The ubiquity of sensors and IoT devices has led to an explosion in data availability in modern agriculture. The large volume and heterogeneity of the data, together with the complexity of data processing requirements, pose huge obstacles for achieving the principles of Findable, Accessible, Interoperable, and Reusable (FAIR). Current data management and analysis paradigms are to a large extent traditional, in which data collecting, curating, integration, loading, storing, sharing and analyzing still involve too much human effort and know-how. The experts, researchers and the farm operators need to understand the data and the whole process of data management pipeline to make full use of the data. The essential problem of the traditional paradigm is the lack of a layer of orchestrational intelligence which can understand, organize and coordinate the data processing utilities to maximize data management and analysis outcome. The emerging reasoning and tool mastering abilities of large language models (LLM) make it a potentially good fit to this position, which helps a shift from the traditional user-driven paradigm to AI-driven paradigm. In this paper, we propose and explore the idea of a LLM based copilot for autonomous agricultural data management and analysis. Based on our previously developed platform of Agricultural Data Management and Analytics (ADMA), we build a proof-of-concept multi-agent system called ADMA Copilot, which can understand user’s intent, makes plans for data processing pipeline and accomplishes tasks automatically, in which three agents: a LLM based controller, an input formatter and an output formatter collaborate together. Different from existing LLM based solutions, by defining a meta-program graph, our work decouples control flow and data flow to enhance the predictability of the behavior of the agents. Experiments demonstrate the intelligence, autonomy, efficacy, efficiency, extensibility, flexibility and privacy of our system. Comparison is also made between ours and existing systems to show the superiority and potential of our system.

Explanations of classifiers can provide recourse to those impacted, i.e., the ability to facilitate those impacted to understand and potentially change the classification. Contrastive explanations provide this recourse by producing an alternative input close to the original such that the label is changed to the desired, implying actions in the difference between the original and contrastive inputs. However, quantitatively evaluating contrastive explanations remains a challenging task. In particular, some state-of-the-art contrastive explanation algorithms for decisions made by deep neural networks can produce inputs that are out-of-distribution or adversarial and are thus either infeasible for a user to achieve or do not change the label according to the underlying data distribution, respectively. Past work has termed contrastive examples valid if they cross the decision boundary of the classifier. However, this definition does not encompass these failure modes and thus is not suitable to evaluate contrastive methods when used for recourse. In this paper, we define a new type of validity, called distributional validity, that checks for these failure modes. We experiment with the distributional validity of state-of-the-art contrastive explanation methods and find that the best contrastive method depends on the architecture of the classification model.

Approximate Nearest Neighbor (ANN) search has become a fundamental operation in numerous applications, including recommendation systems, computer vision, and natural language processing. The advent of Large Language Models (LLMs) arouses new interest in developing more efficient ANN algorithms, which will be the core functionality of vector databases as long-term memory of LLM. Multiple types of index structures, such as hashing-based, tree-based, and quantization-based, have been developed for ANN, and recently, graph-based algorithms have become the SOTA paradigm with the best trade-off between recall rate and query latency. However, almost all the existing graph-based index structures can only be hosted in memory due to the otherwise frequent I/O operations during searching if the graph-based index is stored on disk. The problem follows that for extremely large datasets, it is infeasible to accommodate the whole graph-based index in memory, and furthermore, it is difficult to build the whole index in memory at once. Thus, it is favorable if the graph-based index can be stored purely on disk and loaded into memory on demand during searching on the graph. There are existing efforts, such as DiskANN, which try to store graph-based index structure on disk while still keeping a compressed version of the dataset in memory to reduce disk I/O and speed up distance calculation. In this paper, we introduce LM-DiskANN, a novel dynamic graph-based ANN index that is designed specifically to be hosted on disk while keeping a low memory footprint by storing complete routing information in each node. By conducting extensive experiments on multiple benchmark datasets, we demonstrate that LM-DiskANN achieves a similar recall-latency curve while consuming much less memory compared with SOTA graph-based ANN indexes. Furthermore, its scalability and adaptability make it a promising solution for future big data applications.

