Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Mathematics and Computing, Information sciences”

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 541 records · Page 30
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Using Complex Probability Amplitudes to Simulate Solute Transport in Composite Porous Media

Probability amplitudes are fundamental to quantum mechanics and offer robust descriptions of complicated systems, which have allowed physicists to explain behaviors inaccessible to classical physics. This article ponders how some of the same conceptual underpinnings of the mathematics used for modeling quantum systems might be applied to subsurface water resources problems and speculates how these tools could facilitate applications on quantum computers. A probability amplitude-based model for describing advective-dispersive transport in porous media using linear operators is investigated. The proposed complex valued model decomposes spreading into two “sub-continuum partial dispersion” coefficients, and this recovers classical spreading when the sum of these coefficients is the Fickian dispersion coefficient. However, the probability amplitudes have a manyto-one relationship to a probability distribution, so it embeds a level of heterogeneity into seemingly equivalent functions. Two propagators with different sub-continuum coefficients may have the same macroscopic behavior when either is considered in isolation, but when they act on the other the system’s behavior changes. Additionally, differences in the amplitudes cause a reduction in spreading as velocity correlations are disrupted, despite both propagators having identical dispersion coefficients, and this cannot be achieved using classical methods without changing the dispersion coefficient. The main point is that these amplitude-based models offer a way to embed information about the system into the propagators, instead of just “averaging it out” when making an upscaled model.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2024

High throughput, accurate gene annotation through AI and HPC-enabled structural analysis

With the advances in next generation sequencing technologies, the number of sequenced genomes is growing exponentially, resulting in a technology bottleneck for the translation of sequence information into usable hypotheses about the function of each gene. We have proposed leveraging our leadership high-performance computing (HPC) resources to help break this annotation bottleneck. Here we design an HPC-based framework to infer gene function from gene sequence by incorporating information about protein structure and interactions predicted by deep learning approaches. Accurate functional prediction and gene annotation using computational methods will facilitate breakthroughs in the genomic sciences essential to understanding and harnessing life processes in bacteria, fungi and plants. The development and applications of the state-of-the-art deep neural networks to protein structural modeling, interaction prediction, sequence comparison, and quality assessment of protein structural models will be made possible by leadership computational resources. These HPC-enabled bioinformatics and molecular modeling tools will provide powerful insights into molecular functions of genes.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2024

Bayesian Analysis of TRISO Fuel: Quantifying Model Inadequacy, Incorporating Lower-Length-Scale Effects, and Developing Parallel Active Learning Capabilities

The U.S. Department of Energy (DOE)’s Nuclear Energy Advanced Modeling and Simulation (NEAMS) program aims to develop predictive capabilities by applying computational methods to the analysis and design of advanced reactor and fuel-cycle systems. This program has been providing engineering-scale support for the continued development of BISON, a high-fidelity, high-resolution fuel performance tool. Fuel behavior in nuclear reactors is governed by a complex network of mechanisms that interact with various other physics aspects in the reactor system. Any model developed to represent fuel behavior will likely be idealized, resulting in uncertainties when comparing their predictions against the observed data. In Fiscal Year (FY)-23, we initiated the Uncertainty Quantification (UQ) work by using Bayesian methods to establish a level of model trustworthiness and further improve it, with a particular emphasis on TRI-Structural isOtropic (TRISO) nuclear fuel. This year, we further expanded on that UQ work by investigating an approach to quantifying model inadequacy and accounting for lower-length scale (LLS) effects in TRISO silver (Ag) release modeling. Furthermore, we are implementing parallel active learning capabilities to reduce the computational cost (i.e., required computational resources and elapsed time) of performing UQ. Specifically, we utilized The Kennedy O’Hagan framework for Bayesian uncertainty quantification (KOH) to account for model inadequacy in TRISO Ag release predictions made by BISON. The KOH framework represents an improvement over the standard Bayesian framework used in FY-23. Explicitly accounting for model inadequacy in the Bayesian framework helps establish the level of experimental noise uncertainty in the Advanced Gas Reactor (AGR) data. We compared the inverse UQ results obtained from both the standard Bayesian and KOH frameworks in light of the AGR-2/3/4 data, and also compared the predictive UQ results obtained from these two frameworks in light of the AGR-1 data. Next, we investigated the impact of considering LLS effects in the Ag release simulations. We developed an expanded database of LLS simulated effective diffusivities for Ag, covering a wide range of microstructures and temperatures. Using this database, we developed a framework for incorporating LLS effects into the engineering-scale Ag release UQ. We developed both parametric and non-parametric approaches for bridging the length scales. We then investigated the inverse UQ results in light of the AGR-2/3/4 data and the predictive UQ results in light of the AGR-1 data, and compared the LLS-informed approach and the Arrhenius equation, which does not include microstructure information. Finally, we discussed implementing parallel active learning capabilities in the Multiphysics Object Oriented Simulation Environment (MOOSE)/BISON to reduce the computational cost (i.e., computational resources and elapsed time) of Bayesian UQ. For verification purposes, we first tested these new capabil ities on a species interaction problem. We then demonstrated them on the TRISO Ag release application, showing that parallel active learning capabilities can enhance the accuracy of UQ while also substantially reducing the computational cost in comparison to the reference methods developed in FY-23.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2020

