Search

In this paper, we present a novel and effective approach to truck and trailer classification, which integrates deep learning models and conventional image processing and computer vision techniques. The developed method groups trucks into subcategories by carefully examining the truck classes and identifying key geometric features for discriminating truck and trailer types. We also present three discriminating features that involve shape, texture, and semantic information to identify trailer types. Experimental results demonstrate that the developed hybrid approach can achieve high accuracy with limited training data, where the vanilla deep learning approaches show moderate performance due to over-fitting and poor generalization. Additionally, the models generated are human-understandable.

Camera-based systems are increasingly used for collecting information on intersections and arterials. Unlike loop controllers that can generally be only used for detection and movement of vehicles, cameras can provide rich information about the traffic behavior. Vision-based frameworks for multiple-object detection, object tracking, and near-miss detection have been developed to derive this information. However, much of this work currently addresses processing videos offline. In this article, we propose an integrated two-stream convolutional networks architecture that performs real-time detection, tracking, and near-accident detection of road users in traffic video data. The two-stream model consists of a spatial stream network for object detection and a temporal stream network to leverage motion features for multiple-object tracking. We detect near-accidents by incorporating appearance features and motion features from these two networks. Further, we demonstrate that our approaches can be executed in real-time and at a frame rate that is higher than the video frame rate on a variety of videos collected from fisheye and overhead cameras.

A high proportion of crashes happen at or near intersections. To improve intersection safety, trajectory prediction of vehicles has been studied intensively, mostly for automated vehicle (AV) and advanced driver assistance system (ADAS) applications. These approaches with vehicle-mounted sensors still suffer from limited detection range or occluded views. In this work, we propose to perform trajectory prediction on surveillance cameras. As Vehicle-to-Infrastructure (V2I) technology enables low-latency wireless communication, warnings from our prediction algorithm can be sent to vehicles in real-time. Our approach consists of an offline learning phase and an online prediction phase. The offline phase learns common motion patterns from clustering and finds prototype trajectories for each cluster, and updates the prediction model. The online phase predicts the future trajectories for incoming vehicles assuming they follow one of the motion patterns learned from the offline phase. We adopted a long short-term memory encoder-decoder (LSTM-ED) model for the task of trajectory prediction. We also explored the usage of a Curvilinear Coordinate System (CCS), which utilizes the learned prototype and simplifies the trajectory representation. Our model is also able to handle noisy data and variable-length trajectories. Our proposed approach outperforms the baseline Gaussian Process (GP) model, and shows sufficient reliability when evaluated on collected intersection data.

Understanding 3D scenes is a critical prerequisite for autonomous agents. Recently, LiDAR and other sensors have made large amounts of data available in the form of temporal sequences of point cloud frames. In this work, we propose a novel problem -- sequential scene flow estimation (SSFE) -- that aims to predict 3D scene flow for all pairs of point clouds in a given sequence. This is unlike the previously studied problem of scene flow estimation which focuses on two frames. We introduce the SPCM-Net architecture, which solves this problem by computing multi-scale spatiotemporal correlations between neighboring point clouds and then aggregating the correlation across time with an order-invariant recurrent unit. Our experimental evaluation confirms that recurrent processing of point cloud sequences results in significantly better SSFE compared to using only two frames. Additionally, we demonstrate that this approach can be effectively modified for sequential point cloud forecasting (SPF), a related problem that demands forecasting future point cloud frames. Our experimental results are evaluated using a new benchmark for both SSFE and SPF consisting of synthetic and real datasets. Previously, datasets for scene flow estimation have been limited to two frames. We provide non-trivial extensions to these datasets for multi-frame estimation and prediction. Due to the difficulty of obtaining ground truth motion for real-world datasets, we use self-supervised training and evaluation metrics. We believe that this benchmark will be pivotal to future research in this area. All code for benchmark and models will be made accessible., Comment: Accepted for publication in International Journal of Computer Vision, Special Issue on 3D Computer Vision. Code and data: https://github.com/BestSonny/SPCM

Two-dimensional array-based datasets are pervasive in a variety of domains. Current approaches for generative modeling have typically been limited to conventional image datasets and performed in the pixel domain which does not explicitly capture the correlation between pixels. Additionally, these approaches do not extend to scientific and other applications where each element value is continuous and is not limited to a fixed range. In this paper, we propose a novel approach for generating two-dimensional datasets by moving the computations to the space of representation bases and show its usefulness for two different datasets, one from imaging and another from scientific computing. The proposed approach is general and can be applied to any dataset, representation basis, or generative model. We provide a comprehensive performance comparison of various combinations of generative models and representation basis spaces. We also propose a new evaluation metric which captures the deficiency of generating images in pixel space.