Modern agriculture faces grand challenges to meet increased demands for food, fuel, feed, and fiber with population growth under the constraints of climate change and dwindling natural resources. Data innovation is urgently required to secure and improve the productivity, sustainability, and resilience of our agroecosystems. As various sensors and Internet of Things (IoT) instrumentation become more available, affordable, reliable, and stable, it has become possible to conduct data collection, integration, and analysis at multiple temporal and spatial scales, in real-time, and with high resolutions. At the same time, the sheer amount of data poses a great challenge to data storage and analysis, and the de facto data management and analysis practices adopted by scientists have become increasingly inefficient. Additionally, the data generated from different disciplines, such as genomics, phenomics, environment, agronomy, and socioeconomic, can be highly heterogeneous. That is, datasets across disciplines often do not share the same ontology, modality, or format. All of the above make it necessary to design a new data management infrastructure that implements the principles of Findable, Accessible, Interoperable, and Reusable (FAIR). In this paper, we propose Agriculture Data Management and Analytics (ADMA), which satisfies the FAIR principles. Our new data management infrastructure is intelligent by supporting semantic data management across disciplines, interactive by providing various data management/analysis portals such as web GUI, command line, and API, scalable by utilizing the power of high-performance computing (HPC), extensible by allowing users to load their own data analysis tools, trackable by keeping track of different operations on each file, and open by using a rich set of mature open source technologies.

Medical image synthesis is important in diverse healthcare applications, such as computer-aided diagnosis, medical image analysis, and educational tools. While Generative Adversarial Networks (GANs) have shown remarkable success in generating natural images, their application to medical images often falls short in faithfully capturing essential anatomical features. In this paper, we introduce a new approach that focuses on tissue-specific color encoding to enhance medical image synthesis using GANs. Our method deviates from the conventional practice of directly training GANs on gray-scale medical images. Instead, we initiate the process by generating and encoding various gray-scale representations of distinct tissues into separate color channels within composite images. These tissue-specific color images are then utilized to train a GAN model. The GAN, once trained, excels in producing high-quality synthetic images for individual tissues, and when combined, these tissue images yield final synthesized images that better portray the intricate tissue characteristics found in medical data. We have conducted an experimental study to validate the effectiveness of our approach in comparison to alternative methods with both qualitative and quantitative assessments to evaluate the quality of synthesized individual tissues and their combined final results.

Cover crops offer a range of agricultural and environmental benefits, such as reducing soil erosion, increasing carbon and enhancing water storage, increasing forage production, protecting soil nutrients, and so on. However, adoption of cover crop farming remains limited among Nebraska’s farmers. To promote awareness of the value of cover crop farming, we harness modern tools and technology and develop a new web-based tool capable of quantifying potential forage production, forage quality, and environmental benefits when planting cover crops, taking into account factors like climate, soil types, and seeding periods. Our tool incorporates the concept of data provenance to capture simulation configurations and results. This implementation can enhance data integrity and facilitate knowledge sharing within the scientific community, supporting further research and broader public benefits. The tool also includes reporting functions with visualizations illustrating distributions of potential forage, transpiration, nitrogen uptake, and more. Based on factors such as cover crop types, planting and termination dates, locations, and soil types, our tool provides valuable insights, enabling farmers to experiment with different cover crops on their land, ultimately leading to improved environmental outcomes for the broader Nebraska community.

Considering the challenges posed by the space and time complexities in handling extensive scientific volumetric data, various data representations have been developed for the analysis of large-scale scientific data. Multivariate functional approximation (MFA) is an innovative data model designed to tackle substantial challenges in scientific data analysis. It computes values and derivatives with high-order accuracy throughout the spatial domain, mitigating artifacts associated with zero- or first-order interpolation. However, the slow query time through MFA makes it less suitable for interactively visualizing a large MFA model. In this work, we develop the first scalable interactive volume visualization pipeline, MFA-DVV, for the MFA model encoded from large-scale datasets. Our method achieves low input latency through distributed architecture, and its performance can be further enhanced by utilizing a compressed MFA model while still maintaining a high-quality rendering result for scientific datasets. We conduct comprehensive experiments to show that MFA-DVV can decrease the input latency and achieve superior visualization results for big scientific data compared with existing approaches.