On the Convergence of Physics Informed Neural Networks for Linear Second-Order Elliptic and Parabolic Type PDEs

Physics informed neural networks (PINNs) are deep learning based techniques for solving partial differential equations (PDEs) encountered in computational science and engineering. Guided by data and physical laws, PINNs find a neural network that approximates the solution to a system of PDEs. Such a neural network is obtained by minimizing a loss function in which any prior knowledge of PDEs and data are encoded. Despite its remarkable empirical success in one, two or three dimensional problems, there is little theoretical justification for PINNs. As the number of data grows, PINNs generate a sequence of minimizers which correspond to a sequence of neural networks. We want to answer the question: Does the sequence of minimizers converge to the solution to the PDE? We consider two classes of PDEs: linear second-order elliptic and parabolic. By adapting the Schauder approach and the maximum principle, we show that the sequence of minimizers strongly converges to the PDE solution in C 0 . Furthermore, we show that if each minimizer satisfies the initial/boundary conditions, the convergence mode becomes H 1 . Computational examples are provided to illustrate our theoretical findings. To the best of our knowledge, this is the first theoretical work that shows the consistency of PINNs.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

The confluence of machine learning and multiscale simulations

Multiscale modeling has a long history of use in structural biology, as computational biologists strive to overcome the time- and length-scale limits of atomistic molecular dynamics. Contemporary machine learning techniques, such as deep learning, have promoted advances in virtually every field of science and engineering and are revitalizing the traditional notions of multiscale modeling. Deep learning has found success in various approaches for distilling information from fine-scale models, such as building surrogate models and guiding the development of coarse-grained potentials. However, perhaps its most powerful use in multiscale modeling is in defining latent spaces that enable efficient exploration of conformational space. In conclusion, this confluence of machine learning and multiscale simulation with modern high-performance computing promises a new era of discovery and innovation in structural biology.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2020

Redox transistors based on TiO 2 for analogue neuromorphic computing

The ability to train deep neural networks on large data sets have made significant impacts onto artificial intelligence, but consume significant amounts of energy due to the need to move information from memory to logic units. In-memory "neuromorphic" computing presents an alternative framework that processes information directly on memory elements. In-memory computing has been limited by the poor performance of the analogue information storage element, often phase-change memory or memristors. To solve this problem, we developed two types of "redox transistors" using TiO 2 (anatase) which stores analogue information states through the electrochemical concentration of dopants in the crystal. The first type of redox transistor uses lithium as the electrochemical dopant ion, and its key advantage is low operating voltage. The second uses oxygen vacancies as the dopant, which is CMOS compatible and can retain state even when scaled to nanosized dimensions. Both devices offer significant advantages in terms of predictable analogue switching over conventional filamentary-based devices, and provide a significant advance in developing materials and devices for neuromorphic computing.

