UHCS PhD Showcase 2017

Program - 2014

Select a topic below for more detail…


Student Oral Presentations - PGH 232

Schedule

Time Description
9:00 am Welcome & UHCS Points of Pride - Dr. Jaspal Subhlok
9:05 am Overview - Dr. Shishir Shah
9:15 am

Student Oral Presentations

  • Bassam Almogahed
  • Wei Ding
  • Deepak Eachempati
  • Waleed Faris
  • Apurva Gala
  • Binh Le
  • Yen Le
  • Meenakshi Sharma
  • Behnaz Sanati
11:45 am Lunch
12:30 pm Keynote Speaker - Dr. Roberta Ness, "So You Think You Can Innovate?"
1:45 pm

Student Oral Presentations

  • Munara Tolubaeva
  • Xu Yan
  • Paul Hernandez
  • Lijuan Zhao
  • Ming Chih Shih
  • Bassam Almogahed - “Toward Resolving the Data Imbalanced Issue in Supervised Learning Problems”

    We present a framework to address the imbalanced data problem using semi-supervised learning. Specifically, from a supervised problem, we create a semi-supervised problem and then use a semi-supervised learning method to identify the most relevant instances to establish a well-defined training set for the imbalanced dataset.

  • Wei Ding - “Detecting Stepping-Stones Under the Influence of Packet Jittering”

    This type of stepping-stone attack can be detected by applying timing-based correlation algorithms on the connections in and out of a host. However, hackers can add chaff packets or jitter the original packets to decrease the detection rate of these correlation algorithms. This paper proposes a novel method to detect intrusions under the influence of packet jittering.

  • Deepak Eachempati - “Implementation and Optimization Techniques for Fortran 2008”

    Fortran has long been a preferred programming language for use in High Performance Computing. In the most recent version of the standard, Fortran 2008, parallel processing features based on the Partitioned Global Address Space programming model were incorporated into the language specification. The speaker will describe his efforts in implementing these features in an open-source compiler. He will cover techniques developed and implemented to deal with specific technical challenges in implementing PGAS languages and present performance results.

  • Waleed Faris - “Communicating with ALPS:  The Building of a Natural Language Processor”

    Humans use languages as a way of exchanging ideas. The ability to learn a language in order to read and write indicates that there is an underlying link between the language and the knowledge that is being represented. These links are common across multiple languages. The objective of this presentation is to provide ALPS with the knowledge it needs to bridge this gap. The problem is divided into three tasks: learning a grammar, understanding a sentence, and generating a sentence to express a thought.

  • Apurva Gala - “Person Re-identification for Distributed Wide Area Surveillance”

    Person re-identification (Re-ID) is the process of matching images/videos of people taken from different cameras. If the images/videos are taken on different days, traditional assumptions, of same clothing across different cameras does not hold. In this study, we explore gait features for long period person re-ID and understand the limits of gait for Re-ID.

  • Binh Le - “Marker Optimization for Facial Motion Acquisition and Deformation”

    A long-standing problem in marker-based facial motion capture is what are the optimal facial mocap marker layouts. Despite its wide range of potential applications, this problem has not yet been systematically explored to date. In this presentation, I will describe an approach to compute optimized marker layouts for facial motion acquisition as optimization of characteristic control points from a set of high-resolution, ground-truth facial mesh sequences.

  • Yen Le - “PDM-ENLOR:  Learning Ensemble of Local PDM-based Regressions”

    Active Shape Models (ASMs), suffer from their inability to represent a large range of variations of a complex shape and to account for the large errors in detection of model points. We propose a novel method that overcomes these limitations by locating each shape model point individually using an ensemble of local regression models and appearance cues from selected model points. We demonstrate the advantages of our method on the problem of segmenting gene expression images of mouse brain.

  • Meenakshi Sharma - “Detecting Altered Methylation States Using High Throughput DNA Sequencing”

    Altered DNA methylation patterns have been associated with genomic instability and are a hallmark of cancer. High-throughput sequencing (HTS) provides a useful and cost-effective method to identify “methylation signatures” in comparative and genome-wide studies. Here we present a novel computational approach to detect Differentially Methylated Regions in cancer cell lines by minimizing the bias introduced by repeated DNA sequences present in human genome.

  • Munara Tolubaeva - “Compile Time Modeling of Off-Chip Memory Bandwidth for Parallel Loops”

    We present a statistical model to predict the off-chip memory bandwidth required by a parallel loop during its execution. It is a compile-time modeling technique that derives the correlations between memory bandwidth requirement and data access patterns of multithreaded applications.