We propose SparsePipe, an efficient and asynchronous parallelism approach for handling 3D point clouds with multi-GPU training. SparsePipe is built to support 3D sparse data such as point clouds. It achieves this by adopting generalized convolutions with sparse tensor representation to build expressive high-dimensional convolutional neural networks. Compared to dense solutions, the new models can efficiently process irregular point clouds without densely sliding over the entire space, significantly reducing the memory requirements and allowing higher resolutions of the underlying 3D volumes for better performance. SparsePipe exploits intra-batch parallelism that partitions input data into multiple processors and further improves the training throughput with inter-batch pipelining to overlap communication and computing. Besides, it suitably partitions the model when the GPUs are heterogeneous such that the computing is load-balanced with reduced communication overhead. Using experimental results on an eight-GPU platform, we show that SparsePipe can parallelize effectively and obtain better performance on current point cloud benchmarks for both training and inference, compared to its dense solutions., Accepted in 2020 IEEE 27th International Conference on High Performance Computing, Data, and Analytics (HiPC)

We are developing machine learning algorithms and software to fuse real-time feeds from video cameras and traffic sensor data to generate real-time detection, classification, and space-time trajectories of individual vehicles and pedestrians. This information is then transmitted to a cloud-based system and then synthesized to create a real-time city-wide traffic palette. I will discuss our research on: Smart intersections: Space-time trajectories are used to understand and improve the safety and efficiency of the intersection. Using conflict points of the vehicle-pedestrian trajectories, we identify potential collisions, or “near-misses,” and how they are related to the state of the signal cycle (transition from green to yellow, from yellow to red, etc.) and the presence of other vehicles and pedestrians. • Smart system: We are developing efficient signal re-timing for different corridors by time of day and day of the week to reflect the changes in network demand. We are also developing machine learning techniques for real-time detection of incidents and accidents on arterial networks. • Smart interactions with connected and autonomous vehicles: We have developed signalized intersection control strategies and sensor fusion algorithms for jointly optimizing vehicle trajectories and signal control for a mixture of autonomous vehicles and traditional vehicles at every intersection

The past 20 years has seen a progressive evolution of computer vision algorithms for unsupervised 2D image segmentation. While earlier efforts relied on Markov random fields and efficient optimization (graph cuts, etc.), the next wave of methods beginning in the early part of this century were, in the main, stovepiped. Of these 2D segmentation efforts, one of the most popular and, indeed, one that comes close to being a state of the art method is the ultrametric contour map (UCM). The pipelined methodology consists of (i) computing local, oriented responses, (ii) graph creation, (iii) eigenvector computation (globalization), (iv) integration of local and global information, (v) contour extraction, and (vi) superpixel hierarchy construction. UCM performs well on a range of 2D tasks. Consequently, it is somewhat surprising that no 3D version of UCM exists at the present time. To address that lack, we present a novel 3D supervoxel segmentation method, dubbed 3D UCM, which closely follows its 2D counterpart while adding 3D relevant features. The methodology, driven by supervoxel extraction, combines local and global gradient-based features together to first produce a low-level supervoxel graph. Subsequently, an agglomerative approach is used to group supervoxel structures into a segmentation hierarchy with explicitly imposed containment of lower-level supervoxels in higher-level supervoxels. Comparisons are conducted against state of the art 3D segmentation algorithms. The considered applications are 3D spatial and 2D spatiotemporal segmentation scenarios. For the latter comparisons, we present results of 3D UCM with and without optical flow video pre-processing. As expected, when motion correction beyond a certain range is required, we demonstrate that 3D UCM in conjunction with optical flow is a very useful addition to the pantheon of video segmentation methods.

Freight analysis creates a comprehensive picture of freight movements across the state (and country), by capturing and gathering a variety of sources. Understanding and predicting freight flow patterns is useful in planning and policy decisions at the federal, state, and local levels.Trucks are largely in charge of transporting freight, both in terms of tonnage and revenue. The Federal Highway Administration (FHWA) has a methodology for classifying these trucks into nine categories. Determining the class of the truck is useful in understanding the type of commodity that truck is carrying. This paper details a video based machine learning approach for truck classification. Our approach uses an array of techniques from image processing, deep learning and data mining to develop highly accurate classifiers. Additionally, we provide mechanisms for automatically filtering out trucks from video data.