36 MATERIALS SCIENCE↗
DOE OSTI2023

University Coalition for Fossil Fuel Energy Research

Following a nationwide open competition, the University Coalition for Fossil Energy Research (UCFER) was established in October 2015 through a cooperative agreement between Penn State and the Department of Energy (DOE) National Energy Technology Laboratory (NETL). Penn State lead UCFER with the objective of advancing basic and applied research for clean and low-carbon energy based on fossil fuels in support of the DOE’s mission. UCFER focused on research that improves the efficiency of production and use of fossil energy resources, while minimizing the environmental impacts and reducing greenhouse gas emissions. Penn State lead a team of nine universities (Massachusetts Institute of Technology, The Pennsylvania State University, Princeton University, Texas A&M University, University of Kentucky, University of Southern California, The University of Tulsa, University of Wyoming, and Virginia Polytechnic and State University) during the competition stage, adding seven more universities in 2017 (Carnegie Mellon University, Louisiana State University, The Ohio State University, University of North Dakota, University of Pittsburgh, University of Utah, and West Virginia University). This Coalition exhibited a wide geographical distribution across the U.S. bringing a wide variety of fossil energy expertise. This national university alliance was a major collaborative effort with NETL to address specific topics of R&D in NETL’s mission area, which involved one or more of NETL’s five core competencies (Geologic and Environmental Systems, Materials Engineering and Manufacturing, Energy Conversion Engineering, Systems Engineering and Analysis, and Computational Science and Engineering). The first five to six months of the project was the definitization stage. During this period, Penn State worked closely with NETL to finalize the Coalition organizational structure and By- Laws, prepare a statement of substantial involvement and a statement of project objectives, and develop operations and membership plans. A major component of this stage included preparing an execution plan to solicit research, evaluate proposals, recommend selected projects to NETL, and award projects. In addition, a plan was prepared to monitor projects, review projects, disseminate knowledge from research projects and develop an online system for Coalition research portfolio management. This included developing a website and several databases. The first of six rounds of solicitations started in mid-2016. Projects from the sixth solicitation started February 1, 2021, and ended January 31, 2023. Projects that were selected represented twelve technology lines. Approximately $16.6 million in funding was available for the six solicitations. Most of the funding was provided by DOE, Office of Fossil Energy (DOEFE) with the DOE Fuel Cells Technologies Office (DOE-FCTO) providing funding for a few projects. Coalition universities submitted 259 proposals in response to the solicitations, requesting approximately $67.0 million in funding, and forty-three projects were selected. However, one project withdrew after the principal investigator left the university. The management of the Coalition projects was a major activity by Penn State. Managing the Coalition projects consisted of monitoring the projects, reviewing the projects through annual technical review meetings, disseminating knowledge from the research projects, and developing an online system for Coalition research portfolio management. Penn State’s OMT monitored projects to ensure that all milestones (technical, schedule, budget) were met, expenditures were allowable, cost share (when applicable) were reported, and all technical reports were submitted. The OMT also posted the technical reports electronically on a secure members-only website for access and review by the Coalition members. Disseminating knowledge from the research projects was done through a website, newsletters, various meetings, conferences, journal articles, publicity/press releases, and project summaries that were prepared after each project was completed. Penn State kept NETL apprised of UCFER progress through quarterly reports (thirtyone were submitted by Penn State), verbal and written communications, yearly updates at the annual technical review meetings, and cost accrual reports. The UCFER project had a significant impacty. The UCFER program established the first national university alliance in fossil energy research with a major collaboration effort with DOE NETL that addressed specific topics in NETL’s research and development mission areas. It generated inter-university collaborations, which was another program interest. Twenty-two out of 259 proposals contained collaborations (≈8.5%) and three of forty-two funded projects involved inter-university collaborations (≈7.0%). The forty-two funded projects provided support at fourteen universities involving 269 personnel. Research was conducted by 106 faculty, 115 graduate and undergraduate students, forty-six research staff and post-doctoral scholars, and two visiting scholars. Students and post-doctoral scholars were also on-site at NETL through CRADAs. In addition, non-Coalition participants included six universities and sixteen companies and national laboratories. The non-Coalition participants were involved as subcontractors, providers of cost share, performed unpaid consultation and sample analysis, served as advisory board members, or were providers of samples and materials for testing. UCFER also produced visibility in that fifty-six refereed journal articles were published, fifty-seven conference papers and twenty-three posters were prepared, 190 presentations were given, eight patent applications were filed, two books/book chapters were written, and ten software codes were developed. Collaboration between NETL and the individual projects was a major requirement for all funded projects. This included NETL staff time to support collaboration, consultation, technical guidance, sample preparation and analysis, internships at NETL, on-site testing and equipment usage by Coalition participants at NETL, co-mentoring students, and coauthoring journal articles and conference papers. Collaboration was impacted by COVID-19 in that not all on-site activities could be performed. NETL personnel were coauthors on eight of the conference papers (fourteen percent of the conference papers that were prepared) and seventeen of the journal articles (thirty percent of the journal articles that were prepared). A website was developed for an online proposal solicitation and review process and to provide exposure to UCFER. A website analysis highlighted the large amount of member and general public interest in UCFER by interpreting access statistics from March 2016 through June 2023. Visitors to the site originated from many different organizations, businesses, and countries. The website provided a means to disseminate information to both the general public and the UCFER members and was successfully used for outreach activities. In addition, NETL required that RFP release, proposal submission, and proposal reviews all be performed online. Penn State successfully developed these capabilities in a secure section of the website, which were used throughout the UCFER project. It is recognized that each project had its technical successes. In addition, highlighted successes were compiled and summarized from the research projects. Information was requested from the PIs of completed projects. In addition, Penn State’s Operations Management Team reviewed subcontract reports to identify project successes. Examples of information requested from PIs included (not all-inclusive): new projects that have been funded as a result of UCFER funding; new commercial products; establishment of a new center; new patent; new software; best paper awards; highly-cited work; and graduate student successes. A total of forty-eight highlighted successes were reported.