  • Behnaz Sanati - “An Online Partitioned Scheduling of Real-Time Tasks with Reward Constraints”

    Maximizing the benefit gained by soft real-time tasks in many applications is highly needed to provide an acceptable QoS (Quality of Service). In this research, an efficient preemptive benefit-aware technique is proposed. This method considers online choices of two approximation algorithms including greedy and load balancing to partition tasks among homogeneous multi-processors at the time of release without using any statistics. It maximizes total gained benefit while reducing the total idle time of the processors and the makespan of the tasks.

  • Xu Yan - “Modeling Local Behavior for Multi-Person Tracking”

    Human interaction dynamics are known to play an important role in the development of robust pedestrian trackers that are needed for a variety of applications in video surveillance. Recent approaches have begun to leverage interaction, especially by modeling the repulsion forces among pedestrians to improve motion predictions. However, human interaction is more complex and is influenced by multiple social effects. We propose a novel human tracking method by modeling complex social interactions.

  • Paul Hernandez - “One Class Classification for Segmentation of Neurons”

    We propose a novel one-class classification method to segment neurons. First, a new criterion to select a training set consisting of background voxels is proposed. Then, a discriminant function is learned from the training set that allows determining how similar an unlabeled voxel is to the voxels in the background class. Finally, foreground voxels are assigned as those unlabeled voxels that are not classified as background.

  • Lijuan Zhao - “Automated Detection of Breast Contour in 3D Images of the Female Torso”

    The ability to quantify morphological features of the breast facilitates pre-operative planning and post-operative outcome assessment in breast reconstruction. Breast contour is an important attribute for quantitative assessment of breast aesthetics. Based on the detected breast contour, relevant morphological measures such as breast size, shape, symmetry, volume and ptosis can be determined. In this study we present an approach for the automatic contour detection of the lower breast in three-dimensional (3D) images.

  • Ming Chih Shih - “Automatic B Cell Lymphoma Detection Using Flow Cytometry Data”

    Flow cytometry (FC) is popular used to detect B Cell lymphoma. The current advanced technology rapid produce a huge data set but current FC analysis is mostly based on human manual gating, tedious and time-consuming process. We proposed a solution to detect the disease, track the treatment and monitor the relapse.

  • Otto Dobretsberger - “Computational Challenges in the Analysis and Manipulation of Short Genomic Sequences”

    Genomic sequences are composed of only a 4- or 5- letter alphabet (A, T, C, G, & N), which represent the 4 nucleic acids (adenine, tyrosine, cytosine, guanine, & unknown, respectively). This information can be used to develop algorithms specifically tailored to address Next Generation Sequencing (NGS) data. We have developed algorithms to efficiently compress NGS data to a fraction of what alternative compression techniques such as ZIP or GZIP are capable of.

  • Roberto Valerio Molina - “A Data Complexity Approach to Kernel Selection for Support Vector Machines”

    A data complexity approach to kernel selection based on the behavior of Polynomial and Gaussian kernels.

Student Poster Presentations - A.D. Bruce Religion Center, 2nd Floor Atrium Lounge

Schedule

Time Description
3:15 - 5:15 pm Student Poster Presentations
  • Fatih Akdag - “Creating Polygon Model for Spatial Clusters”

    Polygon models are essential in spatial data mining applications in order to analyze relationships and change between clusters. We developed a software framework that takes a spatial cluster as an input and generates polygon model for the cluster as an output. The polygons are generated using Characteristics shapes. We propose polygon fitness functions to automatically select proper input parameter and define a novel polygon emptiness measure that quantifies the presence of empty areas in a polygon.

  • Paul Amalaman - “PATHFINDER: A New Bivariate Decision Tree Induction Approach”

    A new bivariate classification tree induction approach designed for datasets with real-valued input attributes is introduced. Unlike previous approaches that compute and store the decision boundaries in the nodes, in our approach decision tree tests compute the best subset of boundary points for a given test example at runtime. Experimental results show that the proposed approach is capable of inducing shorter trees than widely known multivariate approaches and is capable of obtaining very high accuracies for some datasets for the price of not classifying a few examples.

  • Junmo An - “Localization and Tracking of an MR Compatible Manipulator with Computer-Controlled Optically Detunable Inductively Coupled RF Coils”

    Localization and tracking of interventional devices with MR-visible markers is of paramount importance in procedures with real-time MR guidance. In this study we describe a technique for robust localization and fast tracking of the MR-compatible robotic system using multiple MR-visible markers that are selectively tuned and detuned so that only one or a combination of them is visible each time on the MR image. The described technique can be used to track via imaging or projections in the shaft of steerable catheters and the articulated robots.

  • Malcolm Dcosta - “Peri-Nasal Indicators of Deceptive Behavior”

    We use a new physiological measurement methodology, which quantifies stress-induced facial perspiration responses via thermal imagery. A wavelet based signal processing method is used to construct feature vectors of dominant perspiration frequencies. Finally, a pattern classification algorithm is developed to classify the subjects as deceptive or non-deceptive. We tested on data from 67 subjects who faced stressful interrogation for a mock crime. The results show that the proposed approach has achieved a success rate of 80% in blind predictions.