Background There has been an increasing interest in understanding the usefulness of wrist-based accelerometer data for physical activity (PA) assessment due to the ease of use and higher user compliance than other body placements. PA assessment studies have relied on machine learning methods which take accelerometer data in forms of variables, or feature vectors. Methods In this work, we introduce automated shape feature derivation methods to transform epochs of accelerometer data into feature vectors. As the first step, recurring patterns in the collected data are identified and placed in a codebook. Similarities between epochs of accelerometer data and codebook’s patterns are the basis of feature calculations. In this paper, we demonstrate supervised and unsupervised approaches to learn codebooks. We evaluated these methods and compared them with the standard statistical measures for PA assessment. The experiments were performed on 146 participants who wore an ActiGraph GT3X+ accelerometer on the right wrist and performed 33 activities of daily living. Results Our evaluations show that the shape feature derivation methods were able to perform comparably with the standard wrist model (F1-score: 0.89) for identifying sedentary PAs (F1-scores of 0.86 and 0.85 for supervised and unsupervised methods, respectively). This was also observed for identifying locomotion activities (F1-scores: 0.87, 0.83, and 0.81 for the standard wrist, supervised, unsupervised models, respectively). All the wrist models were able to estimate energy expenditure required for PAs with low error (rMSE: 0.90, 0.93, and 0.90 for the standard wrist, supervised, and unsupervised models, respectively). Conclusion The automated shape feature derivation methods offer insights into the performed activities by providing a summary of repeating patterns in the accelerometer data. Furthermore, they could be used as efficient alternatives (or additions) for manually engineered features, especially important for cases where the latter fail to provide sufficient information to machine learning methods for PA assessment.

Background This paper studies the temporal consistency of health care expenditures in a large state Medicaid program. Predictive machine learning models were used to forecast the expenditures, especially for the high-cost, high-need (HCHN) patients. Results We systematically tests temporal correlation of patient-level health care expenditures in both the short and long terms. The results suggest that medical expenditures are significantly correlated over multiple periods. Our work demonstrates a prevalent and strong temporal correlation and shows promise for predicting future health care expenditures using machine learning. Temporal correlation is stronger in HCHN patients and their expenditures can be better predicted. Including more past periods is beneficial for better predictive performance. Conclusions This study shows that there is significant temporal correlation in health care expenditures. Machine learning models can help to accurately forecast the expenditures. These results could advance the field toward precise preventive care to lower overall health care costs and deliver care more efficiently.

Patients with a large number of health care encounters receive great attention in health care research because their expensive and problematic health care utilization has important implications for the US health care system. The large volume of emergency department (ED) visits and inpatient hospital stays through time provides a unique opportunity to apply data-driven methods for identifying temporal signals associated with these patients. The micro variations in the inter-arrival time of these encounters within this population are not well-studied. Computational approaches for distinguishing various temporal visiting patterns leads to the problem of efficiently clustering asynchronous time series. Thus, we propose a Wasserstein distance based spectral clustering for this problem. Asynchronous time series are first represented as histograms of inter-arrival time and their pairwise similarities are computed under Wasserstein distance. Spectral clustering operates on the similarity matrix as input thereby avoiding the computational bottleneck of Wasserstein barycenters. The effectiveness of this method is demonstrated by synthetic data and application to a large real world health insurance encounters dataset, identifying potential associations between temporal visiting patterns and health factors from a population of frequent ED and inpatient hospital users.

In this paper, we present a novel terrain classification framework for large-scale remote sensing images. A well-performing multi-scale superpixel tessellation based segmentation approach is employed to generate homogeneous and irregularly shaped regions, and a transfer learning technique is sequentially deployed to derive representative deep features by utilizing successful pre-trained convolutional neural network (CNN) models. This design is aimed to overcome the big problem of lacking available ground-truth data and to increase the generalization power of the multi-pixel descriptor. In the subsequent classification step, we train a fast and robust support vector machine (SVM) to assign the pixel-level labels. Its maximum-margin property can be easily combined with a graph Laplacian propagation approach. Moreover, we analyze the advantages of applying a feature selection technique to the deep CNN features which are extracted by transfer learning. In the experiments, we evaluate the whole framework based on different geographical types. Compared with other region-based classification methods, the results show that our framework can obtain state-of-the-art performance w.r.t. both classification accuracy and computational efficiency.