Soil measurement and evaluation are crucial to various aspects of agriculture, including agricultural productivity, nutrient management, water management, and pH Regulation. Hyperspectral imaging is an advanced technique used to capture and analyze a wide range of light wavelengths (or spectral bands) across the electromagnetic spectrum. Hyperspectral imaging in soil research involves the use of this advanced imaging technique to analyze the spectral properties of soils. It allows researchers to capture detailed information about the composition, texture, and conditions of soil across a wide range of wavelengths in the electromagnetic spectrum. This in-depth spectral analysis provides valuable insights for studying soil health, nutrient content, moisture levels, and other critical parameters. However, existing hyperspectral analysis of soil relies on using imaging systems to exclusively capture information from the soil surface. This yields a two-dimensional image in which each pixel represents a spectrum vector. In this paper, we provide a new 3D hyperspectral data capturing features deep into the soil where each voxel represents a spectrum vector. For effective analysis of this type of new hyperspectral data, we develop a 3D visualization tool to not only directly visualize individual spectrum of the soil volume but also provide a way to cluster such high dimensional data leveraging a deep learning-based method through autoencoder.

We present a novel web-based framework, named Pixel-Based Edge Bundling (PBEB), for effectively and interactively visualizing large graphs. Our framework combines an image-based edge-bundling method and a parallel texture-based processing scheme, allowing us to effectively and efficiently compute edge similarities using kernel density estimation and subsequently group these edges into bundles based on their similarities. We discuss several challenges related to developing large-graph visualization on web-based platforms. To accelerate the edge bundling process and enable interactivity in web-based environments, we leverage texture-based parallel processing, a standard feature of WebGL. Our framework optimizes an end-to-end process, from bundling to rendering, enabling practical and interactive visualization of large graphs in a web-based setting. We demonstrate the superior performance of our framework by conducting comparisons with existing web-based and CUDA-based edge-bundling methods using various standard graphics cards on different devices.

Deep Neural Networks (DNNs) have performed admirably in classification tasks. However, the characterization of their classification uncertainties, required for certain applications, has been lacking. In this work, we investigate the issue by assessing DNNs’ ability to estimate conditional probabilities and propose a framework for systematic uncertainty characterization. Denoting the input sample as x and the category as y, the classification task of assigning a category y to a given input x can be reduced to the task of estimating the conditional probabilities p(y|x), as approximated by the DNN at its last layer using the softmax function. Since softmax yields a vector whose elements all fall in the interval (0, 1) and sum to 1, it suggests a probabilistic interpretation to the DNN’s outcome. Using synthetic and real-world datasets, we look into the impact of various factors, e.g., probability density f(x) and inter-categorical sparsity, on the precision of DNNs’ estimations of p(y|x), and find that the likelihood probability density and the inter-categorical sparsity have greater impacts than the prior probability to DNNs’ classification uncertainty.

Studying the growth dynamics of developing plants is of critical importance in plant sciences. The traditional methods rely on either manual measurement, which involves tedious labor work, or 2D image-based approaches, which cannot fully characterize plants in 3D. Given the advances of scanners and 3D reconstruction methods, scientists begin to pay more attention to 3D models to improve accuracy. However, existing methods mostly focus on the growth of a whole plant rather than its detailed substructures. In this paper, we have developed an end-to-end pipeline to detect the key events on both the whole plant and the specific components. Our method is achieved by building 3D models from images, segmenting individual components, and capturing traits. We implement an experiment on maizes for evaluation and successfully detect events in the process of growth.