  • Kinjal Dhar Gupta - “Domain Adaptation Under Data Misalignment: An Application to Cepheid Variable Star Classification”

    Domain adaptation helps to build models when the train and the test data have different distributions. We address the case in which the target data differs in distribution from the source data due to displacement in all or some of the features but shares the same feature space. We propose a method that uses maximum likelihood to find the right amount of shift to align the source data with the target data such that the source model itself can be used to classify the target data. We test our method on two classes of Cepheid stars across three galaxies.

  • Tao Feng - “Context-Aware Touch Screen Based User Identity Recognition Under Uncontrolled Environment”

    Due to the increasing popularity of mobile technologies, sensitive user information is often accessed and stored on mobile devices. However, the essential task of mobile user identity recognition is rendered more challenging by the conflicting requirements of security and usability. In this poster, we present a context-aware identity change detection solutions employing touch screen data, motion data, and voice data under uncontrolled environment that takes steps toward addressing this trade-off.

  • Xifeng Gao - “Structured Volume Decomposition via Generalized Sweeping”

    We introduce a volumetric partitioning strategy to partition the volume of an input triangle mesh into a series of deformed cuboids. This is achieved by a user-designed volumetric harmonic function that guides the decomposition of the input volume into a sequence of 2-manifold level sets. A skeletal structure whose corners correspond to corner vertices of a 2D parameterization is extracted for each level set. Intersections of the surface sheets of the skeletal surface correspond to the singular edges of the generated hexahedral representations.

  • Ushasi Ghosh - “Extraction of Underlying Soil Structure from Seismic Data Using Data Mining Techniques”

    Recent advances in data capture, processing power and storage capabilities has enabled us to analyze large volumes of seismic data. In this study we report on the implementation of machine learning and data mining techniques for analysis of seismic data to reveal salt deposits underneath the soil. Several seismic attributes have been extracted from these datasets. Using information gain, the best six attributes have been selected for further classification. Finally we compared the results obtained using different clustering techniques.
  • Dong Han - “Revealing Protocol Information and Activity from Energy Instrumentation in Wireless Sensor Network”

    We present a novel approach to study and reveal network information from energy data in wireless sensor network. Unlike prior approaches, which focused on analyzing the statistical measurement values of power consumption data across the nodes, our approach studies the radio activities on the network. Results from instrumentation on actual testbed indicate that our approach can achieve up to 95% accuracy to identify the routing protocols, and infer the network topology with 80% accuracy on average, 100% accuracy to detect the sink node change.

  • Charu Hans - “Automated Analysis of Zebrafish Vasculature Using Confocal Images”

    Zebrafish has recently emerged as an invaluable model for bio-medical research. Advance imaging technique and rapid development of the zebrafish embryo, is the driving factor for their wide applicability in research area. A high fertility makes zebrafish a good candidate for high throughput application, which turns out to be a highly cost effeffective. However, the lack of tools for automated analysis of complex images presents an obstacle to utilizing the zebrafish as a high-throughput screening model.

  • Kyeongan Kwon - “Interfacing Information in User Studies with Mixed Methods”

    There has been proliferation of mixed methods in user studies. At the same time, user studies have been growing larger in terms of size and duration. This is enthusiasm but also challenging. Researchers have to organize and analyze large amounts of data. Proper statistical analysis is critical and has received a lot of attention. One aspect that has been neglected is a visual interface to the study’s results. Such an interface can support qualitative understanding, conveying at a glance the study’s of sympathetic responses in students taking exams.

  • Yu Li - “Scheduling Transparent Real-Time Virtual Resources”

    The Regularity-based Resource Partition (RRP) Model aims to provide maximal transparency to the task-level scheduling in a 2-layer hierarchical real-time system. This paper investigates the resource-level scheduling problem of the RRP Model based on partitioned scheduling for the first time, introduces the MulZ partitioned scheduling algorithm with multiple EABSes to reduce the approximation overhead, and shows that MulZ performs much better than the current global scheduling algorithms via simulation experiments.

  • Pranav Mantini - “Context Based Trajectory Forecasting”

    One challenging problem in video surveillance or robotics is the design of human motion trajectory forecasting algorithms. We build a model to estimate the occupancy behavior of humans based on the geometry and social norms. We also develop a trajectory forecasting algorithm that understands this occupancy and leverages it for trajectory forecasting in previously unseen geometries (environments). Experimental data suggests a significant enhancement in t

  • Behrang Mehrparvar - “Deep Learning in Pattern Recognition”

    Since 2006, deep learning as a growing branch of machine learning and artificial intelligence has gained huge attention of scholars in academia and industry. The basic idea of deep learning is to provide the ability to learn models of the raw data in hierarchical representations without the limitations of hand crafted feature detectors. In this presentation, we will introduce an overview of application of deep learning in pattern recognition and compare proposed architectures for each specific problem.