Building an efficient and accurate pixel-level labeling framework for large-scale and high-resolution satellite imagery is an important machine learning application in the remote sensing area. Due to the very limited amount of the ground-truth data, we employ a well-performing superpixel tessellation approach to segment the image into homogeneous regions and then use these irregular-shaped regions as the foundation for the dense labeling work. A deep model based on generative adversarial networks is trained to learn the discriminating features from the image data without requiring any additional labeled information. In the subsequent classification step, we adopt the discriminator of this unsupervised model as a feature extractor and train a fast and robust support vector machine to assign the pixel-level labels. In the experiments, we evaluate our framework in terms of the pixel-level classification accuracy on satellite imagery with different geographical types. The results show that our dense-labeling framework is very competitive compared to the state-of-the-art methods that heavily rely on prior knowledge or other large-scale annotated datasets.

Data association is a key step within the multi-object tracking pipeline that is notoriously challenging due to its combinatorial nature. A popular and general way to formulate data association is as the NP-hard multidimensional assignment problem (MDAP). Over the last few years, data-driven approaches to assignment have become increasingly prevalent as these techniques have started to mature. We focus this survey solely on learning algorithms for the assignment step of multi-object tracking, and we attempt to unify various methods by highlighting their connections to linear assignment as well as to the MDAP. First, we review probabilistic and end-to-end optimization approaches to data association, followed by methods that learn association affinities from data. We then compare the performance of the methods presented in this survey, and conclude by discussing future research directions., Comment: Accepted for publication in ACM Computing Surveys

Most physical activity (PA) assessment studies use wearable accelerometers attached to the hip. However, there is significant and recent interest in understanding the usefulness of wrist based accelerometer data collection due to ease of use and higher compliance. This paper develops machine learning methods for identifying activity types and computing energy expenditures. Our approach converts the raw time series data into intermediate variables or features using standard statistical methods as well as bag-of-words (BoW) approach. We tested this approach to assess type of physical activities as well as estimate the required corresponding energy expenditure. The method was evaluated on 17 participants. Each of the participants wore an Actigraph GT3X+ accelerometer on the right wrist and performed 33 activities of daily living. Energy expenditure was measured in parallel by a portable indirect calorimetry system. Our results show that the BoW approach resulted in a more accurate model for PA identification (F1-score = 0.88 and 0.91 for sedentary and locomotion detection, respectively), compared with standard statistical summaries. The BoW approach preserved additional details about the accelerometer data, which resulted in distinguishing different activities that belonged to the same higher-level category (e.g., distinguishing leisure walk from stair ascent where both PAs belong to the locomotion class) and consequently yielding an accurate energy expenditure estimation model for PAs (rMSE = 0.93 and R2 = 0.69).

While computer vision has made noticeable advances in the state of the art for 2D image segmentation, the same cannot be said for 3D volumetric datasets. In this work, we present a scalable approach to volumetric segmentation. The methodology, driven by supervoxel extraction, combines local and global gradient-based features together to first produce a low level supervoxel graph. Subsequently, an agglomerative approach is used to group supervoxel structures into a segmentation hierarchy with explicitly imposed containment of lower level supervoxels in higher level supervoxels. Comparisons are conducted against state of the art 3D segmentation algorithms. The considered applications are 3D spatial and 2D spatiotemporal segmentation scenarios.

We applied and developed several machine learning techniques, including linear model, tree-based model, and deep neural networks to forecast expenditures for high utilizers in a very large public health program. The results show promise for predicting health care expenditures for these high utilizers. To improve interpretability, we quantified the contributions of influential input variables to the prediction score. These results help to advance the field toward targeted preventive care to lower overall health care costs.

With Moore's law supplying billions of transistors on-chip, embedded systems are undergoing a transition from single-core to multi-core to exploit this high transistor density for high performance. However, the optimal layout of these multiple cores along with the memory subsystem (caches and main memory) to satisfy power, area, and stringent real-time constraints is a challenging design endeavor. The short time-to-market constraint of embedded systems exacerbates this design challenge and necessitates the architectural modeling of embedded systems to reduce the time-to-market by expediting target applications to device/architecture mapping. In this paper, we present a queueing theoretic approach for modeling multi-core embedded systems that provides a quick and inexpensive performance evaluation both in terms of time and resources as compared to the development of multi-core simulators and running benchmarks on these simulators. We verify our queueing theoretic modeling approach by running SPLASH-2 benchmarks on the SuperESCalar simulator (SESC). Results reveal that our queueing theoretic model qualitatively evaluates multi-core architectures accurately with an average difference of 5.6% as compared to the architectures' evaluations from the SESC simulator. Our modeling approach can be used for performance per watt and performance per unit area characterizations of multi-core embedded architectures, with varying number of processor cores and cache configurations, to provide a comparative analysis. Queueing theory-based modeling technique for evaluating multi-core architectures.Enables quick and inexpensive architectural evaluation.Architectural evaluation for workloads with any computing requirements.Can be used for performance per watt & performance per unit area characterizations.Provides insights about shared last-level caches (LLCs) orchestration.

