Engineering Papers⌕ Search

Engineering topics

Vatsavai, Raju

Publications and source records attributed to Vatsavai, Raju.

12 records
All sourcesNASA NTRSDOE OSTI
DOE OSTI2016

Detecting Extreme Events in Gridded Climate Data

Detecting and tracking extreme events in gridded climatological data is a challenging problem on several fronts: algorithms, scalability, and I/O. Successful detection of these events will give climate scientists an alternate view of the behavior of different climatological variables, leading to enhanced scientific understanding of the impacts of events such as heat and cold waves, and on a larger scale, the El Nin o Southern Oscillation. Recent advances in computing power and research in data sciences enabled us to look at this problem with a different perspective from what was previously possible. In this paper we present our computationally efficient algorithms for anomalous cluster detection on climate change big data. We provide results on detection and tracking of surface temperature and geopotential height anomalies, a trend analysis, and a study of relationships between the variables. We also identify the limitations of our approaches, future directions for research and alternate approaches.

climate extremes temperature anomalies reanalysis↗
DOE OSTI2016

Evaluation of Graph Pattern Matching Workloads in Graph Analysis Systems

Graph analysis has emerged as a powerful method for data scientists to represent, integrate, query, and explore heterogeneous data sources. As a result, graph data management and mining became a popular area of research, and led to the development of plethora of systems in recent years. Unfortunately, the number of emerging graph analysis systems and the wide range of applications, coupled with a lack of apples-to-apples comparisons, make it difficult to understand the trade-offs between different systems and the graph operations for which they are designed. A fair comparison of these systems is a challenging task for the following reasons: multiple data models, non-standardized serialization formats, various query interfaces to users, and diverse environments they operate in. To address these key challenges, in this paper we present a new benchmark suite by extending the Lehigh University Benchmark (LUBM) to cover the most common capabilities of various graph analysis systems. We provide the design process of the benchmark, which generalizes the workflow for data scientists to conduct the desired graph analysis on different graph analysis systems. Equipped with this extended benchmark suite, we present performance comparison for nine subgraph pattern retrieval operations over six graph analysis systems, namely NetworkX, Neo4j, Jena, Titan, GraphX, and uRiKA. Through the proposed benchmark suite, this study reveals both quantitative and qualitative findings in (1) implications in loading data into each system; (2) challenges in describing graph patterns for each query interface; and (3) different sensitivity of each system to query selectivity. We envision that this study will pave the road for: (i) data scientists to select the suitable graph analysis systems, and (ii) data management system designers to advance graph analysis systems.

Hong, Seokyong↗
DOE OSTI2015

Machine Learning and Understanding for Intelligent Extreme Scale Scientific Computing and Discovery. DOE Workshop Report, January 7–9, 2015, Rockville, MD

Large-scale parallel simulations and data analysis drive scientific discovery across many disciplines. To drive larger and more detailed simulations, and deal with larger data volumes, exascale machines capable of 1018 operations per second are expected within the next five to ten years. However, the complexity of developing and adapting modern simulation codes for these architectures is increasing rapidly. Workloads on exascale machines will be billion-way parallel. Exploiting the full capability of the machine will require carefully assigning application tasks to cores, accelerators, deep memories, and other heterogeneous compute resources, while simultaneously optimizing for time to solution, data movement, power, and resilience. Optimizations that improve performance on one machine may slow down another. Worse, applications themselves are dauntingly complex. Production simulations comprise millions of lines of code and use sophisticated, adaptive algorithms whose performance is input-dependent. Complex workflows can couple multi-physics simulation with data preprocessing and post-processing modules.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2014

Analytics For Distracted Driver Behavior Modeling in Dilemma Zone

In this paper, we present the results obtained and insights gained through the analysis of TRB contest data. We used exploratory analysis, regression, and clustering models for gaining insights into the driver behavior in a dilemma zone while driving under distraction. While simple exploratory analysis showed the distinguishing driver behavior patterns among different popu- lation groups in the dilemma zone, regression analysis showed statically signification relationships between groups of variables. In addition to analyzing the contest data, we have also looked into the possible impact of distracted driving on the fuel economy.

Distracted driver behavior↗
DOE OSTI2014

Advances In very high resolution satellite imagery analysis for Monitoring human settlements

The high rate of urbanization, political conflicts and ensuing internal displacement of population, and increased poverty in the 20th century has resulted in rapid increase of informal settlements. These unplanned, unauthorized, and/or unstructured homes, known as informal settlements, shantytowns, barrios, or slums, pose several challenges to the nations, as these settlements are often located in most hazardous regions and lack basic services. Though several World Bank and United Nations sponsored studies stress the importance of poverty maps in designing better policies and interventions, mapping slums of the world is a daunting and challenging task. In this paper, we summarize our ongoing research on settlement mapping through the utilization of Very high resolution (VHR) remote sensing imagery. Most existing approaches used to classify VHR images are single instance (or pixel-based) learning algorithms, which are inadequate for analyzing VHR imagery, as single pixels do not contain sufficient contextual information (see Figure 1). However, much needed spatial contextual information can be captured via feature extraction and/or through newer machine learning algorithms in order to extract complex spatial patterns that distinguish informal settlements from formal ones. In recent years, we made significant progress in advancing the state of art in both directions. This paper summarizes these results.