Nowadays, floating-point temporal-spatial datasets are routinely generated from scientific observational apparatuses or computer simulations at an unprecedented pace. The sheer amount of these large volumetric datasets on the order of terabytes or petabytes consume massive resources in terms of bandwidth, storage and computational power. On the other hand, scientists, equipped with low-end post-analysis machines, often find it impossible to visualize and analyze these massive datasets with such limited resources in hand, not to mention their ultimate goal of real time analysis and visualization. To solve this discrepancy, a compact data representation has to be generated and a trade-off between resource consumption and analytical precision has to be found. There are many existing volumetric representation generating methods, almost all of which adopts some kind of hand-engineered heuristics to extract the effective portion of the datasets. However, the trade-off between resource consumption and analytical quality could not be well established due to the introduction of hand-engineered heuristics. In this paper, we present a deep learning based method that can adaptively capture the inherently complicated dynamics of temporal-spatial volumetric datasets without introducing any hand engineered features. We train an autoencoder based neural network with quantization and adaptation. Compared with existing methods, our method could learn data representation at a much lower compressed/uncompressed rate while preserving the details of original datasets. Also, our method could adapt with different data distribution and conduct compression and decompression in real time. Through extensive experiments, we show the effectiveness and efficiency of our approach over existing methods.

We explore the usage of deep convolutional neural network for clustering the time steps of a spatial-temporal scientific dataset. Our approach first takes the scientific dataset as training data and trains a deep convolutional autoencoder. A low-dimensional feature space or latent space can be extracted by inferencing the encoding part of the network. As a result, each time step is transformed into a feature descriptor that can be compared with each other in the feature space. In this way, we can cluster time steps according to their feature descriptors, and each group of time steps has a similar characterization. We demonstrate the effectiveness of our approach using a real-world simulation dataset of water contamination. Multiple variables and their combinations of this dataset are fed into our approach. The trained network enables the clustering of the time steps and facilitates scientists to examine their large spatial-temporal datasets.

In this paper, we present a unified graph clustering framework based on an asynchronous approach. We study the similarities among the Louvain algorithm and the Infomap algorithm. Based on their common features, we build an end-to-end optimized distributed framework for implementing both algorithms. By extending the existing asynchronous distributed framework for large-scale graphs traversal, we ensure both workload and communication balanced. Our extensive experiments show that our framework is correct and effective with different large real-world and synthetic datasets using up to 32,768 processors for the Louvain algorithm and 16,384 processors for the Infomap algorithm. The quality and the scalability of our framework are superior to the existing work.

In large-scale enterprise data centers for big data analytics, long batched jobs and short interactive jobs are usually mixed. Hybrid job schedulers, consisting of one centralized scheduler for long jobs and multiple distributed schedulers for short jobs, have become a promising alternative because they can significantly shorten latencies of short jobs via independent and parallelized assignment of short tasks by distributed schedulers and lower chances of head-of-line blocking via a number of performance optimization techniques.However, short jobs are still facing long job latencies under hybrid job schedulers due to workload fluctuation and straggler task problem. In this paper, we propose Eirene to optimize the latency performance of short jobs via two schemes tightly coupled into the general architecture of hybrid job schedulers. Coordinated Cold Data Migration leverages high task waiting time of short jobs under heavily-loaded periods and migrates cold data from disks to local memory for the initial phase of reading input so as to shorten task runtime and queueing time. On the other hand, Scheduler-Aware Task Cloning exploits spare computing resources under lightly-loaded periods and performs proactive task cloning for short jobs to mitigate the straggler problem.We implement a prototype of Eirene based on Eagle, a state-of-the-art hybrid job scheduler. Experimental results show that, under heavy loads, Eirene is able to improve 50-percentile (P50), 75-percentile (P75), 90-percentile (P90) latency performance of short jobs by up to 44.4%, 80.3%, 84.1% respectively compared with Eagle under the Facebook trace with a cluster of 50000 nodes.