The reconstruction of nxn-pixel Synthetic Aperture Radar imagery using Back Projection algorithm incurs O(n2·m) cost, where m is the number of pulses. This paper presents dynamic data driven multiresolution algorithms to speed up SAR backprojection on multiple GPUs. A critical part of this spatially variant reconstruction process is load balancing, which circumvents asymmetric work assignment. Our algorithms achieve 15 TFLOPS using 128 GPUs.

Considering the graph of a feature function as an embedded surface in three dimensions is a standard device in computer vision. When multiple feature functions (eg. multiple images) are available, the natural extension of the above concept is to a higher-dimensional embedded surface. This has received surprisingly little attention. In this paper, we advocate for this view by showing the utility of surface area for estimating spatial transformations between images for the purposes of registration. In contrast to the entropy of the stack of images (when represented as continuous and differentiable functions) being zero, the surface area turns out to be useful measure for image registration. We show that during the progression of standard registration algorithms like Congealing, the area of the stack of images being registered decreases. Our own algorithm for affine registration based on image stack surface area (ISSA) minimization has several advantages compared with Congealing and Mutual Information (MI) registration. Finally, we highlight the generality of our framework by also showcasing experiments on affine point-set registration.

Motivation: Phylogenetics, or reconstructing the evolutionary relationships of organisms, is critical for understanding evolution. A large number of heuristic algorithms for phylogenetics have been developed, some of which enable estimates of trees with tens of thousands of taxa. Such trees may not be robust, as small changes in the input data can cause major differences in the optimal topology. Tools that can assess the quality and stability of phylogenetic tree estimates and identify the most reliable parts of the tree are needed. Results: We define measures that assess the stability of trees, subtrees and individual taxa with respect to changes in the input sequences. Our measures consider changes at the finest granularity in the input data (i.e. individual nucleotides). We demonstrate the effectiveness of our measures on large published datasets. Our measures are computationally feasible for phylogenetic datasets consisting of tens of thousands of taxa. Availability: This software is available at http://bioinformatics.cise.ufl.edu/phylostab Contact: sheikh@cise.ufl.edu

We explore novel algorithms for DVS (Dynamic Voltage Scaling) based energy minimization of DAG (Directed Acyclic Graph) based applications on parallel and distributed machines in dynamic environments. Static DVS algorithms for DAG execution use the estimated execution time. The estimated time in practice is overestimated or underestimated. Therefore, many tasks may be completed earlier or later than expected during the actual execution. For overestimation, the extra available slack can be added to future tasks so that energy requirements can be reduced. For underestimation, the increased time may cause the application to miss the deadline. Slack can be reduced for future tasks to reduce the possibility of not missing the deadline. In this paper, we present novel dynamic scheduling algorithms for reallocating the slack for future tasks to reduce energy and/or satisfy deadline constraints. Experimental results show that our algorithms are comparable to static algorithms applied at runtime in terms of energy minimization and deadline satisfaction, but require considerably smaller computational overhead.

Energy consumption is a critical issue in parallel and distributed embedded systems. We present a novel algorithm for energy efficient scheduling of Directed Acyclic Graph (DAG) based applications on Dynamic Voltage Scaling (DVS) enabled systems. Experimental results show that our algorithm provides near optimal solutions for energy minimization with considerably smaller computational time and memory requirements compared to an existing algorithm that provides near optimal solutions.

Sensor networks are finding significant applications in large scale distributed systems. One of the basic operations in sensor networks is in-network aggregation. Among the various approaches to in-network aggregation, such as gossip and tree, including the hash-based techniques, the tree-based approaches have better performance and energy-saving characteristics. However, sensor networks are highly prone to failures. Numerous techniques suggested in the literature to counteract the effect of failures have not been carefully analyzed. In this paper, we focus on the performance of these tree-based aggregation techniques in the presence of failures. First, we identify a fault model that captures the important failure traits of the system. Then, we analyze the correctness of simple tree aggregation with our fault model. We then use the same fault model to analyze the techniques that utilize redundant trees to improve the variance. The impact of techniques for maintaining the correctness under faults, such as rebuilding or locally fixing the tree, is then studied under the same fault model. We also do the cost-benefit analysis of using the hash-based schemes which are based on FM sketches. We conclude that these fault tolerance techniques for tree aggregation do not necessarily result in substantial improvement in fault tolerance.

