Search NASA⌕ Search

SEARCH · Search NASA

Results for “Adaptive convolution.”

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 127 records · Page 7
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Analyzing inference workloads for spatiotemporal modeling

Ensuring power grid resiliency, forecasting climate conditions, and optimization of transportation infrastructure are some of the many application areas where data is collected in both space and time. Spatiotemporal modeling is about modeling those patterns for forecasting future trends and carrying out critical decision-making by leveraging machine learning/deep learning. Once trained offline, field deployment of trained models for near real-time inference could be challenging because performance can vary significantly depending on the environment, available compute resources and tolerance to ambiguity in results. Users deploying spatiotemporal models for solving complex problems can benefit from analytical studies considering a plethora of system adaptations to understand the associated performance-quality trade-offs. To facilitate the co-design of next-generation hardware architectures for field deployment of trained models, it is critical to characterize the workloads of these deep learning (DL) applications during inference and assess their computational patterns at different levels of the execution stack. In this paper, we develop several variants of deep learning applications that use spatiotemporal data from dynamical systems. We study the associated computational patterns for inference workloads at different levels, considering relevant models (Long short-term Memory, Convolutional Neural Network and Spatio-Temporal Graph Convolution Network), DL frameworks (Tensorflow and PyTorch), precision (FP16, FP32, AMP, INT16 and INT8), inference runtime (ONNX and AI Template), post-training quantization (TensorRT) and platforms (Nvidia DGX A100 and Sambanova SN10 RDU). Overall, our findings indicate that although there is potential in mixed-precision models and post-training quantization for spatiotemporal modeling, extracting efficiency from contemporary GPU systems might be challenging. Instead, co-designing custom accelerators by leveraging optimized High Level Synthesis frameworks (such as SODA High-Level Synthesizer for customized FPGA/ASIC targets) can make workload-specific adjustments to enhance the efficiency.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Exploring Randomly Wired Neural Networks for Climate Model Emulation

Exploring the climate impacts of various anthropogenic emissions scenarios is key to making informed decisions for climate change mitigation and adaptation. State-of-the-art Earth system models can provide detailed insight into these impacts but have a large associated computational cost on a per-scenario basis. This large computational burden has driven recent interest in developing cheap machine learning models for the task of climate model emulation. In this paper, we explore the efficacy of randomly wired neural networks for this task. We describe how they can be constructed and compare them with their standard feedforward counterparts using the ClimateBench dataset. Specifically, we replace the serially connected dense layers in multilayer perceptrons, convolutional neural networks, and convolutional long short-term memory networks with randomly wired dense layers and assess the impact on model performance for models with 1 million and 10 million parameters. We find that models with less-complex architectures see the greatest performance improvement with the addition of random wiring (up to 30.4% for multilayer perceptrons). Furthermore, of 24 different model architecture, parameter count, and prediction task combinations, only one had a statistically significant performance deficit in randomly wired networks relative to their standard counterparts, with 14 cases showing statistically significant improvement. We also find no significant difference in prediction speed between networks with standard feedforward dense layers and those with randomly wired layers. These findings indicate that randomly wired neural networks may be suitable direct replacements for traditional dense layers in many standard models.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS2022

Multi-Mission Terrain Classifier for Safe Rover Navigation and Automated Science

We previously presented Soil Property and Object Classification (SPOC), a machine learning-based terrain classifier for Mars rovers, for automatically segmenting rover images by its surface type such as sand and bedrock. This paper presents a number of practical improvements to pave the way for potential future onboard deployment. First, we achieved 97.0% overall pixel accuracy, evaluated against the classification generated by human experts on images from Mars Science Laboratory (MSL) missions. The substantial increase in accuracy was primarily enabled by the sheer volume of data used for training; we created a new large-scale dataset of Martian terrain labels, namely AI4Mars, which contains more than 400k labels contributed by citizen scientists for 50k images taken by the Mars Exploration Rovers (MER) and Mars Science Laboratory (MSL) rover. Second, we demonstrated that SPOC can quickly adapt to a new mission landed on a previously unseen site. Specifically, we pretrained a model with MER and MSL data from the AI4Mars dataset and then adapted to the Mars 2020 Rover (M2020) by feeding a small volume of data between Sol 0 and 157; the adapted model was tested on Sol 200-203 and resulted in 84.2% overall pixel accuracy and 93.4% reliability (recall) for detecting sand, the most concerning class for rover’s traversability. Third, we found that pretraining can substantially mitigate the decline of accuracy over time. We showed that the performance of a SPOC model pretrained with the ImageNet dataset and then trained by MSL images only up to Sol 390 remains comparable to a model trained by images up to Sol 1689 on the test data after Sol 1689. Fourth, we reimplemented SPOC with a light-weight convolutional neural network (CNN), MobileNetV2, which typically runs within tens of milliseconds (ms) on mobile processors such as Qualcomm’s Snapdragon. Finally, we released the AI4Mars dataset to the public to encourage open innovation.

