Search

Parsing untrusted data is notoriously challenging. Failure to handle maliciously crafted data correctly can (and does) lead to a wide range of vulnerabilities. The Language-theoretic security (LangSec) philosophy seeks to obviate the need for developers to apply ad hoc solutions by, instead, offering formally correct and verifiable input handling throughout the software development lifecycle. One of the key components in developing secure parsers is a broad coverage corpus that enables developers to understand the problem space for a given format and to use, potentially, as seeds for fuzzing and other automated testing. In this paper, we offer an update on work initially reported at the LangSec 2020 conference on the development of a file observatory to gather and enable analysis on a diverse collection of files at scale. The initial focus of the observatory is on Portable Document Format (PDF) files and file formats typically embedded in PDFs. In this paper, we report on the addition of a bug tracker corpus and new analytic methods.

The development and deployment of machine learning (ML) systems can be executed easily with modern tools, but the process is typically rushed and means-to-an-end. The lack of diligence can lead to technical debt, scope creep and misaligned objectives, model misuse and failures, and expensive consequences. Engineering systems, on the other hand, follow well-defined processes and testing standards to streamline development for high-quality, reliable results. The extreme is spacecraft systems, where mission critical measures and robustness are ingrained in the development process. Drawing on experience in both spacecraft engineering and ML (from research through product across domain areas), we have developed a proven systems engineering approach for machine learning development and deployment. Our Machine Learning Technology Readiness Levels (MLTRL) framework defines a principled process to ensure robust, reliable, and responsible systems while being streamlined for ML workflows, including key distinctions from traditional software engineering. Even more, MLTRL defines a lingua franca for people across teams and organizations to work collaboratively on artificial intelligence and machine learning technologies. Here we describe the framework and elucidate it with several real world use-cases of developing ML methods from basic research through productization and deployment, in areas such as medical diagnostics, consumer computer vision, satellite imagery, and particle physics.

While there are more than 7000 languages in the world, most translation research efforts have targeted a few high-resource languages. Commercial translation systems support only one hundred languages or fewer, and do not make these models available for transfer to low resource languages. In this work, we present useful tools for machine translation research: MTData, NLCodec, and RTG. We demonstrate their usefulness by creating a multilingual neural machine translation model capable of translating from 500 source languages to English. We make this multilingual model readily downloadable and usable as a service, or as a parent model for transfer-learning to even lower-resource languages., Comment: To-appear: ACL 2021 System Demonstrations

MAARS (Machine leaning-based Analytics for Automated Rover Systems) is an ongoing JPL effort to bring the latest self-driving technologies to Mars, Moon, and beyond. The ongoing AI revolution here on Earth is finally propagating to the red planet as the High Performance Spaceflight Computing (HPSC) and commercial off-the-shelf (COTS) system-on-a-chip (SoC), such as Qualcomm's Snapdragon, become available to rovers. In this three year project, we are developing, implementing, and benchmarking a wide range of autonomy algorithms that would significantly enhance the productivity and safety of planetary rover missions. This paper is to provide the latest snapshot of the project with broad and high-level description of every capability that we are developing, including scientific scene interpretation, vision-based traversability assessment, resource-aware path planning, information-theoretic path planning, on-board strategic path planning, and on-board optimal kinematic settling for accurate collision checking. All of the onboard software capabilities will be integrated into JPL's Athena test rover using ROS (Robot Operating System).

Nowcasting of precipitation is a difficult spatiotemporal task because of the non-uniform characterization of meteorological structures over time. Recently, convolutional LSTM has been shown to be successful in solving various complex spatiotemporal based problems. In this research, we propose a novel precipitation nowcasting architecture 'Convcast' to predict various short-term precipitation events using satellite data. We train Convcast with ten consecutive NASA's IMERG precipitation data sets each at intervals of 30 minutes. We use the trained neural network model to predict the eleventh precipitation data of the corresponding ten precipitation sequence. Subsequently, the predicted precipitation data are used iteratively for precipitation nowcasting of up to 150 minutes lead time. Convcast achieves an overall accuracy of 0.93 with an RMSE of 0.805 mm/h for 30 minutes lead time, and an overall accuracy of 0.87 with an RMSE of 1.389 mm/h for 150 minutes lead time. Experiments on the test dataset demonstrate that Convcast consistently outperforms other state-of-the-art optical flow based nowcasting algorithms. Results from this research can be used for nowcasting of weather events from satellite data as well as for future on-board processing of precipitation data.