Fault tolerance and scalability are important considerations in the design of sensor network applications. Data aggregation is an essential operation in sensor networks. Multiple techniques have been proposed recently to tackle the issues of scalability and fault tolerance of aggregation in sensor networks. In this article, we analyze the impact of using a few of the more reliable, though expensive, nodes-such as the Intel XScale-called microservers, in addition to the standard motes, on the fault tolerance and scalability of the aggregation algorithms in sensor networks. In particular, we propose a simple model that captures the essence of tree aggregation in such heterogeneous sensor networks. We validate this theoretical model with simulation results. We also study the effective impact on the sustainable probability of failure, and perform cost-benefit analysis. We also show how hybrid aggregation can be utilized instead of tree, to improve the performance of aggregation in heterogeneous sensor networks. We show that our work can be applied for effectively optimizing the use of expensive hardware while designing fault-tolerant, distributed sensor networks.

Recurrent chromosomal alterations provide cytological and molecular positions for the diagnosis and prognosis of cancer. Comparative genomic hybridization (CGH) has been useful in understanding these alterations in cancerous cells. CGH datasets consist of samples that are represented by large dimensional arrays of intervals. Each sample consists of long runs of intervals with losses and gains. In this article, we develop novel SVM-based methods for classification and feature selection of CGH data. For classification, we developed a novel similarity kernel that is shown to be more effective than the standard linear kernel used in SVM. For feature selection, we propose a novel method based on the new kernel that iteratively selects features that provides the maximum benefit for classification. We compared our methods against the best wrapper-based and filter-based approaches that have been used for feature selection of large dimensional biological data. Our results on datasets generated from the Progenetix database, suggests that our methods are considerably superior to existing methods. Availability: All software developed in this article can be downloaded from http://plaza.ufl.edu/junliu/feature.tar.gz Contact: juliu@cise.ufl.edu

When anomaly detection software is used as a data analysis tool, finding the hardest-to-detect anomalies is not the most critical task. Rather, it is often more important to make sure that those anomalies that are reported to the user are in fact interesting. If too many unremarkable data points are returned to the user labeled as candidate anomalies, the software can soon fall into disuse. One way to ensure that returned anomalies are useful is to make use of domain knowledge provided by the user. Often, the data in question includes a set of environmental attributes whose values a user would never consider to be directly indicative of an anomaly. However, such attributes cannot be ignored because they have a direct effect on the expected distribution of the result attributes whose values can indicate an anomalous observation. This paper describes a general purpose method called conditional anomaly detection for taking such differences among attributes into account, and proposes three different expectation-maximization algorithms for learning the model that is used in conditional anomaly detection. Experiments with more than 13 different data sets compare our algorithms with several other more standard methods for outlier or anomaly detection

Urban neighborhood classification using very high resolution (VHR) remote sensing imagery is a challenging and {\em emerging} application. A semi-supervised learning approach for identifying neighborhoods is presented which employs superpixel tessellation representations of VHR imagery. The image representation utilizes homogeneous and irregularly shaped regions termed superpixels and derives novel features based on intensity histograms, geometry, corner and superpixel density and scale of tessellation. The semi-supervised learning approach uses a support vector machine (SVM) to obtain a preliminary classification which is then subsequently refined using graph Laplacian propagation. Several intermediate stages in the pipeline are presented to showcase the important features of this approach. We evaluated this approach on four different geographic settings with varying neighborhood types and compared it with the recent Gaussian Multiple Learning algorithm. This evaluation shows several advantages, including model building, accuracy, and efficiency which makes it a great choice for deployment in large scale applications like global human settlement mapping and population distribution (e.g., LandScan), and change detection.

Classification of very high resolution (VHR) remote sensing imagery is a rapidly emerging discipline but faces several challenges owing to the huge scale of the pixel data involved, indiscernibility in the traditionally used features to represent various regions, and the lack of available ground truth data. This paper provides a framework which elegantly overcomes these hurdles by providing a novel semi-supervised learning approach which employs multiscale superpixel tessellation representations of VHR imagery. Superpixels are homogeneous and irregularly shaped regions which form the backbone of our approach and are used to derive novel features by learning a decision tree. Our semi-supervised learning approach works on a superpixel graph and seamlessly combines the large margin capability of a support vector machine (SVM) with a graph-based Laplacian label propagation approach to obtain a novel objective function. Further we also provide a self-contained and easily parallelisable linear iterative optimisation approach based on the principle of majorisation-minimisation. We evaluate this approach on four different geographic settings with varying neighbourhood types and draw comparisons with the popular and widely used Gaussian multiple instance learning algorithm. Our results showcase several advantages in accuracy and efficiency, which coupled with the ease of model building and inherently parallelisable optimisation make our framework a great choice for deployment in large scale applications like global human settlement mapping and population distribution, and change detection.