Ono, Masahiro↗
NASA NTRS2016

On Holo-Hilbert Spectral Analysis: A Full Informational Spectral Representation for Nonlinear and Non-Stationary Data

The Holo-Hilbert spectral analysis (HHSA) method is introduced to cure the deficiencies of traditional spectral analysis and to give a full informational representation of nonlinear and non-stationary data. It uses a nested empirical mode decomposition and Hilbert-Huang transform (HHT) approach to identify intrinsic amplitude and frequency modulations often present in nonlinear systems. Comparisons are first made with traditional spectrum analysis, which usually achieved its results through convolutional integral transforms based on additive expansions of an a priori determined basis, mostly under linear and stationary assumptions. Thus, for non-stationary processes, the best one could do historically was to use the time- frequency representations, in which the amplitude (or energy density) variation is still represented in terms of time. For nonlinear processes, the data can have both amplitude and frequency modulations (intra-mode and inter-mode) generated by two different mechanisms: linear additive or nonlinear multiplicative processes. As all existing spectral analysis methods are based on additive expansions, either a priori or adaptive, none of them could possibly represent the multiplicative processes. While the earlier adaptive HHT spectral analysis approach could accommodate the intra-wave nonlinearity quite remarkably, it remained that any inter-wave nonlinear multiplicative mechanisms that include cross-scale coupling and phase-lock modulations were left untreated. To resolve the multiplicative processes issue, additional dimensions in the spectrum result are needed to account for the variations in both the amplitude and frequency modulations simultaneously. HHSA accommodates all the processes: additive and multiplicative, intra-mode and inter-mode, stationary and nonstationary, linear and nonlinear interactions. The Holo prefix in HHSA denotes a multiple dimensional representation with both additive and multiplicative capabilities.

Huang, Norden E.↗
NASA NTRS2022

Spatial Grid-Based Object Localization from A Single Passive Sensor: A Deep Learning-Integrated Approach

Ongoing efforts at NASA’s Langley Research Center have produced a single passive sensor system for detecting ground objects and pinpointing their real-world location to a desired level of precision. The Langley center serves as a test range for unmanned aerial systems (UAS) and real-time knowledge about the location of people on campus is needed to inform least-risk UAS flight operations. The proposed system provides this knowledge through a camera combined with a convolutional neural network and an algorithm that projects an imaginary grid of square cells from the ground plane onto the perspective view of the camera. The position of detected objects on the camera’s projected grid determines their location in the real-world. The imaginary grid is easily mapped to a universal coordinate system, such as longitude and latitude, to provide both relative and absolute positional information of the detected objects. This simple system is shown to be accurate and effective, with decisive advantages over alternative multi-sensor and active sensor approaches. Extensions to the system are described to allow adaptation to a variety of other use cases.