Analysis of hyperspectral images is of great importance in many scientific disciplines. Obtaining the spectral and spatial information simultaneously from time-varying hyperspectral images is a challenging task due to their high dimensionality. In this paper, we design an interface that allows users to study hyperspectral images interactively and obtain spectral features and enhanced images at the same time. The image fusion results change dynamically with the regions of interest selected by users and convey both the spatial and spectral information. We show the usefulness of our approach using time-varying hyperspectral plant images. We compare our method with existing hyperspectral image analysis techniques. Our evaluation indicates that our interface can help users determine important bands, identify regions of interest, and generate image fusion results for time-varying hyperspectral plant images.

Generating 3D digital representations of plants is indispensable for researchers to gain a detailed understanding of plant dynamics. Emerging high-throughput plant phenotyping techniques can capture plant point clouds that, however, often contain imperfections and make it a changeling task to generate accurate 3D reconstructions. We present an end-to-end pipeline to reconstruct surfaces from point clouds of maize and rice plants. In particular, we propose a two-step clustering approach to accurately segment the points of each individual plant component according to maize and rice properties. We further employ surface fitting and edge fitting to ensure the smoothness of resulting surfaces. Realistic visualization results are obtained through post-processing, including texturing and lighting. Our experimental study has explored the parameter space and demonstrated the effectiveness of our pipeline for high-throughput plant phenotyping.

Medical imaging enables researchers and practitioner to uncover the characteristics of diseases (e.g., human cancer) in great detail. However, the sheer size of resulting imaging data and the high dimension of derived features become a major challenge in data analysis, diagnosis, and knowledge discovery. We present a novel visual analytics system, named iVAR, targeted at observing the comprehensive quantification of tumor phenotypes by effectively exploring a large number of quantitative image features. Our system is comprised of multiple linked views combining visualization of three-dimensional volumes and tumors reconstructed by computed tomography (CT) images, and a radiomic analysis of high-dimensional features quantifying tumor image intensity, shape and texture, and three non-image clinical features. Thus, it offers insights into the overall distribution of quantitative imaging features and also enables detailed analysis of the relationship between features. We demonstrate our system through use case scenarios on a real-world large-scale CT dataset with lung cancer.

We have devised and implemented a key technology, SpatioTemporal Adaptive-Resolution Encoding (STARE), in an array database management system, i.e. SciDB, to achieve unparalleled variety scaling for Big Earth Data, enabling rapid-response visual analytics. STARE not only serves as a unifying data representation homogenizing diverse varieties of Earth Science Datasets, but also supports spatiotemporal data placement alignment of these datasets to optimize a major class of Earth Science data analyses, i.e. those requiring spatiotemporal coincidence. Using STARE, we tailor a data partitioning and distribution strategy for the data access patterns of our scientific analysis, leading to optimal use of distributed resources. With STARE, rapid-response visual analytics are made possible through a high-level query interface, allowing geoscientists to perform data exploration visually, intuitively and interactively. We envision a system based on these innovations to relieve geoscientists of most laborious data management chores so that they may focus better on scientific issues and investigations. A significant boost in scientific productivity may thus be expected. We demonstrate these advantages with a prototypical system including comparisons to alternatives.

We investigate the impact of data placement on two Big Data technologies, Spark and SciDB, with a use case from Earth Science where data arrays are multidimensional. Simultaneously, this investigation provides an opportunity to evaluate the performance of the technologies involved. Two datastores, HDFS and Cassandra, are used with Spark for our comparison. It is found that Spark with Cassandra performs better than with HDFS, but SciDB performs better yet than Spark with either datastore. The investigation also underscores the value of having data aligned for the most common analysis scenarios in advance on a shared nothing architecture. Otherwise, repartitioning needs to be carried out on the fly, degrading overall performance.

