Search NASA⌕ Search

SEARCH · Search NASA

Results for “Analyst”

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 685 records · Page 38

The VAST Challenge: History, Scope, and Outcomes: An introduction to the Special Issue

Visual analytics aims to facilitate human insight from complex data via a combination of visual representations, interaction techniques, and supporting algorithms. To create new tools and techniques that achieve this goal requires that researchers have an understanding of analytical questions to be addressed, data that illustrates the complexities and ambiguities found in realistic analytic settings, and methods for evaluating whether the plausible insights are gained through use of the new methods. However, researchers do not, generally speaking, have access to analysts who can articulate their problems or operational data that is used for analysis. To fill this gap, the Visual Analytics Science and Technology (VAST) Challenge has been held annually since 2006. The VAST Challenge provides an opportunity for researchers to experiment with realistic but not real problems, using realistic synthetic data with known events embedded. Since its inception, the VAST Challenge has evolved along with the visual analytics research community to pose more complex challenges, ranging from text analysis to video analysis to large scale network log analysis. The seven years of the VAST Challenge have seen advancements in research and development, education, evaluation, and in the challenge process itself. This special issue of Information Visualization highlights some of the noteworthy advancements in each of these areas. Some of these papers focus on important research questions related to the challenge itself, and other papers focus on innovative research that has been shaped by participation in the challenge. This paper describes the VAST Challenge process and benefits in detail. It also provides an introduction to and context for the remaining papers in the issue.

visual analytics, sensemaking, evaluation, crowdso↗

TexTonic: Interactive Visualization for Exploration and discovery of Very Large Text Collections

TexTonic is a visual analytic system for interactive exploration of very large unstructured text collections. TexTonic visualizes hierarchical clusters of representative terms, snippets, and documents in a single, multi-scale spatial layout. Exploration is supported by interacting with the visualization and directly manipulating the terms in the visualization using semantic interactions. These semantic interactions steer the underlying analytic model by translating user interactions within the visualization to contextual updates to the supporting data model. The combination of semantic interactions and information visualization at multiple levels of the data hierarchy helps users manage information overload so that they can more effectively explore very large text collections. In this paper we describe TexTonic’s data processing and analytic pipeline, user interface and interaction design principles, and the results of a user study conducted mid-development with experienced data analysts. We also discuss the implications TexTonic could have on visual exploration and discovery tasks

human centered computing, Visual analytics, Inform↗

A systematic decision-making methodology to formalize the selection of degree of realism in screening analysis of probabilistic risk assessment

In the nuclear power domain, Probabilistic Risk Assessment (PRA) is used to inform decision-making for Nuclear Power Plants (NPPs). Recently, there has been an increase in the utilization of modeling and simulation (M&S) to support the estimation of PRA inputs. Risk analysts should carefully select the PRA items that require M&S and their degree of realism (DoR) with consideration of the required resources. To support this selection, this article formulates a systematic decision-making approach for the DoR selection. The DoR selection is made based on two predictive decision-making attributes: the predicted differences in safety risk estimate (ΔSaRi) and the cost of analysis (ΔCAN). This research also develops and quantifies causal models to estimate ΔSaRi and ΔCAN. The causal model-based prediction of ΔSaRi and ΔCAN helps reduce the trial-and-error nature of the DoR selection in the PRA screening analysis and provides insights for DoR selection and the gradual refinements of PRA realism. This approach is demonstrated for a case study on fire PRA of NPPs, where an adequate DoR is selected from two fire models: an engineering correlation and a zone model.