object localization↗
DOE OSTI2025

Neural Network Analysis of Nuclear Magnetic Resonance and Infrared Spectra

Nuclear magnetic resonance (NMR) spectroscopy and infrared (IR) spectroscopy are powerful chemical characterization techniques with broad general usage. However, the manual evaluation of the resulting spectra is time-consuming and requires significant expertise, preventing insights from being used in real-time applications. With recent advances in computation and artificial intelligence (AI), new tools are available for automating spectral interpretation. In this work, machine learning (ML) algorithms using 1-dimensional convolutional neural networks (CNNs) were applied to identify common functional groups from spectral information. Raw spectra were collected virtually from the Human Metabolome Database (HMDB) and National Institute of Standards and Technology (NIST) Chemistry WebBook and processed into a suitable standard. Algorithm design was tailored to best fit the nature of the problem, with built-in flexibility to accommodate relevant parameters beyond the raw spectral input, specifically solvent identity and magnetic frequency for NMR. The predictive capability of the algorithm in identifying functional groups is displayed in several examples. This methodology has been compiled into a code repository and could easily be modified to adapt alternative data sources, including other spectrum types. To mitigate overfitting, a common problem in mathematical modeling where overfamiliarity with training data produces trends that are not representative of the general data, a novel metric was developed, referred to as Accufit. Accufit includes a parameter that penalizes substantial differences in the training accuracy and the accuracy of an independent validation set. Examples are presented showing the effectiveness of Accufit in maintaining the model’s predictive capability while controlling the overfitting when used as a custom metric for hyperparameter tuning.

Sturgill, James↗
DOE OSTI2023

Lorentz group equivariant autoencoders

Abstract There has been significant work recently in developing machine learning (ML) models in high energy physics (HEP) for tasks such as classification, simulation, and anomaly detection. Often these models are adapted from those designed for datasets in computer vision or natural language processing, which lack inductive biases suited to HEP data, such as equivariance to its inherent symmetries. Such biases have been shown to make models more performant and interpretable, and reduce the amount of training data needed. To that end, we develop the Lorentz group autoencoder (LGAE), an autoencoder model equivariant with respect to the proper, orthochronous Lorentz group $$\textrm{SO}^+(3,1)$$ SO + ( 3 , 1 ) , with a latent space living in the representations of the group. We present our architecture and several experimental results on jets at the LHC and find it outperforms graph and convolutional neural network baseline models on several compression, reconstruction, and anomaly detection metrics. We also demonstrate the advantage of such an equivariant model in analyzing the latent space of the autoencoder, which can improve the explainability of potential anomalies discovered by such ML models.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2023

Explainable AI (XAI)-driven vibration sensing scheme for surface quality monitoring in a smart surface grinding process

Local Interpretable and Model-agnostic Explanation (LIME), an explainable artificial intelligence (XAI) approach is adapted to identify the globally important time-frequency bands for predicting average surface roughness (Ra) in a smart grinding process. The smart grinding setup consisted of a Supertech CNC precision surface grinding machine, instrumented with a Dytran piezoelectric accelerometer attached to the tailstock quill along the tangential direction (Y-axis). For every grinding pass, vibration signatures were captured, and the ground truth surface roughness values were recorded using a Mahr Marsurf M300C portable surface roughness profilometer. The roughness values ranged from 0.06 to 0.14 microns over the complete set of experiments. Time-frequency domain spectrogram frames were extracted for each of the vibration signals collected during the grinding process. Convolutional Neural Networks (CNNs) were modeled to predict the surface roughness based on these spectrogram frames and their image augmentations. The best CNN model was able to predict the roughness values with an overall R2-score of 0.95, training R2-score of 0.99, and testing R2-score of 0.81 with only 80 sets of vibration signals corresponding to 4 experiments with 20 trials each. Although the data size is not large enough to guarantee such performance metrics in real-world scenarios, one can extract statistically consistent explanations underlying the relationships these complex deep learning models capture. Further, the LIME methodology was implemented on the developed surface roughness CNN model to identify the important time-frequency bands (i.e., the superpixels of a spectrogram) influencing the predictions. Based on the identified important regions on the spectrogram frames, the corresponding frequency characteristics were determined that influence the surface roughness predictions. The important frequency range based on LIME results was approximately 11.7 to 19.1 kHz. The power of XAI was demonstrated by cutting down the sampling rate from 160 kHz to 30, 20, 10, and 5 kHz based on the important frequency range and considering Nyquist criteria. Separate CNN models were developed for these ranges by only extracting time-frequency contents below their corresponding Nyquist cut-offs. A proper data acquisition strategy is proposed by comparing the model performances to argue the selection of a sufficient sampling rate to capture the grinding process successfully and robustly.

