steps are (I) clustering many genes by their expression time-course data into a minimal set of clusters of co-expressed genes, (2) theoretically modeling the various conditions under which the time-courses are measured using a continuous-time analog recurrent neural network for the cluster mean time-courses, (3) fitting such a regulatory model to the cluster mean time courses by simulated annealing with weight decay, and (4) analysing several such fits for commonalities in the circuit parameter sets including the connection matrices. This procedure can be used to assess the adequacy of existing and future gene expression time-course data sets for determining transcriptional regulatory relationships such as coregulation.
Mjolsness, Eric↗
Construction of generalized quasilinear diffusion coefficient using neural networks with physical restrictions
The quasilinear diffusion coefficient (D QL ) derived from our machine learning framework shows comparable trends with the ground truth D QL obtained from GENRAY-CQL3D simulations. Additionally, for the strong absorption cases, the radial current drive profiles generated using the D QL from our model exhibit consistent behavior with those obtained from the original simulation. These findings indicate the potential of our surrogate modeling approach with physical restrictions to replicate key wave–plasma interaction characteristics while reducing computational costs. Traditionally, calculating D QL for wave–particle interactions relies on computationally intensive wave simulations coupled with Fokker–Planck solvers. To address this challenge, we developed a machine learning-based surrogate model with physical restrictions derived from cold plasma theory and bounce-averaged damping effects. First, we establish the propagation domain of Lower Hybrid Waves in the (N∥, ρ) space by identifying the accessibility limit and determining the upper and lower bounds of N∥ using the Potential Power Deposition (PPD) method. Subsequently, leveraging a database constructed using Latin hypercube sampling alongside the underlying physical restrictions (e.g. PPD), machine learning methods including U-Net and Recurrent Neural Networks are employed to design a physics-restricted machine learning framework capable of reconstructing D QL .
70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
Monitoring of Temperature Measurements for Different Flow Regimes in Water and Galinstan with Long Short-Term Memory Networks and Transfer Learning of Sensors
Temperature sensing is one of the most common measurements of a nuclear reactor monitoring system. The coolant fluid flow in a reactor core depends on the reactor power state. We investigated the monitoring and estimation of the thermocouple time series using machine learning for a range of flow regimes. Measurement data were obtained, in two separate experiments, in a flow loop filled with water and with liquid metal Galinstan. We developed long short-term memory (LSTM) recurrent neural networks (RNNs) for sensor predictions by training on the sensor’s own prior history, and transfer learning LSTM (TL-LSTM) by training on a correlated sensor’s prior history. Sensor cross-correlations were identified by calculating the Pearson correlation coefficient of the time series. The accuracy of LSTM and TL-LSTM predictions of temperature was studied as a function of Reynolds number (Re). The root-mean-square error (RMSE) for the test segment of time series of each sensor was shown to linearly increase with Re for both water and Galinstan fluids. Using linear correlations, we estimated the range of values of Re for which RMSE is smaller than the thermocouple measurement uncertainty. For both water and Galinstan fluids, we showed that both LSTM and TL-LSTM provide reliable estimations of temperature for typical flow regimes in a nuclear reactor. The LSTM runtime was shown to be substantially smaller than the data acquisition rate, which allows for performing estimation and validation of sensor measurements in real time.
Pantopoulou, Stella↗
Deep neural operators can predict the real-time response of floating offshore structures under irregular waves
The use of neural operators in a digital twin model of an offshore floating structure holds the potential for a significant shift in the prediction of structural responses and health monitoring, offering valuable real-time control insights. In this work, we investigate the effectiveness of three neural operators, namely the deep operator network (DeepONet), the Fourier neural operator (FNO), and the Wavelet neural operator (WNO), to accurately capture the responses of a floating structure under six different sea state codes (3 − 8) based on the wave characteristics described by the World Meteorological Organization (WMO). To further enhance the accuracy of the vanilla architecture of the neural operators, novel extensions, such as wavelet-DeepONet and self-adaptive WNO, are proposed in this paper. The results demonstrate that these high-precision neural operators can deliver structural responses more efficiently, up to two orders of magnitude faster than a dynamic analysis using conventional numerical solvers. Additionally, compared to gated recurrent units (GRUs), a commonly used recurrent neural network for time-series estimation, neural operators are both more accurate and efficient, especially in situations with limited data availability. Taken together, our study shows that FNO outperforms all other operators for approximating the mapping of one input functional space to the output space as well as for responses that have small bandwidth of the frequency spectrum. Conversely, DeepONet, with historical states, proves most accurate in learning the mapping of multiple input functions to the output space and capturing responses within a broad frequency spectrum.