Alkhatib, Sari [Department of Nuclear, Plasma, and↗

Cilantro microbiome before and after nonselective pre-enrichment for Salmonella using 16S rRNA and metagenomic sequencing

Background: Salmonella enterica is a common cause of foodborne gastroenteritis in the United States and is associated with outbreaks in fresh produce such as cilantro. Salmonella culture-based detection methods are complex and time consuming, and improvements to increase detection sensitivity will benefit consumers. In this study, we used 16S rRNA sequencing to determine the microbiome of cilantro. We also investigated changes to the microbial community prior to and after a 24-hour nonselective pre-enrichment culture step commonly used by laboratory analysts to resuscitate microorganisms in foods suspected of contamination with pathogens. Cilantro samples were processed for Salmonella detection according to the method in the United States Food and Drug Administration Bacteriological Analytical Manual. Genomic DNA was extracted from culture supernatants prior to and after a 24-hour nonselective pre-enrichment step and 454 pyrosequencing was performed on 16S rRNA amplicon libraries. A database of Enterobacteriaceae 16S rRNA sequences was created, and used to screen the libraries for Salmonella, as some samples were known to be culture positive. Additionally, culture positive cilantro samples were examined for the presence of Salmonella using shotgun metagenomics on the Illumina MiSeq. Results: Time zero uncultured samples had an abundance of Proteobacteria while the 24-hour enriched samples were composed mostly of Gram-positive Firmicutes. Shotgun metagenomic sequencing of Salmonella culture positive cilantro samples revealed variable degrees of Salmonella contamination among the sequenced samples. Conclusions: Our cilantro study demonstrates the use of high-throughput sequencing to reveal the microbiome of cilantro, and how the microbiome changes during the culture-based protocols employed by food safety laboratories to detect foodborne pathogens. Finding that culturing the cilantro shifts the microbiome to a predominance of Firmicutes suggests that changing our culture-based methods will improve detection sensitivity for foodborne enteric pathogens

59 BASIC BIOLOGICAL SCIENCES↗

A parametric analysis of oilfield design factors affecting the detectability and characterization of electrically conductive hydrofracks

Electrical responses in the vicinity of energized steel-cased well sources offer significant potential for monitoring induced fractures. However, the high complexity of well-fracture-host models spanning multiple length scales compels analysts to simplify their numerical models due to enormous computational costs. This consequently limits our understanding regarding monitoring capabilities and the limitations of electrical measurements on realistic hydraulically fracturing systems. In this paper, we use the hierarchical finite element approach to construct geoelectric models in which geometrically complex fractures and steel-cased wells are discretely represented in 3D conducting media without sacrificing the model realism and computation efficiency. We have discovered systematic numerical analyses of the electrical responses to evaluate the influences of borehole material conductivity and the source type as well as the effects of well geometry, conductivity contrast, source location, fracture growth, and fracture propagation. Furthermore, the numerical results indicate that the borehole material property has a strong control on the electrical potentials along the production and monitoring wells. The monopole source located at a steel-cased well results in a current density distribution that decays away from the source location throughout the well length, whereas the dipole source produces a current density that dominates mainly along the dipole length. Moreover, the conductivity contrast between the fractures and host does not change the overall pattern of the electrical potentials but varies its amplitude. The fracture models near different well systems indicate that the well geometry controls the entire distribution of potentials, while the characteristics of the voltage difference profiles along the wells before and after fracturing are insensitive to the well geometry and the well in which the source is located. Further, the hydraulic-fracturing models indicate that the voltage differences along the production well before and after fracturing have strong sensitivity to fracture growth and fracture set propagation.

58 GEOSCIENCES↗

Chapter 13: All-Electric Vertical Take-Off and Landing Aircraft (eVTOL) for Sustainable Urban Travel

Increasing urbanization, ground congestion, and greenhouse gas emissions have spurred aircraft pioneers to capitalize on advancements in battery technology, electric motors, software, and other advances to develop all-electric vertical take-off and landing (eVTOL) aircraft intended to save travelers time within densely populated urban regions. This study served as an example for conducting a life-cycle analysis (LCA) of an eVTOL aircraft, analyzing its manufacturing process, its in-life operation as a passenger transportation service between urban locations, and (in part) end-of-life treatment. This case study aims to provide guidance on how to accurately conduct eVTOL LCA, particularly on how to define the problem, how to obtain the right information from engineers inside and outside their own organizations, and what areas to focus on. As LCA results are only as accurate as the assumptions that were made, a more rigorous LCA that includes sensitivity analysis and uncertainty characterization can help further understand the limitations of those assumptions. New technologies such as blockchain may also play an important role in LCA, including the life-cycle inventory and improving confidence intervals. Completing a credible LCA is central to supporting the argument in favor of introducing the new technology. Completing such a study will allow companies to demonstrate to relevant stakeholders the likely impact of their aircraft on the environment by quantifying the operational climate footprint of the aircraft and the manufacturing process used to build it. To this end, this case study also provides examples of how to report and present LCA results, optimizing for communications to non-LCA experts. The LCA can also assist in identifying opportunities to improve the environmental performance at various life-cycle stages, informing decision makers as products and manufacturing processes evolve or are redesigned. Given the nascency of both eVTOL aircraft manufacturing and operations, it is strongly recommended that LCA analysts treat reports on eVTOL aircraft as works-in-progress. As new information is found, design changes are made, and the aircraft or operation gains maturity, it is inevitable that some amendments will need to be made to the LCA. It is suggested that a complete overhaul of the LCA be made every 18-24 months throughout the R&D, design, and operational scale-up phases of the project.