42 ENGINEERING↗
DOE OSTI2023

Reprint of: Explainable AI (XAI)-driven vibration sensing scheme for surface quality monitoring in a smart surface grinding process

Local Interpretable and Model-agnostic Explanation (LIME), an explainable artificial intelligence (XAI) approach is adapted to identify the globally important time-frequency bands for predicting average surface roughness (Ra) in a smart grinding process. The smart grinding setup consisted of a Supertech CNC precision surface grinding machine, instrumented with a Dytran piezoelectric accelerometer attached to the tailstock quill along the tangential direction (Y-axis). For every grinding pass, vibration signatures were captured, and the ground truth surface roughness values were recorded using a Mahr Marsurf M300C portable surface roughness profilometer. The roughness values ranged from 0.06 to 0.14 microns over the complete set of experiments. Time-frequency domain spectrogram frames were extracted for each of the vibration signals collected during the grinding process. Convolutional Neural Networks (CNNs) were modeled to predict the surface roughness based on these spectrogram frames and their image augmentations. The best CNN model was able to predict the roughness values with an overall R2-score of 0.95, training R2-score of 0.99, and testing R2-score of 0.81 with only 80 sets of vibration signals corresponding to 4 experiments with 20 trials each. Although the data size is not large enough to guarantee such performance metrics in real-world scenarios, one can extract statistically consistent explanations underlying the relationships these complex deep learning models capture. Further, the LIME methodology was implemented on the developed surface roughness CNN model to identify the important time-frequency bands (i.e., the superpixels of a spectrogram) influencing the predictions. Based on the identified important regions on the spectrogram frames, the corresponding frequency characteristics were determined that influence the surface roughness predictions. The important frequency range based on LIME results was approximately 11.7 to 19.1 kHz. The power of XAI was demonstrated by cutting down the sampling rate from 160 kHz to 30, 20, 10, and 5 kHz based on the important frequency range and considering Nyquist criteria. Separate CNN models were developed for these ranges by only extracting time-frequency contents below their corresponding Nyquist cut-offs. A proper data acquisition strategy is proposed by comparing the model performances to argue the selection of a sufficient sampling rate to capture the grinding process successfully and robustly.

47 OTHER INSTRUMENTATION↗
NASA NTRS2018

The Utilization Profiles of the CCSDS Unified Space Link Protocol (USLP)