We have implemented a flexible User Defined Operator (UDO) for labeling connected components of a binary mask expressed as an array in SciDB, a parallel distributed database management system based on the array data model. This UDO is able to process very large multidimensional arrays by exploiting SciDB's memory management mechanism that efficiently manipulates arrays whose memory requirements far exceed available physical memory. The UDO takes as primary inputs a binary mask array and a binary stencil array that specifies the connectivity of a given cell to its neighbors. The UDO returns an array of the same shape as the input mask array with each foreground cell containing the label of the component it belongs to. By default, dimensions are treated as non-periodic, but the UDO also accepts optional input parameters to specify periodicity in any of the array dimensions. The UDO requires four stages to completely label connected components. In the first stage, labels are computed for each subarray or chunk of the mask array in parallel across SciDB instances using the weighted quick union (WQU) with half-path compression algorithm. In the second stage, labels around chunk boundaries from the first stage are stored in a temporary SciDB array that is then replicated across all SciDB instances. Equivalences are resolved by again applying the WQU algorithm to these boundary labels. In the third stage, relabeling is done for each chunk using the resolved equivalences. In the fourth stage, the resolved labels, which so far are “flattened” coordinates of the original binary mask array, are renamed with sequential integers for legibility. The UDO is demonstrated on a 3-D mask of 0(10n) elements, with 0(108) foreground cells and o(106) connected components. The operator completes in 19 minutes using 84 SciDB instances.

We have implemented an updated Hierarchical Triangular Mesh (HTM) as the basis for a unified data model and an indexing scheme for geoscience data to address the variety challenge of Big Earth Data. In the absence of variety, the volume challenge of Big Data is relatively easily addressable with parallel processing. The more important challenge in achieving optimal value with a Big Data solution for Earth Science (ES) data analysis, however, is being able to achieve good scalability with variety. With HTM unifying at least the three popular data models, i.e. Grid, Swath, and Point, used by current ES data products, data preparation time for integrative analysis of diverse datasets can be drastically reduced and better variety scaling can be achieved. HTM is also an indexing scheme, and when applied to all ES datasets, data placement alignment (or co-location) on the shared nothing architecture, which most Big Data systems are based on, is guaranteed and better performance is ensured. With HTM most geospatial set operations become integer interval operations with further performance advantages.

We present a study of scientific data analytics on heterogeneous architectures using the Legion runtime system. Legion is a new programming model and runtime system targeting distributed heterogeneous architectures. It introduces logical regions as a new abstraction for describing the structures and usages of program data. We describe how to leverage logical regions to express important properties of program data, such as locality and independence, for scientific data analytics that can consist of multiple operations with different data types. Our approach can help users simplify programming on the data partition, data organization, and data movement for distributed-memory heterogeneous architectures, thereby facilitating a simultaneous execution of multiple analytics operations on modern and future supercomputers. We demonstrate the scalability and the usability of our approach by a hybrid data partitioning and distribution scheme for different data types using both CPUs and GPUs on a heterogeneous system.

We present a novel geohydrologic data visualization framework and apply the interface automata theory in support of time-varying multivariate data visualization tasks. The framework tackles heterogeneous geohydrologic data that has unique and complex data structures. The interface automata can generate a series of interactions and interfaces that are adapted to user selection and provide an intuitive method for visualizing and analyzing geohydrologic data. The interface automata can not only clearly guide user exploration, but also enhance user experience by eliminating automation surprises. In addition, our design can significantly reduce the entire system maintenance overhead, and enhance the system extendability for new datasets and data types. Our framework has been applied to a scientific geohydrologic visualization and analysis system, named INSIGHT, for the Nebraska Department of Natural Resources (NDNR). The new framework has brought many advantages that do not exist in the previous approaches, and is more efficient and extendable for visualizing geohydrologic data.

Directly visualizing a large graph as a node-link diagram often incurs visual clutter. Edge bundling can effectively address this issue and concisely reveal the main graph structure with reduced visual clutter. Although researchers have devoted noticeable efforts to develop acceleration methods, it remains a challenging task to efficiently conduct edge bundling on devices with a limited computing capacity, such as ubiquitous smart mobile devices. We present a new method for visualizing a node-link diagram based on force-directed edge bundling. We use textures to encode the data of lines and forces, and employ shaders to conduct the iterative line refinement on GPUs. We name this method as Texture-Based Edge Bundling (TBEB) as the major steps are done using textures. We demonstrate the high performance of TBEB using standard graphics cards. TBEB makes it feasible to interactively visualize large graphs on web-based platforms.