Author(s): Ramapriyan, Hampapuram; Downs, Robert; Dozier, Jeff; Duerr, Ruth; Folk, Mike; Frew, James; Hoebelheinrich, Nancy; Mattmann, Chris; Peng, Ge | Abstract: Bruce R. Barkstrom, principal investigator for NASA missions involved with understanding Earth’s radiation budget, committed his life to analyzing, interpreting, and stewarding Earth science data.

Snow cover and its melt dominate regional climate and water resources in many of the world's mountainous regions. Snowmelt timing and magnitude in mountains are controlled predominantly by absorption of solar radiation and the distribution of snow water equivalent (SWE), and yet both of these are very poorly known even in the best-instrumented mountain regions of the globe. Here we describe and present results from the Airborne Snow Observatory (ASO), a coupled imaging spectrometer and scanning lidar, combined with distributed snow modeling, developed for the measurement of snow spectral albedo/broadband albedo and snow depth/SWE. Snow density is simulated over the domain to convert snow depth to SWE. The result presented in this paper is the first operational application of remotely sensed snow albedo and depth/SWE to quantify the volume of water stored in the seasonal snow cover. The weekly values of SWE volume provided by the ASO program represent a critical increase in the information available to hydrologic scientists and resource managers in mountain regions.

The knowledge we gain from research in climate science depends on the generation, dissemination, and analysis of high-quality data. This work comprises technical practice as well as social practice, both of which are distinguished by their massive scale and global reach. As a result, the amount of data involved in climate research is growing at an unprecedented rate. Some examples of the types of activities that increasingly require an improved cyberinfrastructure for dealing with large amounts of critical scientific data are climate model intercomparison (CMIP) experiments; the integration of observational data and climate reanalysis data with climate model outputs, as seen in the Observations for Model Intercomparison Projects (Obs4MIPs), Analysis for Model Intercomparison Projects (Ana4MIPs), and Collaborative Reanalysis Technical Environment-Intercomparison Project (CREATE-IP) activities; and the collaborative work of the Intergovernmental Panel on Climate Change (IPCC). This article provides an overview of some of climate science's big data problems and the technical solutions being developed to advance data publication, climate analytics as a service, and interoperability within the Earth System Grid Federation (ESGF), which is the primary cyberinfrastructure currently supporting global climate research activities.

Planetary exploration is full of challenges. Data bandwidth is very limited between planetary rovers and ground-based data system. What’s worse, even though NASA has accumulated over 34 million images from various missions, it requires significant effort and is hardly possible for any scientist to go through all of them. In order to improve the degree of automation and the efficiency of these processes, we propose a system leveraging machine learning for planetary rovers to actively look for scientifically interesting and valuable features according to text instructions from scientists and prioritize the images captured onboard with those features for downlink. Such an image prioritization mechanism can also be naturally applied to content-based image search through text description in any local planetary image data server, allowing scientists to search for images with desired features without going through them one by one. Besides theoretical and engineering details of our proposed approach, we also present both quantitative and qualitative evaluation of the system along with some concrete examples.

Transferring large volumes of information from one location to potentially many others that are geographically distributed and across varying networks is still prevalent in modern scientific data systems. This is despite the movement to push computation to the data and to reduce data movement needed to compute answers to challenging scientific problems, to disseminate information to the scientific community, and to acquire data for curation and enrichment. Because of this, it is imperative that decisions made regarding data movement systems and architectures be backed by both analytical rigor, and also by empirical evidence and measurement. The purpose of this study is to expand on the work performed by our research team over the last decade and to take a fresh look at the evaluation of multiple topical data transfer technologies in use cases derived from data-intensive scientific systems and applications in the areas of Earth science. We report on the evaluation of a set of data movement technologies against a set of empirically derived comparison dimensions. Based on this evaluation, we make recommendations towards the selection of appropriate data movement technologies in scientific applications and scenarios.