The purpose of this paper is to identify the utilization profiles for interfacing the Data Protocol Sublayer using the Unified Space Link Protocols (USLP) (reference 1) with the space link coding procedures as specified in the CCSDS Coding & Synchronization Blue Books (references 2 through 5), used in both telecommand and telemetry applications. This paper describes how the USLP Protocol utilizes the coding and synchronization sublayer to support: a. Direct to Earth (DTE) telemetry links for engineering and science data b. Direct to Earth (DTE) telemetry links for very high rate science data c. Direct from Earth (DFE) command, sequencing and flight software loads d. Space to Space Links (Proximity) utilized by orbiters for data exchange to/from surface bound assets. The CCSDS has divided the functions of the Data Link Layer into two sublayers: the Data Link Protocol Sublayer (DLP-SL) and the Coding and Synchronization Sublayer (CS-SL). The Data Link Protocol Sublayer (DLP-SL) interfaces to the users, accepting the data that is to be transported, on the sending side of the link, and delivering that data on the receiving end. The Transfer Frame is the data unit that is transferred across the Data Link Protocol Sublayer and the Coding and Synchronization Sublayer boundary. The Coding and Synchronization Sublayer (CS-SL) provides the encoding, randomization, and frame synchronization functions that prepares the USLP Transfer Frame for transport across the space link. The CS-SL is divided into 2 processes: 1) The Frame Interface Processes (FIP) performs the interface functions required to prepare the data for delivery to the Coding/Decoding Process (CDP). This process includes prepending a Frame Start Marker to the provided frame, when management has designated that the frame is not to be aligned to the codeblock or when there is no block code used. 2) The Coding/Decoding Process (CDP) performs the forward error correction processes that are used to optimize the performance of the link and minimize the error rate. The CDP creates the symbol stream that is delivered to the Physical Layer. The transfer of the USLP transfer frames across different types of space links is the focus of this paper. The Protocol Data Unit (PDU) that is passed in both directions between the Data Link Protocol Sublayer (DLP-SL) and Coding and Synchronization Sublayer (CS-SL) is the transfer frame. The USLP frame structure provides flexibility that can be constrained by the functions utilized within the CS-SL that prepare the transfer frame for transit. For example, the USLP transfer frame contains a length field that enables the frame to be of variable length but CS-SL under certain conditions may constrain the frame to be fixed in length. This paper describes 5 operational modes available for use by the Data Link Layer to provide data exchange across the USLP space link. These modes are different because different operational requirements apply to vastly different types of space links and thus the communications implementation requirements differ. The environmental issues include the power or energy available, the distance between the end points of the link, the complexity of the equipment available at those end points, the atmospheric conditions and radiometric frequency selection. The CS-SL utilizes different forward error correcting codes supported by specific operational modes to configure the data for transit. This paper describes all of the operational modes in a series of data models which decompose the functionality between the Data Link Protocol Sublayer and the Coding and Synchronization sublayer. The operational modes described are: 1. Uncoded Mode: has been used for short links that contain significant available power to provide an acceptable frame error rate. The frames in this mode can be variable in length and typically use an error detection algorithm (i.e., CRC) to determine if there are errors in the received frame. 2. Convolutional Only Mode: is currently the prime forward error correction coding used for the proximity links. The frames in this mode can be variable in length and typically use an error detection algorithm (i.e., CRC) to determine if there are errors in the received frame. 3. Variable Length Frame Aligned to Variable Length Codeblock (TC): is used for Direct from Earth links were power levels are high and the simple, least complex code i.e., the BCH code is used. This mode has been in use since the early 1970s. The BCH code is a short code and the decoder is easy to implement. 4. Fixed Length Frame Aligned to Fixed Length Codeblock (AOS/TM): was introduced when the concatenated Convolutional and Reed-Solomon Code was formulated to provide significant reduction in link data error rate and the ability to determine if there was an error in the decoded codeblock. The frame is aligned to the codeblock so that there is a one to one relationship of frame errors to codeblock errors without additional error detection coding being added. This mode requires the protocol frames to be the exact size of the message portion of the codeblock. 5. Frames Unaligned to Fixed Length Codeblocks (Currently used for very high rates and space to space links): This mode is currently used for missions that have a very high data rate that can be controlled adaptively as the environment changes and as the next generation operating mode for the proximity link. This mode from a coded data stream point of view is exactly like that described in 4. above, except that the frame need not be aligned to the codeblock. There is no requirement on frame length when using this mode. Thus when using USLP it can be used to support links that require short or long frames. There is also no mandatory requirement that frames cannot be separated by idle data reducing the tight data rate connection requirements between the data link protocol sublayer and the coding & synchronization sublayer. In conclusion, how these operational modes can be put to use in mission operational scenarios is described for Direct from Earth links (DFE), Direct to Earth links (DTE), and Proximity links.

Greenberg, E.↗
DOE OSTI2019

Pairwise Association of Seismic Arrivals with Convolutional Neural Networks

Correctly determining the association of seismic phases across a network is crucial for developing accurate earthquake catalogs. Nearly all established methods use travel time information as the main criterion for determining associations, and in problems in which earthquake rates are high and many false arrivals are present, many standard techniques may fail to resolve the problem accurately. As an alternative approach, in this work we apply convolutional neural networks (CNNs) to the problem of associations; we train CNNs to read earthquake waveform arrival pairs between two stations and predict the binary classification of whether the two waveforms are from a common source or different sources. Applying the method to a large training dataset of previously cataloged earthquakes in Chile, we obtain >80% true positive prediction rates for high-frequency data (>2 Hz) and stations separated in excess of 100 km. As a secondary benefit, the output of the neural network can also be used to infer predicted phase types of arrivals. Here, the method is ideally applied in conjunction with standard travel-time-based association routines and can be adapted for arbitrary network geometries and applications, so long as sufficient training data are available.