01 COAL, LIGNITE, AND PEAT↗
DOE OSTI2024

Non-Intrusive Parallel-in-Time Solvers for Partial Differential Equations (Final Report)

Many time-dependent problems and simulations are often modeled using Partial Differential Equations. Traditional modeling approaches that use sequential time-stepping are reaching a bottleneck in optimizing efficiency. The Center of Applied Science and Computing at Lawrence Livermore National Laboratory extensively works on parallelizing these algorithms to leverage the increasing computational power from the growing number of processors in computer hardware. In particular, they aim to design non-intrusive algorithms that can generalize to a variety of problems and sizes without requiring additional information from or modifications on the original problems. Multigrid Reduction in Time (MGRIT) is a parallel-in-time algorithm that is designed to be non-intrusive. This project focuses on increasing the efficiency of MGRIT by approximating the coarse-grid operator using machine learning approaches as a means to find the most non-intrusive, or general, solution.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Partial Support of the Fast-Track Consensus Study on Foundational Research Gaps and Future Directions for Digital Twins (Final Report)

This study from the National Academies of Sciences, Engineering, and Medicine was launched to explore the foundational research gaps and opportunities for digital twins. As part of the information gathering process, the committee organized three targeted workshops—in engineering, climate sciences, and biomedical sciences—to better understand domain-specific nuances and barriers to developing digital twins. These sessions enabled cross-sector experts to surface field-specific needs, challenges, and open questions related to digital twins. Thousands of participants across multiple domains engaged in the discussions, which workshops were summarized in three separate Proceedings-in-Brief.

42 ENGINEERING↗
DOE OSTI2025

QUCODE: End-to-End Qubit Co-Design

The design of a quantum computer can be broken down into different steps, e.g., the material science aspect of designing qubits and devices, considerations of controlling the state of the qubits and their environment, the computer science aspects of mapping algorithms to the available primitives of the quantum computer, and the programming of an application in terms of the available algorithms. Research in these areas is currently fairly isolated, and there is framework for an end-to-end design approach where a desired application informs the choice of materials for the qubits and their environment, and vice versa.We identify knowledge gaps and opportunities for research that builds on existing PNNL capabilities.

36 MATERIALS SCIENCE↗
DOE OSTI2024

WETO Software Stack Best Practices

Wind energy researchers typically share one key characteristic: a passion for increasing wind energy in the global energy mix. The U.S. Department of Energy (DOE) supports this mission in a number of ways including allocating funding directly to various aspects of wind energy research through the Office of Energy Efficiency and Renewable Energy (EERE) via the Wind Energy Technologies Office (WETO). While the traditional output of research is academic publication, software development efforts are increasingly a major focus. Software tools in the research environment allow researchers to describe an idea and quickly increase the scope and scale as they study it further. As a product of research, these tools represent a direct pipeline from researcher to industry practitioners since they are the implementation of ideas described in academic publications. Given this vital role in wind energy research and commercial development, the broad research software portfolio supported by WETO must maintain a minimum level of quality to support the wind energy field in the growing transition to renewable energy. This report outlines a series o f best practices to be adopted by all WETO-supported software projects, as well as expectations that the communities interacting with these projects should have of the developers and tools themselves. Wind energy research software has a unique standing in the field of scientific software. The stakeholders are varied with a subset being: (1) DOE EERE leadership, (2) DOE WETO leadership and program managers, (3) National lab leadership, (4) Associated project principle investigators, (5) Research software engineers, (6) Wind energy researchers in academia (including graduate students, post docs, and national lab staff), (7) Industry researchers and practitioners, (8) Commercial software developers, and (9) The general public interested in wind energy. These software are typically the end-user of other generic software libraries, so the funding cycles are often tied to applied research rather than the development of the software itself. Since the developers are also wind energy researchers, these tools are typically designed in a way that closely resembles the application in which they're used. Additionally, the expertise and incentives for the developers have a high variability, and often neither are aligned with software engineering or computer science. Given the unique environment in which wind energy research software is produced and consumed, it is critical for model owners to understand the context of their software. A framework for developing this understanding is to answer the following questions of a given software project: What is it's purpose? What is its role in the field of wind energy? What is the profile of the expected users? For how long will it be relevant? What is the expected impact? These questions allow model owners to identify the appropriate methods for the design, development, and long term maintenance of their software. Additionally, the answer provide context for future planners to understand why particular decisions were made and discern the consequences of changing course. The information is aggregated from experience within WETO-supported software development groups as well as external organizations and efforts to define the craft of research software engineering. These best practices aim to make the collaborative development process efficient and effective while improving the model understanding across stakeholders. Additionally, the general adoption of a common framework for software quality ensures that the end users of WETO software can trust these tools and accurately understand the risks to workflow integration.