Uncertainty in calculating the spatial- and interannual variability of precipitation over India and Tibet from widely used gridded precipitation datasets is examined for the 29-year period from 1979 to 2007. Uncertainty is defined in terms of the spread among the variability calculated from multiple datasets, a useful method when multiple datasets of similar or unknown accuracy are available for analyses. The resulting uncertainty varies for regions and seasons. Geographical variations are clearly seen in the signal-to-noise ratio (SNR), with the largest uncertainty in the Jammu and Kashmir (J&K), western India-eastern Pakistan, Tibet, Hindu-Kush mountains, and Western Ghats which are characterized by either dry climate or complex terrain, or both. Seasonally, the uncertainty is larger for the December–February period (DJF) than for the June–September period (JJAS) in most of the region except J&K which is characterized by two wet seasons: winter and summer. The uncertainty in the interannual variability also varies according to regions especially in J&K where the calculated interannual variability varies by nearly a factor of two among the datasets. The uncertainty range in the calculated interannual variability is determined largely by two gauge-based data of the finest resolution, Asian Precipitation – Highly-Resolved Observational Data Integration Towards Evaluation of water resources (smallest) and India Meteorological Department (largest) in all regions. The regional and seasonal variations in the uncertainty do not appear to depend on either the spatial resolution or the length of records. This implies that analysis methodology such as the quality control of input data, spatial/temporal interpolation, and retrieval algorithms used in producing these gridded datasets plays a crucial role in determining the characteristics of precipitation climatology represented by individual datasets. Our results show that calculating precipitation characteristics must be accompanied by careful examinations of uncertainty among available datasets, especially for dry seasons and arid/mountainous regions.

Methodology is developed and applied to evaluate the characteristics of daily surface temperature distributions in a six-member regional climate model (RCM) hindcast experiment conducted as part of the North American Regional Climate Change Assessment Program (NARCCAP). A surface temperature dataset combining gridded station observations and reanalysis is employed as the primary reference. Temperature biases are documented across the distribution, focusing on the median and tails. Temperature variance is generally higher in the RCMs than reference, while skewness is reasonably simulated in winter over the entire domain and over the western United States and Canada in summer. Substantial differences in skewness exist over the southern and eastern portions of the domain in summer. Four examples with observed long-tailed probability distribution functions (PDFs) are selected for model comparison. Long cold tails in the winter are simulated with high fidelity for Seattle, Washington, and Chicago, Illinois. In summer, the RCMs are unable to capture the distribution width and long warm tails for the coastal location of Los Angeles, California, while long cold tails are poorly realized for Houston, Texas. The evaluation results are repeated using two additional reanalysis products adjusted by station observations and two standard reanalysis products to assess the impact of observational uncertainty. Results are robust when compared with those obtained using the adjusted reanalysis products as reference, while larger uncertainties are introduced when standard reanalysis is employed as reference. Model biases identified in this work will allow for further investigation into associated mechanisms and implications for future simulations of temperature extremes.

A domain-specific software architecture (DSSA) represents an effective, generalized, reusable solution to constructing software systems within a given application domain. In this paper, we revisit the widely cited DSSA for the domain of grid computing. We have studied systems in this domain over the last ten years. During this time, we have repeatedly observed that, while individual grid systems are widely used and deemed successful, the grid DSSA is actually underspecified to the point where providing a precise answer regarding what makes a software system a grid system is nearly impossible. Moreover, every one of the existing purported grid technologies actually violates the published grid DSSA. In response to this, based on an analysis of the source code, documentation, and usage of eighteen of the most pervasive grid technologies, we have significantly refined the original grid DSSA. We demonstrate that this DSSA much more closely matches the grid technologies studied. Our refinements allow us to more definitively identify a software system as a grid technology, and distinguish it from software libraries, middleware, and frameworks.

We describe efforts to bring new methods of search analytics, machine learning, natural language processing and data visualization to address the challenge of finding and extracting meaning from unstructured text and multimedia content. We use the Polar domain to motivate the problem and our proposed solution. However, our techniques are applicable and scalable to other domains., This work was supported by the DARPA XDATA/Memex program and by the NSF EarthCube program under award ICER- 1639753. In addition, the NSF Polar Cyberinfrastructure award numbers PLR-1348450 and PLR-144562 funded a portion of the work. Effort supported in part by JPL, managed by the California Institute of Technology on behalf of NASA.