58 GEOSCIENCES↗
DOE OSTI2024

Measuring stomatal and guard cell metrics for plant physiology and growth using StoManager1

Abstract Automated guard cell detection and measurement are vital for understanding plant physiological performance and ecological functioning in global water and carbon cycles. Most current methods for measuring guard cells and stomata are laborious, time-consuming, prone to bias, and limited in scale. We developed StoManager1, a high-throughput tool utilizing geometrical, mathematical algorithms, and convolutional neural networks to automatically detect, count, and measure over 30 guard cell and stomatal metrics, including guard cell and stomatal area, length, width, stomatal aperture area/guard cell area, orientation, stomatal evenness, divergence, and aggregation index. Combined with leaf functional traits, some of these StoManager1-measured guard cell and stomatal metrics explained 90% and 82% of tree biomass and intrinsic water use efficiency (iWUE) variances in hardwoods, making them substantial factors in leaf physiology and tree growth. StoManager1 demonstrated exceptional precision and recall (mAP@0.5 over 0.96), effectively capturing diverse stomatal properties across over 100 species. StoManager1 facilitates the automation of measuring leaf stomatal and guard cells, enabling broader exploration of stomatal control in plant growth and adaptation to environmental stress and climate change. This has implications for global gross primary productivity (GPP) modeling and estimation, as integrating stomatal metrics can enhance predictions of plant growth and resource usage worldwide. Easily accessible open-source code and standalone Windows executable applications are available on a GitHub repository (https://github.com/JiaxinWang123/StoManager1) and Zenodo (https://doi.org/10.5281/zenodo.7686022).

Wang, Jiaxin (ORCID:0000000348085085)↗
DOE OSTI2021

Deep Kronecker neural networks: A general framework for neural networks with adaptive activation functions

Here we propose a new type of neural networks, Kronecker neural networks (KNNs), that form a general framework for neural networks with adaptive activation functions. KNNs employ the Kronecker product, which provides an efficient way of constructing a very wide network while keeping the number of parameters low. Our theoretical analysis reveals that under suitable conditions, KNNs induce a faster decay of the loss than that by the feed-forward networks. This is also empirically verified through a set of computational examples. Furthermore, under certain technical assumptions, we establish global convergence of gradient descent for KNNs. As a specific case, we propose the Rowdy activation function that is designed to get rid of any saturation region by injecting sinusoidal fluctuations, which include trainable parameters. The proposed Rowdy activation function can be employed in any neural network architecture like feed-forward neural networks, Recurrent neural networks, Convolutional neural networks etc. The effectiveness of KNNs with Rowdy activation is demonstrated through various computational experiments including function approximation using feed-forward neural networks, solution inference of partial differential equations using the physics-informed neural networks, and standard deep learning benchmark problems using convolutional and fully-connected neural networks.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Non-Blind Deblurring for Fluorescence: A Deformable Latent Space Approach with Kernel Parameterization

We report N\non-blind deblurring (NBD) is a modeling method of the image deblurring problem in computer vision, where the blurring kernel is known or can be externally estimated. In this paper, we attempt to solve a parametric NBD problem, inspired by the simultaneous acquisition of ptychography and fluorescent imaging (FI). Ptychography is an imaging method that favors larger probes, i.e. convolutional kernels, while FI relies on a small probe for high resolution. Also, the kernel can be solved during ptychographic reconstruction. With Ptycho-FI using the same larger kernel, we can perform NBD on the blurred fluorescent images to achieve high-resolution FI, and thus speed up the experiments. To this end, we design a deep latent space deformation network that is directly parameterized by the kernel. The network consists of three components: encoder, deformer, and decoder, where the deformer is specifically meant to rectify the latent space representations of blurred images to a standard latent space, regardless of the kernel. The deformation network is trained with a two-stage training scheme. We conduct extensive experiments to confirm that our parametric model can adapt to drastically different blurring kernels and perform robust deblurring.

77 NANOSCIENCE AND NANOTECHNOLOGY↗
NASA NTRS2018

Automated Fiber Placement Defects: Automated Inspection and Characterization

Automated Fiber Placement (AFP) is an additive composite manufacturing technique, and a pressing challenge facing this technology is defect detection and repair. Manual defect inspection is time consuming, which led to the motivation to develop a rapid automatic method of inspection. This paper suggests a new automated inspection system based on convolutional neural networks and image segmentation tasks. This creates a pixel by pixel classification of the defects of the whole part scan. This process will allow for greater defect information extraction and faster processing times over previous systems, motivating rapid part inspection and analysis. Fine shape, height, and boundary detail can be generated through our system as opposed to a more coarse resolution demonstrated in other techniques. These scans are analyzed for defects, and then each defect is stored for export, or correlated to machine parameters or part design. The network is further improved through novel optimization techniques. New training instances can also be created with every new part scan by including the machine operator as a post inspection check on the accuracy of the system. Having a continuously adapting inspection system will increase accuracy for automated inspections, cutting down on false readings.