97 MATHEMATICS AND COMPUTING↗
DroughtCast: A Machine Learning Forecast of the United States Drought Monitor
Drought is one of the most ecologically and economically devastating natural phenomena affecting the United States, causing the U.S. economy billions of dollars in damage, and driving widespread degradation of ecosystem health. Many drought indices are implemented to monitor the current extent and status of drought so stakeholders such as farmers and local governments can appropriately respond. Methods toforecast drought conditions weeks to months in advance are less common but would provide a more effective early warning system to enhance drought response, mitigation, and adaptation planning. To resolve this issue, we introduce DroughtCast, a machine learning framework for forecasting the United States Drought Monitor (USDM). DroughtCast operates on the knowledge that recent anomalies in hydrology and meteorology drive future changes in drought conditions. We use simulated meteorology and satellite observed soil moisture as inputs into a recurrent neural network to accurately forecast the USDM between 1 and 12 weeks into the future. Our analysis shows that precipitation, soil moisture, and temperature are the most important input variables when forecasting future drought conditions. Additionally, a case study of the 2017 Northern Plains Flash Drought shows that DroughtCast was able to forecast a very extreme drought event up to 12 weeks before its onset. Given the favorable forecasting skill of the model, DroughtCast may provide a promising tool for land managers and local governments in preparing for and mitigating the effects of drought.
Machine Learning↗
Implementation of AI/DEEP learning disruption predictor into a plasma control system
Abstract This paper reports on advances in the state‐of‐the‐art deep learning disruption prediction models based on the Fusion Recurrent Neural Network (FRNN) originally introduced in a 2019 NATURE publication [ https://doi.org/10.1038/s41586‐019‐1116‐4 ]. In particular, the predictor now features not only the “disruption score,” as an indicator of the probability of an imminent disruption, but also a “sensitivity score” in real time to indicate the underlying reasons for the imminent disruption. This adds valuable physics interpretability for the deep learning model and can provide helpful guidance for control actuators now implemented into a modern plasma control system (PCS). The advance is a significant step forward in moving from modern deep learning disruption prediction to real‐time control and brings novel AI‐enabled capabilities relevant for application to the future burning plasma ITER system. Our analyses use large amounts of data from JET and DIII‐D vetted in the earlier NATURE publication. In addition to “when” a shot is predicted to disrupt, this paper addresses reasons “why” by carrying out sensitivity studies. FRNN is accordingly extended to use more channels of information, including measured DIII‐D signals such as (i) the “n1rms” signal that is correlated with the n = 1 modes with finite frequency, including neoclassical tearing mode and sawtooth dynamics; (ii) the bolometer data indicative of plasma impurity control; and (iii) “q‐min”—the minimum value of the safety factor relevant to the key physics of kink modes. The additional channels and interpretability features expand the ability of the deep learning FRNN software to provide information about disruption subcategories as well as more precise and direct guidance for the actuators in a PCS.
70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
A deep learning upscaling framework: Reactive transport and mineral precipitation in fracture-matrix systems
Pore-scale modeling has limited applicability at large scales due to its high computational cost. One common approach to upscale pore-scale models is the use of effective medium theories, which homogenize small-scale features in a porous structure and characterize the medium by macroscale properties (e.g., permeability) and equations (e.g., Darcy’s law). However, there are classes of physical processes for which effective medium approximations may become inaccurate, e.g., mineral precipitation and clogging during reactive transport. We have developed a deep learning upscaling framework, in which pore-scale modeling is directly employed in macroscale systems, without relying on effective medium approximations. The upscaling framework is first developed for general multiscale systems and then applied to modeling reactive transport with mineral precipitation in the altered layer in fracture-matrix structures. Solute transport from the fractures to the matrix is modeled as a wall boundary condition for the fractures, which, in turn, is predicted by recurrent neural networks using the concentration histories at the fracture-matrix boundary. Specifically, we consider a meter-scale fracture network embedded in sandstones, where the smallest feature is at the micron scale. Here the proposed framework allows us to span five orders of magnitude in length scales by capturing mineral precipitation in the altered layer of the rock matrix at the pore scale across the entire meter-scale fracture network.
42 ENGINEERING↗