This paper describes the applications of deep learning-based image recognition in the DARPA Memex program and its repository of 1.4 million weapons-related images collected from the Deep web. We develop a fast, efficient, and easily deployable framework for integrating Google's Tensorflow framework with Apache Tika for automatically performing image forensics on the Memex data. Our framework and its integration are evaluated qualitatively and quantitatively and our work suggests that automated, large-scale, and reliable image classification and forensics can be widely used and deployed in bulk analysis for answering domain-specific questions.

Mesoscale convective systems are high impact convectively driven weather systems that contribute large amounts to the precipitation daily and monthly totals at various locations globally. As such, an understanding of the lifecycle, characteristics, frequency and seasonality of these convective features is important for several sectors and studies in climate studies, agricultural and hydrological studies, and disaster management. This study explores the applicability of graph theory to creating a fully automated algorithm for identifying mesoscale convective systems and determining their precipitation characteristics from satellite datasets. Our results show that applying graph theory to this problem allows for the identification of features from infrared satellite data and the seamlessly identification in a precipitation rate satellite-based dataset, while innately handling the inherent complexity and non-linearity of mesoscale convective systems.

This article includes the following "mini-articles" about applications and examples of computing in astronomy: "Visualizing the Universe: Using Modern Graphics Cards to Understand the Physical World," by F. Alexander Bogert, Nicholas Smith, and John Holdener; "Visualizing Big Data in Astronomy: The Automated Movie Production Environment Distribution and Display (AMPED) Pipeline," by Eric M. De Jong; "Supporting Distributed, Collaborative Review and Classification of Fast Transient Events," by Andrew F. Hart, Luca Cinquini, Shakeh E. Khudikyan, David R. Thompson, Chris A. Mattmann, Kiri Wagstaff, Joseph Lazio, and Dayton L. Jones; "Big Data Technologies at JPL," by Dayton L. Jones; and "The Astrophysics Source Code Library: ascl.net," by Alice Allen, Lior Shamir, and Peter Teuben.

This study evaluates model fidelity in simulating relationships between seasonally averaged precipitation, cloud fraction and surface insolation from the North American Regional Climate Change Assessment Project (NARCCAP) hindcast using observational data from ground stations and satellites. Model fidelity is measured in terms of the temporal correlation coefficients between these three variables and the similarity between the observed and simulated joint probability distribution functions (JPDFs) in 14 subregions over the conterminous United States. Observations exhibit strong negative correlations between precipitation/cloud fraction and surface insolation for all seasons, whereas the relationship between precipitation and cloud fraction varies according to regions and seasons. The skill in capturing these observed relationships varies widely among the NARCCAP regional climate models, especially in the Midwest and Southeast coast regions where observations show weak (or even negative) correlations between precipitation and cloud fraction in winter due to frequent non-precipitating stratiform clouds. Quantitative comparison of univariate and JPDFs indicates that model performance varies markedly between regions as well as seasons. This study also shows that comparison of JPDFs is useful for summarizing the performance of and highlighting problems with some models in simulating cloud fraction and surface insolation. Our quantitative metric may be useful in improving climate models by highlighting shortcomings in the formulations related with the physical processes involved in precipitation, clouds and radiation or other multivariate processes in the climate system.

We present further work on SciSpark, a Big Data framework that extends Apache Spark's inmemory parallel computing to scale scientific computations. SciSpark's current architecture and design includes: time and space partitioning of highresolution geo-grids from NetCDF3/4; a sciDataset class providing N-dimensional array operations in Scala/Java and CF-style variable attributes (an update of our prior sciTensor class); parallel computation of time-series statistical metrics; and an interactive front-end using science (code & visualization) Notebooks. We demonstrate how SciSpark achieves parallel ingest and time/space partitioning of Earth science satellite and model datasets. We illustrate the usability, extensibility, and early performance of SciSpark using several Earth science Use cases, here presenting benchmarks for sciDataset Readers and parallel time-series analytics. A three-hour SciSpark tutorial was taught at an ESIP Federation meeting using a dozen “live” Notebooks.