ADVANCED PROPULSION SYSTEMS↗

Tuning an Electronic Structure Code

As discussed in previous chapters, performance analysts have many tools at their disposal. The performance issues presented by any particular application determine the specific utility of each of these tools, including which we should even use. However, experience frequently shows that issues arise that require features not available in any existing tool.

97 MATHEMATICS AND COMPUTING↗

Development of a Scalable Risk-informed Predictive Maintenance Cloud-based Strategy at Nuclear Power Plants

The fact that light-water reactor operation and maintenance costs are prohibitively expensive and contribute to the premature decommissioning of nuclear power plants is partly due to how the equipment is monitored. In recent years, cloud computing has emerged as a dominant technology, as its low cost, computing and storage adaptability, and ability to host applications across numerous virtual infrastructures potentially make it a cost-effective alternative to onsite storage and diagnostics. In this paper, a technological assessment is carried out on a provisional cloud deployment architecture for a nuclear power plant predictive monitoring system. This cloud-based monitoring system would enable maintenance and diagnostic analysts and other authorized plant users to remotely monitor equipment functionality, thus enabling early fault detection and effective predictive maintenance practices. To provide data processing and storage, sensor device networking, and database management, the Microsoft Azure cloud platform is utilized as part of the proposed cloud architecture; however, this analysis could be extended to other cloud computing service providers as well. The focus of this paper is on application of cloud resources for enabling predictive maintenance, identification of technological hurdles associated with moving to a cloud-computing-based architecture, and potential benefits from moving to a centralized cloud system.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Estimating the Contributions to Human Error Probability from the Convolution of the Distribution of Time Available and Time Required

As part of their duties, Human Reliability Analysis must often evaluate if crews in nuclear power plants (NPPs) can complete tasks associated with a human-failure event within time limits. For example, the time required in NPP scenarios is determined by systematic and structured walkthroughs, feasibility studies, recorded times from training exercises, and interviews with experienced operators and experts. Typically, a point estimate is derived for the estimate (mean, maximum, or 95th percentile of time required). Using point-estimate values can mask the risk associated with variability among crews, plant conditions and set-up, environmental conditions, and other impact factors under which these actions are executed. While point estimates for time required and time available have served the industry well, without considering the uncertainty they could lead to biased understanding about the risk. The Integrated Human Event Analysis System - General Methodology (IDHEAS-G) model (developed by the US Nuclear Regulatory Commission, NRC) for human error probability calculates human error probability by summing two probabilities: insufficient time and cognitive error. As such, the model takes a more holistic approach by considering the full distributions for time required and time available to calculate the human error probability because the time available to complete the task is insufficient. In this study, we expand on the work of the NRC and discuss methods for estimating these time considerations. For example, for the time required, the impact of Performance Influencing Factors (PIFs) on the distribution was divided into impacts that are aleatory in nature, such as crew-to-crew variability, and those that are epistemic (i.e., the PIFs). Starting with the factors that introduce aleatory uncertainty, a first-order distribution was developed from a large set of time required (i.e., NPP task completion times) data for the range of operator actions that occur in the NPP control room under simulated accident conditions. The first-order distribution can then be adjusted to account for epistemic uncertainty using research associated with the impact of applicable PIFs on the time required. We also develop guidance for analysts to address the probability distributions for the time available. The guidance we developed on how to estimate time required and time available distributions is based on the identification of pertinent research and data, data analyses, and expert knowledge elicitation.