Vatsavai, Raju↗
DOE OSTI2013

Complex Settlement Pattern Extraction With Multi-instance Learning

Per-pixel (or single instance) based classification schemes which have proven to be very useful in thematic classification have shown to be inadequate when it comes to analyzing very high resolution remote sensing imagery. The main problem being that the pixel size (less than a meter) is too small as compared to the typical object size (100s of meters) and contains too little contextual information to accurately distinguish complex settlement types. One way to alleviate this problem is to consider a bigger window or patch/segment consisting a group of adjacent pixels which offers better spatial context than a single pixel. Unfortunately, this makes per-pixel based classification schemes ineffective. In this work, we look at a new class of machine learning approaches, called multi-instance learning, where instead of assigning class labels to individual instances (pixels), a label is assigned to the bag (all pixels in a window or segment). We applied this multi-instance learning approach for identifying two important urban patterns, namely formal and informal settlements. Experimental evaluation shows the better performance of multi-instance learning over several well-known single-instance classification schemes.

multi-instance learning↗
DOE OSTI2013

Gaussian Multiple Instance Learning Approach for Mapping the Slums of the World Using Very High Resolution Imagery

In this paper, we present a computationally efficient algo- rithm based on multiple instance learning for mapping infor- mal settlements (slums) using very high-resolution remote sensing imagery. From remote sensing perspective, infor- mal settlements share unique spatial characteristics that dis- tinguish them from other urban structures like industrial, commercial, and formal residential settlements. However, regular pattern recognition and machine learning methods, which are predominantly single-instance or per-pixel classi- fiers, often fail to accurately map the informal settlements as they do not capture the complex spatial patterns. To overcome these limitations we employed a multiple instance based machine learning approach, where groups of contigu- ous pixels (image patches) are modeled as generated by a Gaussian distribution. We have conducted several experi- ments on very high-resolution satellite imagery, represent- ing four unique geographic regions across the world. Our method showed consistent improvement in accurately iden- tifying informal settlements.

Vatsavai, Raju↗
DOE OSTI2013

Evaluating fusion techniques for multi-sensor satellite image data

Satellite image data fusion is a topic of interest in many areas including environmental monitoring, emergency response, and defense. Typically any single satellite sensor cannot provide all of the benefits offered by a combination of different sensors (e.g., high-spatial but low spectral resolution vs. low-spatial but high spectral, optical vs. SAR). Given the respective strengths and weaknesses of the different types of image data, it is beneficial to fuse many types of image data to extract as much information as possible from the data. Our work focuses on the fusion of multi-sensor image data into a unified representation that incorporates the potential strengths of a sensor in order to minimize classification error. Of particular interest is the fusion of optical and synthetic aperture radar (SAR) images into a single, multispectral image of the best possible spatial resolution. We explore various methods to optimally fuse these images and evaluate the quality of the image fusion by using K-means clustering to categorize regions in the fused images and comparing the accuracies of the resulting categorization maps.

Martin, Benjamin W↗
DOE OSTI2013

Image Change Detection via Ensemble Learning

The concept of geographic change detection is relevant in many areas. Changes in geography can reveal much information about a particular location. For example, analysis of changes in geography can identify regions of population growth, change in land use, and potential environmental disturbance. A common way to perform change detection is to use a simple method such as differencing to detect regions of change. Though these techniques are simple, often the application of these techniques is very limited. Recently, use of machine learning methods such as neural networks for change detection has been explored with great success. In this work, we explore the use of ensemble learning methodologies for detecting changes in bitemporal synthetic aperture radar (SAR) images. Ensemble learning uses a collection of weak machine learning classifiers to create a stronger classifier which has higher accuracy than the individual classifiers in the ensemble. The strength of the ensemble lies in the fact that the individual classifiers in the ensemble create a mixture of experts in which the final classification made by the ensemble classifier is calculated from the outputs of the individual classifiers. Our methodology leverages this aspect of ensemble learning by training collections of weak decision tree based classifiers to identify regions of change in SAR images collected of a region in the Staten Island, New York area during Hurricane Sandy. Preliminary studies show that the ensemble method has approximately 11.5% higher change detection accuracy than an individual classifier.

Martin, Benjamin W↗