17 WIND ENERGY↗
DOE OSTI2026

CONUS-wide Projected Flood Frequency and Uncertainty Estimates, Version 1.0

This dataset presents a large-ensemble of CONUS-wide projected flood frequency and uncertainty estimates across ~2.7 million NHDPlusV2 river reaches over the CONUS. The framework producing this dataset leverages a multi-model, uncertainty-aware modeling framework that allows evaluating shifts in flood frequences at the stream reach level across the CONUS. CONUS-wide ensemble streamflow projections generated from hydrologic simulations driven by downscaled and bias-corrected Coupled Model Intercomparison Project Phase 6 (CMIP6) outputs are used to derive these flood frequency and uncertainty estimates over the period 1980 - 2099. A spatially consistent regional L-moment algorithm is applied across clusters defined by the US Hydrologic Unit Code Subregions (HUC4s and HUC8s) and NHDPlusV2 stream orders to estimate flood frequencies. The dataset also includes at-site based flood estimates that allow for the comparison between local and regional approach-based estimates, assess projected changes, and characterize their uncertainties. For more reliable estimation of rare flood frequencies such as 500 and 1000-year return periods, super-ensemble based estimates are also included in the dataset. This dataset is derived to support the "Impact-Informed Dam Safety Risk Assessment for Securing Hydropower Assests" project for the US Department of Energy (DOE) Hydropower and Hydrokinetic Office (H2O). For further details, refer to Kao et al. (2022), Ghimire et al. (2023), Ghimire et al. (2025), and Hosking and Wallis (1997).

Ghimire, Ganesh [ORNL] (ORCID:0000000242843941)↗
DOE OSTI2023

Specifications and a Prototype Software to Demonstrate a Data Catalog for Hanford Datasets

Environmental management activities at the Hanford Site produce extensive data about site conditions, contaminants, and cleanup activities. Managing, archiving, and accessing that data requires a high degree of collaboration among site contractors and a high level of awareness by project managers and staff. The Hanford Site has a range of databases (e.g., Hanford Environmental Information System [HEIS]) and their associated user interfaces (e.g., Environmental Dashboard Application [EDA], Virtual Library [VL]), as well as other document management systems (e.g., Integrated Document Management System [IDMS]). However, Hanford lacks a single unified resource to find data (which itself comes in multiple formats) amongst the multiple disparate systems, not to mention ad hoc data not contained in an official repository/database.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

User’s Manual for RESRAD-RDD&IND Code Version 2: Vol. 1—Methodology and Models Used in RESRAD-RDD&IND Code

RESRAD-RDD&IND is part of the RESRAD family of codes that Argonne National Laboratory developed for the U.S. Department of Energy (DOE). An earlier version, RESRAD-RDD published in 2009, dealt only with radiological dispersal device (RDD) incidents (DOE 2009). This new version, RESRAD-RDD&IND, is designed to evaluate both RDD incidents and improvised nuclear device (IND) incidents. This report is Volume 1 of the RESRAD-RDD&IND User’s Manual that documents the methodology, models, and radionuclide-specific data used in RESRAD-RDD&IND code Version 2.0. Volume 2 of the RESRAD-RDD&IND User’s Manual is called the User’s Guide for RESRAD-RDD&IND Code . It describes how to use RESRAD-RDD&IND code Version 2.0 and includes screen shots and parameter information.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2022