human error probability, human performance, time e↗

Natural Language Processing-Enhanced Nuclear Industry Operating Experience Data Analysis: Aggregation and Interpretation of Multi-Report Analysis Results

Industry-wide operating experience is a critical source of raw data for reliability and risk model parameter estimations for nuclear power plants. A large portion of operating experience data are failure events stored as reports that contain unstructured data, such as narratives. In current practice, a failure report is usually reviewed and manually coded by analysts. The coding is based on extracting several event characteristics such as system name, component type, sub-part type, failure mode, and failure cause. Event narratives are mostly used to help understand events and extract their characteristics. In this line of research, we aim to maximize the usage of event narratives by leveraging natural language processing (NLP) methods to automatically convert an event narrative to a causal graph. This research has promise to improve physical understanding of failure initiation and propagation and to facilitate use of non-failure data (e.g., near-misses and degradations) to complement the limited data pool of failures. In our previous work, we developed an NLP tool and applied it to analyze a number of licensee event reports submitted by U.S. nuclear power plants to the Nuclear Regulatory Commission. In this paper, we will report our recent research progress in aggregating the results of multiple reports, developing network model(s), and drawing statistical insights.

99 GENERAL AND MISCELLANEOUS↗

Sensitivity Calculations for Systems with Polyethylene Reflector Materials Using CLUTCH

The SCALE 6.2.4 code package contains four sequences for calculating $k_{eff}$ sensitivity coefficients. Two of these sequences use deterministic transport solvers: a one-dimensional (1D) capability based on XSDRN, and a two-dimensional (2D) capability based on NEWT. These sequences are restricted to the multigroup (MG) treatment of neutron energy. The three-dimensional (3D) sequences use the KENO V.a or KENO-VI Monte Carlo transport codes and can be used to calculate sensitivity coefficients with either MG or continuous-energy (CE) transport. The 3D sensitivities are ultimately reported in an MG structure, regardless of the method used in the transport calculations. If desired, the sensitivity coefficients can be reported with very fine energy resolution from a CE calculation, but they are calculated only in the MG library structure in the MG mode. CE TSUNAMI methods are available in SCALE starting in SCALE version 6.2. Sensitivity coefficients were generated using the 3D sequences as part of the generation of the SCALE 6.2.2 Validation Report; difficulties encountered when using the CLUTCH method for thick, fissionable-material reflectors were discussed and investigated as documented in a previous paper. This paper discusses the difficulties encountered in generating accurate sensitivity coefficients using the CLUTCH technique for polyethylene reflectors for two fast spectrum benchmarks. Direct perturbation (DP) calculations were performed to confirm the accuracy of the total sensitivity coefficient for important isotopes with large sensitivities in the system. Discrepancies were detected for CLUTCH-calculated sensitivity coefficients in the reflector of a critical experiment with a radial polyethylene reflector. A simple polyethylene-reflected plutonium sphere was then used to further investigate the discrepancy. Calculations performed using the iterated fission probability (IFP) method generated accurate sensitivity coefficients in both cases. The results of this study emphasize the need to confirm CLUTCH sensitivity results with DP calculations. IFP calculations are generally less efficient but more reliable than CLUTCH calculations. Improvements to the CLUTCH methodology that retain the greater efficiency but address identified difficulties are therefore potentially useful to analysts.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Preliminary Design of Critical Experiments Involving Commercially Available B 4 C Neutron Absorber Plates with Low-Enriched UO 2 Fuel