Sacco, Christopher↗
DOE OSTI2019

Prediction of atomization energy using graph kernel and active learning

Data-driven prediction of molecular properties presents unique challenges to the design of machine learning methods concerning data structure/dimensionality, symmetry adaption, and confidence management. In this paper, we present a kernel-based pipeline that can learn and predict the atomization energy of molecules with high accuracy. The framework employs Gaussian process regression to perform predictions based on the similarity between molecules, which is computed using the marginalized graph kernel. To apply the marginalized graph kernel, a spatial adjacency rule is first employed to convert molecules into graphs whose vertices and edges are labeled by elements and interatomic distances, respectively. We then derive formulas for the efficient evaluation of the kernel. Specific functional components for the marginalized graph kernel are proposed, while the effects of the associated hyperparameters on accuracy and predictive confidence are examined. We show that the graph kernel is particularly suitable for predicting extensive properties because its convolutional structure coincides with that of the covariance formula between sums of random variables. Using an active learning procedure, we demonstrate that the proposed method can achieve a mean absolute error of 0.62 ± 0.01 kcal/mol using as few as 2000 training samples on the QM7 dataset.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

DeepAdversaries: examining the robustness of deep learning models for galaxy morphology classification

With increased adoption of supervised deep learning methods for work with cosmological survey data, the assessment of data perturbation effects (that can naturally occur in the data processing and analysis pipelines) and the development of methods that increase model robustness are increasingly important. In the context of morphological classification of galaxies, we study the effects of perturbations in imaging data. In particular, we examine the consequences of using neural networks when training on baseline data and testing on perturbed data. We consider perturbations associated with two primary sources: (a) increased observational noise as represented by higher levels of Poisson noise and (b) data processing noise incurred by steps such as image compression or telescope errors as represented by one-pixel adversarial attacks. We also test the efficacy of domain adaptation techniques in mitigating the perturbation-driven errors. We use classification accuracy, latent space visualizations, and latent space distance to assess model robustness in the face of these perturbations. For deep learning models without domain adaptation, we find that processing pixel-level errors easily flip the classification into an incorrect class and that higher observational noise makes the model trained on low-noise data unable to classify galaxy morphologies. On the other hand, we show that training with domain adaptation improves model robustness and mitigates the effects of these perturbations, improving the classification accuracy up to 23% on data with higher observational noise. Domain adaptation also increases up to a factor of ${\approx}2.3$ the latent space distance between the baseline and the incorrectly classified one-pixel perturbed image, making the model more robust to inadvertent perturbations. Successful development and implementation of methods that increase model robustness in astronomical survey pipelines will help pave the way for many more uses of deep learning for astronomy.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2023

Quarknet Project: Simulating Double-Source Plane Lenses and Einstein Rings for Domain Adaptation

Double-Source Plane Lenses (DSPLs) are a form of strong gravitational lensing that can be used to uniquely constrain cosmological parameters like the dark energy equation of state. These objects are exceedingly rare: to date, only four have been discovered in astronomical surveys, like the Sloan Digital Sky Survey and the HyperSuprimeCam Survey. Additionally, the main features of these objects are typically co-located sets of arcs, which look like Einstein Rings of single-source galaxy-galaxy lenses. Larger astronomical surveys typically have low image resolution, which significantly exacerbates issues when trying to find DSPLs. Future surveys are likely to include many more of these objects and have higher image resolution. Traditional methods for finding strong lenses struggle to identify simple galaxy-galaxy lenses. Recently, neural networks have become the standard for lens identification, but these tools have not yet been used to search for DSPLs. In this work, we use the software deeplenstronomy to simulate varied images of DSPLs and Einstein Rings – with and without noise and artifacts of observational surveys, forming a comprehensive benchmark dataset of 40,000 images. We utilise this dataset to perform experiments with convolutional neural networks to discern between the two phenomena. Sofia Grimm and Jerry Zhou are co-first authors on this work.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