A stochastic covariance shrinkage approach to particle rejuvenation in the ensemble transform particle filter

Rejuvenation in particle filters is necessary to prevent the collapse of the weights when the number of particles is insufficient to properly sample the high-probability regions of the state space. Rejuvenation is often implemented in a heuristic manner by the addition of random noise that widens the support of the ensemble. This work aims at improving canonical rejuvenation methodology by the introduction of additional prior information obtained from climatological samples; the dynamical particles used for importance sampling are augmented with samples obtained from stochastic covariance shrinkage. A localized variant of the proposed method is developed. Numerical experiments with the Lorenz '63 model show that modified filters significantly improve the analyses for low dynamical ensemble sizes. Furthermore, localization experiments with the Lorenz '96 model show that the proposed methodology is extendable to larger systems.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Colloidal State Machines as Smart Tracers for Chemical Reactor Analysis

A widely utilized tool in reactor analysis is passive tracers that report the residence time distribution, allowing estimation of the conversion and other properties of the system. Recently, advances in microrobotics have introduced powered and functional entities with sizes comparable to some traditional tracers. This has motivated the concept of Smart Tracers that could record the local chemical concentrations, temperature, or other conditions as they progress through reactors. Herein, the design constraints and advantages of Smart Tracers by simulating their operation in a laminar flow reactor model conducting chemical reactions of various orders are analyzed. It is noted that far fewer particles are necessary to completely map even the most complex concentration gradients compared with their conventional counterparts. Design criteria explored herein include sampling frequency, memory storage capacity, and ensemble number necessary to achieve the required accuracy to inform a reactor model. Cases of severe particle diffusion and sensor noise appear to bind the functional upper limit of such probes and require consideration for future design. The results of the study provide a starting framework for applying the new technology of microrobotics to the broad and impactful set of problems classified as chemical reactor analysis.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

In situ compression artifact removal in scientific data using deep transfer learning and experience replay

The massive amount of data produced during simulation on high-performance computers has grown exponentially over the past decade, exacerbating the need for streaming compression and decompression methods for efficient storage and transfer of this data---key to realizing the full potential of large-scale computational science. Lossy compression approaches such as JPEG when applied to scientific simulation data realized as a stream of images can achieve good compression rates but at the cost of introducing compression artifacts and loss of information. This paper develops a unified framework for in situ compression artifact removal in which the fully convolutional neural network architectures are combined with scalable training, transfer learning, and experience replay to achieve superior accuracy and efficiency while significantly decreasing the storage footprint as compared with the traditional optimization-based approaches. We demonstrate the proposed approach and compare it with compressed sensing postprocessing and other baseline deep learning models using climate simulations and nuclear reactor simulations, both of which are driven by hyperbolic partial differential equations. Our approach when applied to remove the compression artifacts on the JPEG-compressed nuclear reactor simulation data (using a transfer-trained model that was pretrained on the climate simulation data and updated incrementally as the nuclear reactor simulation progressed), achieved a significant improvement---mean peak signal-to-noise ratio of 42.438 as compared with 27.725 obtained with the compressed sensing approach.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Experimental Observations of the Topology of Convolutional Neural Network Activations

Topological data analysis (TDA) is a branch of computational mathematics, bridging algebraic topology and data science, that provides compact, noise-robust representations of complex structures. Deep neural networks (DNNs) learn millions of parameters associated with a series of transformations defined by the model architecture resulting in high-dimensional, difficult to interpret internal representations of input data. As DNNs become more ubiquitous across multiple sectors of our society, there is increasing recognition that mathematical methods are needed to aid analysts, researchers, and practitioners in understanding and interpreting how these models' internal representations relate to the final classification. In this paper we apply cutting edge techniques from TDA with the goal of gaining insight towards interpretability of convolutional neural networks used for image classification. We use two common TDA approaches to explore several methods for modeling hidden layer activations as high-dimensional point clouds, and provide experimental evidence that these point clouds capture valuable structural information about the model's process. First, we demonstrate that a distance metric based on persistent homology can be used to quantify meaningful differences between layers and discuss these distances in the broader context of existing representational similarity metrics for neural network interpretability. Second, we show that a mapper graph can provide semantic insight as to how these models organize hierarchical class knowledge at each layer. These observations demonstrate that TDA is a useful tool to help deep learning practitioners unlock the hidden structures of their models.

topological data analysis, deep learning↗