The International Criticality Safety Benchmark Evaluation Project (ICSBEP) is an initiative to provide high-quality benchmark data in a standardized format for criticality safety analysts to use to validate calculational tools and nuclear data. In the last released 2022 version, the ICSBEP handbook contains over 5,000 critical experiment descriptions, results, and associated models. This paper describes the critical configurations proposed and calculations performed as part of the conceptual design phase of Integral Experiment Request (IER) 554 (IER-554). The goal of IER-554 is to add to the benchmark experiments available to the criticality safety community by designing critical experiments that can analyze how adding commercially available Boralcan plates to an assembly of low-enriched UO2 fuel rods affects the effective neutron multiplication factor (k eff ). Boralcan, a neutron absorber product developed by Rio Tinto, is made of B4C in an Al 1100 matrix and is commonly used for criticality suppression in fuel storage pools. In the current version of the ICSBEP handbook, only a few dated critical experiments involve B4C materials, and none involve Boralcan in a thin plate shape. The experiments designed as part of IER-554 are intended to be performed at the Sandia Pulsed Reactor Facility/Critical Experiments (SPRF/CX) apparatus at Sandia National Laboratories. SPRF/CX is a well-characterized assembly considered trustworthy by the benchmarking community because of the numerous high-quality evaluations with very low experimental uncertainties included in the ICSBEP handbook. Figure 1 illustrates the assembly for a particular configuration from LCT-078, one of the published benchmarks in the ICSBEP handbook. The experiments will be moderated and reflected by light water at ambient atmospheric pressure. Before the plates are inserted in the critical assembly, they will be characterized with x-ray computed tomography (XCT) to identify the sizes and distribution of the B 4 C powder particles inside the plates. The preliminary design calculations were performed with SCALE 6.3.0 using the KENO V.a sequence for the criticality calculations, the CE-TSUNAMI-3D sequence for the sensitivity and uncertainty studies, and the ENDF/B-VII.1 Continuous Energy cross section library. All the calculation results presented have 10 pcm statistical uncertainties. A set of critical experiments with Boralcan plates could increase the confidence of the criticality safety community in its modeling methods when using this type of neutron absorber material. Experiments would also help validate k eff calculations, and the industry could use these validations to change the B loading credit limits from the US Nuclear Regulatory Commission standard review plan for dry cask storage of spent nuclear fuel. This paper summarizes only part of the analysis documented in the preliminary design report.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

l3d2vtk: An MCNPTools Utility to Enable LNK3DNT File Visualization & Post-Processing

This paper describes recent work to develop and incorporate a utility program into the MCNPTools package that enables LNK3DNT file visualization & post-processing. A LNK3DNT file is a binary Cartesian, cylindrical, or spherical structured mesh file that describes material and material relative density distributions by mesh voxel. This mesh file format is principally used by the PARTISN discrete-ordinates radiation-transport code. However, the MCNP ® Monte Carlo radiation-transport code can use LNK3DNT files to specify geometry, and the MCNP code can produce LNK3DNT files representing a user-defined region of the MCNP constructive solid geometry (CSG). Because of the use of LNK3DNT files as MCNP input and output and with other computer codes, improving the ability to visualize and/or quantitatively post-process LNK3DNT files benefits end users. In particular, it is helpful to visually verify that the material and density distributions are as expected. Also, an analyst can orient viewers with respect to other in formation, e.g., by co-plotting a voxelized representation of MCNP CSG with mesh tally and/or particle track results. However, until now, the only capability to interactively interrogate LNK3DNT files is through a 2-D slice plotter provided by the Monte Carlo Application ToolKit (MCATK), which is not currently distributed outside of Los Alamos National Laboratory. The utility created as part of this work converts LNK3DNT binary files to an ASCII XML-based StructuredMesh VTK (.vts) file. The geometries supported are 1-D spherical and cylindrical ( r ), 1-D Cartesian ( x ), 2-D cylindrical ( r-z and r-θ ) and Cartesian ( x-y ), and 3-D cylindrical ( r-z-θ ) and Cartesian ( x-y-z ). This file format is chosen to minimize external dependencies (e.g., on an HDF5 library) and to provide a file that can be interrogated by the user in a text editor or standard visualization applications (e.g., ParaView or VisIt). However, other file formats may provide more utility such as XDMF, which can use HDF5 and is discussed further as something to explore as future work. This paper is organized as follows: a brief introduc tion to the MCNPTools library and utility programs is given, the LNK3DNT-to-VTK (l3d2vtk) utility functionality is de scribed, and then an example post-processing workflow using the ParaView application is given.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Disease Prediction Models and Operational Readiness