We present a novel framework for semisupervised labeling of regions in remote sensing image datasets. Our approach works by decomposing the image into irregular patches or superpixels and derives novel features based on intensity histograms, geometry, corner density, and scale of tessellation. Our classification pipeline uses either k-nearest neighbors or SVM to obtain a preliminary classification which is then refined using Laplacian propagation algorithm. Our approach is easily parallelizable and fast despite the high volume of data involved. Results are presented which showcase the accuracy as well as different stages of our pipeline.

This paper describes deterministic communication-efficient algorithms for performing dynamic permutations on a coarse-grained parallel machine. Our analysis shows that the general permutation operation can be completed inC?n/p(+ lower order terms) time and is optimal and scalable providedn?p3+p2?/? (nis the size of the permutation or the number of elements distributed across thepprocessors, ? is the start-up overhead, and 1/? is the data transfer rate).Cis a small constant typically between 2 and 3 for write permutations, slightly higher for read permutations. Modifications to exploit locality of access are presented. Special classes of permutations that are optimal for smaller sizes are also described. A companion paper 22] deals with the problem of random data accesses with hot spots.

This paper describes deterministic communication-efficient algorithms for performing random data accesses with hot spots on a coarse-grained parallel machine. The general random access read-write operations with hot spots can be completed inC?n/p(+ lower order terms) time and is optimal and scalable providedn?p3+p2?/? (nis the number of elements distributed acrosspprocessors, ? is the start-up overhead and 1/? is the data transfer rate).Cis a small constant between 3 and 4 for the random access write operation, slightly higher for the random access read operation. Monotonic random access reads/writes can be completed with smaller constants and are optimal for smallernas well. A companion paper 26] deals with the problem of performing dynamic permutations.

In this paper we discuss the runtime support required for the parallelization of unstructured data-parallel applications on nonuniform and adaptive environments. We describe several optimization techniques for fast remapping of data and for reducing the amount of communications between machines when the computational load on the machines changes during the execution of data-parallel applications. The approach presented is reasonably general and is applicable to a wide variety of regular as well as irregular applications. We present performance results for the solution of an unstructured mesh on a cluster of workstations.

In this paper we present fast parallel algorithms for remapping a class of irregular and adaptive problems on coarse-grained distributed memory machines. We show that the remapping of these applications, using simple index-based mapping algorithm, can be reduced to sorting a nearly sorted list of integers or merging an unsorted list of integers with a sorted list of integers. By using the algorithms we have developed, the remapping of these problems can be achieved at a fraction of the cost of mapping from scratch.

The imaging of ground objects by circular synthetic aperture radar (CSAR) is a well-known technique that can benefit from the use of multiple receivers in a multistatic configuration. Although static receivers have been employed to determine the location of one or more airborne objects, the use of multiple dynamically positioned airborne receivers with an airborne transmitter represents a novel application of SAR technology for the detection of ground objects. This paper presents theory, algorithms, and experimental results for dynamic multistatic synthetic aperture radar (DMSAR) for airborne sensing of ground-based objects. We emphasize the proper placement of receivers in relationship to the transmitter and object area, in order to achieve a specific quality of image reconstruction. Additionally, we discuss performance issues pertaining to the computation of DMSAR on high-performance platforms such as clusters of graphics processing units or hybrid processors. Example results and analysis are taken from synthetic or public-domain data.

PACK/UNPACK are Fortran 9O/HPF array construction functions which derive new arrays from existing arrays. We present algorithms for performing these operations on coarse-grained parallel machines. Our algorithms are relatively architecture independent and can be applied to arrays of arbitrary dimensions with arbitrary distribution along every dimension. Experimental results are presented on the CM-5.

Two-dimensional arrays are useful in a large variety of scientific and engineering applications. Parallelization of these applications requires the decomposition of array elements among different machines. Several data-decomposition techniques have been studied in the literature for machines with uniform computational power. In this paper, we develop new methods for decomposing arrays into a cluster of machines with nonuniform computational power. Simulation results show that our methods provide superior decomposition over naive schemes.