INTRODUCTION: The objective of this manuscript is to present a systematic review of biosurveillance models that operate on select agents and can forecast the occurrence of a disease event. One of the primary goals of this research was to characterize the viability of biosurveillance models to provide operationally relevant information for decision makers to identify areas for future research. Two critical characteristics differentiate this work from other infectious disease modeling reviews. First, we reviewed models that attempted to predict the disease event, not merely its transmission dynamics. Second, we considered models involving pathogens of concern as determined by the US National Select Agent Registry (as of June 2011). Methods: We searched dozens of commercial and government databases and harvested Google search results for eligible models utilizing terms and phrases provided by public health analysts relating to biosurveillance, remote sensing, risk assessments, spatial epidemiology, and ecological niche-modeling, The publication date of search results returned are bound by the dates of coverage of each database and the date in which the search was performed, however all searching was completed by December 31, 2010. This returned 13,767 webpages and 12,152 citations. After de-duplication and removal of extraneous material, a core collection of 6,503 items was established and these publications along with their abstracts are presented in a semantic wiki at http://BioCat.pnnl.gov. Next, PNNL’s IN-SPIRE visual analytics software was used to cross-correlate these publications with the definition for a biosurveillance model resulting in the selection of 54 documents that matched the criteria resulting Ten of these documents, However, dealt purely with disease spread models, inactivation of bacteria, or the modeling of human immune system responses to pathogens rather than predicting disease events. As a result, we systematically reviewed 44 papers and the results are presented in this analysis.

biosecurity↗

Data analysis and modeling pipelines for controlled networked social science experiments

There is large interest in networked social science experiments for understanding human behavior at-scale. Significant effort is required to perform data analytics on experimental outputs and for computational modeling of custom experiments. Moreover, experiments and modeling are often performed in a cycle, enabling iterative experimental refinement and data modeling to uncover interesting insights and to generate/refute hypotheses about social behaviors. The current practice for social analysts is to develop tailor-made computer programs and analytical scripts for experiments and modeling. This often leads to inefficiencies and duplication of effort. In this work, we propose a pipeline framework to take a significant step towards overcoming these challenges. Our contribution is to describe the design and implementation of a software system to automate many of the steps involved in analyzing social science experimental data, building models to capture the behavior of human subjects, and providing data to test hypotheses. The proposed pipeline framework consists of formal models, formal algorithms, and theoretical models as the basis for the design and implementation. We propose a formal data model, such that if an experiment can be described in terms of this model, then our pipeline software can be used to analyze data efficiently. The merits of the proposed pipeline framework is elaborated by several case studies of networked social science experiments.

97 MATHEMATICS AND COMPUTING↗

Ground Magnetic Data for West-Central Colorado

Modeled ground magnetic data was extracted from the Pan American Center for Earth and Environmental Studies database at http://irpsrvgis08.utep.edu/viewers/Flex/GravityMagnetic/GravityMagnetic_CyberShare/ on 2/29/2012. The downloaded text file was then imported into an Excel spreadsheet. This spreadsheet data was converted into an ESRI point shapefile in UTM Zone 13 NAD27 projection, showing location and magnetic field strength in nano-Teslas. This point shapefile was then interpolated to an ESRI grid using an inverse-distance weighting method, using ESRI Spatial Analyst. The grid was used to create a contour map of magnetic field strength.

15 GEOTHERMAL ENERGY↗

Gravity Data for West-Central Colorado

Modeled Bouger-Corrected Gravity data was extracted from the Pan American Center for Earth and Environmental Studies Gravity Database of the U.S. at http://irpsrvgis08.utep.edu/viewers/Flex/GravityMagnetic/GravityMagnetic_CyberShare/ on 2/29/2012. The downloaded text file was opened in an Excel spreadsheet. This spreadsheet data was then converted into an ESRI point shapefile in UTM Zone 13 NAD27 projection, showing location and gravity (in milligals). This data was then converted to grid and then contoured using ESRI Spatial Analyst. Data from From University of Texas: Pan American Center for Earth and Environmental Studies

15 GEOTHERMAL ENERGY↗

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↗