We study two classes of sigmoids: the simple sigmoids, defined to be odd, asymptotically bounded, completely monotone functions in one variable, and the hyperbolic sigmoids, a proper subset of simple sigmoids and a natural generalization of the hyperbolic tangent. We obtain a complete characterization for the inverses of hyperbolic sigmoids using Euler's incomplete beta functions, and describe composition rules that illustrate how such functions may be synthesized from others. These results are applied to two problems. First we show that with respect to simple sigmoids the continuous Cohen-Grossberg-Hopfield model can be reduced to the (associated) Legendre differential equations. Second, we show that the effect of using simple sigmoids as node transfer functions in a one-hidden layer feedforward network with one summing output may be interpreted as representing the output function as a Fourier series sine transform evaluated at the hidden layer node inputs, thus extending and complementing earlier results in this area. Copyright © 1996 Elsevier Science Ltd

In this paper, we present several algorithms for performing all-to-many personalized communication on distributed memory parallel machines. We assume that each processor sends a different message (of potentially different size) to a subset of all the processors involved in the collective communication. The algorithms are based on decomposing the communication matrix into a set of partial permutations. We study the effectiveness of our algorithms from both the view of static scheduling and runtime scheduling.

The POWER (PrObabilistic Weighted Ensemble of Roles) model[1] is a Bayesian mixture model where a single data point is generated by multiple components with varying degrees of influence. This is unlike standard mixture models where each data point is generated by a single mixture component. The POWER model allows for capturing various hidden and complexly overlapping components and roles of entities that contribute to generating the data points. However, the POWER model suffers from a very slow learning time. The highest complexity time of the parameters’ learning steps is \(O\left(n \cdot k \cdot d^{2}\right)\), with n being the number of data points, k the number of components of the model and d the number of data attributes.

Region growing is a general technique for image segmentation, where image characteristics are used to group adjacent pixels together to form regions. This paper presents a parallel algorithm for solving the region growing problem based on the split-and-merge approach, and uses it to test and compare various parallel architectures and programming models. The implementations were done on the Connection Machine, models CM-2 and CM-5, in the data parallel and message passing programming models. Randomization was introduced in breaking ties during merging to increase the degree of parallelism, and only one- and two-dimensional arrays of data were used in the implementations.

Distributed memory multiprocessors are increasingly being used to provide high performance for advanced calculations with scientific applications. Distributed memory machines offer significant advantages over their shared memory counterparts in terms of cost and scalability, though it is widely accepted that they are difficult to program given the current status of software technology. Currently, distributed memory machines are programmed using a node language and a message passing library. This process is tedious and error prone because the user must perform the task of data distribution and communication for non-local data access. This thesis describes an advanced compiler that can generate efficient parallel programs when the source programming language naturally represents an application's parallelism. Fortran 90D/HPF described in this thesis is such a language. Using Fortran 90D/HPF, parallelism is represented with parallel constructs, such as array operations, where statements, forall statements, and intrinsic functions. The language provides directives for data distribution. Fortran 90D/HPF gives the programmer powerful tools to express a problem with natural data parallelism. To validate this hypothesis, a prototype of Fortran 90D/HPF was implemented. The compiler is organized around several major units: language parsing, partitioning data and computation, detecting communication and generating code. The compiler recognizes the presence of communication patterns in the computations in order to generate appropriate communication calls. Specifically, this involves a number of tests on the relationships among subscripts of various arrays in a statement. The compiler includes a specially designed algorithm to detect communications and to generate appropriate collective communication calls to execute array assignments and forall statements. The Fortran 90D/HPF compiler performs several types of communication and computation optimizations to improve the performance of the generated code. Empirical measurements show that the performance of the output of the Fortran 90D/HPF compiler is comparable to that of corresponding hand-written codes on several systems. We hope that this thesis assists in the widespread adoption of parallel computing technology and leads to a more attractive and powerful software development environment to support application parallelism that many users need.

Microarray experiments often measure expressions of genes taken from sample tissues in the presence of external perturbations such as medication, radiation, or disease. Typically in such experiments, gene expressions are measured before and after the application of external perturbation. In this paper, we focus on an important class of such microarray experiments that inherently have two groups of tissue samples. The external perturbation can change the expressions of some genes directly or indirectly through gene interaction network. When such different groups exist, the expressions of genes after the perturbation can be different between the two groups. It is not only important to identify the genes that respond differently across the two groups, but also to mine the reason behind this differential response. In this paper, we aim to identify the cause of this differential behavior of genes, whether because of the perturbation or due to interactions with other genes in two group perturbation experiments. We propose a new probabilistic Bayesian method with Markov Random Field to find such genes. Our method incorporates information about relationship from gene networks as prior information. Experimental results on synthetic and real datasets demonstrate the superiority of our method compared to existing techniques.