  • Ahmad Qawasmeh - “OpenMP Observability via Collector APIs and Tool Support”

    OpenMP tasks present a new dimension of concurrency, which creates new challenges with respect to its observability. We propose new extensions to the OpenMP Runtime API (ORA) to profile task level parallelism. We implement the proposed extensions in OpenUH, which is an open-source OpenMP compiler. With negligible overheads, we capture important events like task creation, execution, suspension, and exiting at both task instance and construct levels. We also integrate our ORA into the performance tool TAU to visualize OpenMP task measurements.

  • Mahbubur Rahman - “A Multiscale Computational Framework to Understand Vascular Adaptation”

    The objective of this work is to understand the vein graft adaptation using a computational framework. The adaptation of vein graft corresponds to two scenarios; either it is intimal or medial growth, which results to the failure of blood circulation through this vein. We would like to explain these scenarios and the key parameters responsible for this. Finally, our research results can be integrated with specific genes responsible for this vein graft adaptation, which will help to determine specific drugs to stabilize the vein graft adaptation.

  • Remi Salmon - “Modeling and Simulation for Breast Conserving Therapy”

    Breast cancer is the most common cancer among women in the world. While early detection and improvements in surgical techniques have increased survival rates, the outcome of breast cancer surgery remains highly visible and affects the quality of life of the patient. We built and validated a multi-scale model for breast conserving therapy that can be used to help the surgeon in his decision-making process.

  • Nripun Sredar - “Examining in vivo Changes in Optic Nerve Head of Non-Human Primates During the Progression of Experimental Glaucoma”

    Glaucoma is the leading cause of irreversible blindness worldwide. The lamina cribrosa, a three-dimensional meshwork of collagenous beams within the optic nerve head, is suggested to be the initial site of damage to retinal ganglion cell axons in glaucoma. We sought to characterize changes in predominant anterior lamina cribrosa surface (ALCS) microarchitecture in vivo during the progression of experimental glaucoma (EG) using adaptive optics scanning laser ophthalmoscope imaging and spectral density optical coherence tomography.

  • Peng Sun - “High Level Programming Model for Heterogeneous MPSoCs Using Industry Standard APIs”

    This poster summarizes the design and implementations created for heterogeneous embedded systems using the multicore communication APIs (MCAPI). The target platform that we have used to evaluate our implementation is Freescale QorlQ P4080 multicore platform, which is of a state-of-the-art eight core Power Architecture. We achieved a higher level and portable abstraction of the communications between the Power host to the Pattern Matching Engine by MCAPI.

  • Salah Aldeen Taamneh - “What Sympathetic Responses Can Tell About Children’s Performance in Reading”

    We investigate the relationship between the children’s performance and their sympathetic responses in reading. We conducted a significance test on the signals from 2 reading tasks assigned for each participant, and we associated that test with the performance. The poor performers exhibited highly expected significant sympathetic responses, while the good performers exhibited mostly expected low sympathetic responses. Additionally, the responses of the good performers with low self-confidence tends to be more similar to responses of the poor performers.

  • Xiaonan Tian - “OpenUH - An Open Source OpenACC Compiler”

    OpenACC is an emerging directive based programming model for heterogeneous system. In this poster, we present the research and development challenges, and solutions to create an open source OpenACC compiler based on main stream compiler framework.

  • Tayfun Tuna - “Text Based Indexing to Ease Navigation in Lecture Video”

    In this work, we present “automatic video indexing” that facilitates easy access to video content and enhances the user experience. Video indexing divides a video lecture into segments indicating different topics by analyzing the text content in the lecture video. Assumption is the lecture includes a sequence of words and each topics within the video has different sets of words or terms. Experiments show that our proposed text based indexing algorithm provides far more accuracy than image based indexing algorithms, 74% vs 43%.

  • Ilyas Uyanik - “Revealing Walking Behaviors via a Mobile App”

    Quantitatively studying walking behavior has been difficult, due to the large of number of subjects needed, the 24/7 nature of monitoring required. We introduce an effective method to quantify calories expended in walking via interpretation of the phone’s accelerometer values. Hundreds of users around the globe took to the application (iBurnCalorie) over the course of half a year, providing a treasure trove of anonymous data. These data give an insight into walking behavior and paint quantitative profiles of ideal users, opening future interventions.

  • Sujing Wang - “New Spatio-temporal Clustering Algorithms for Polygons”

    Polygons provide natural representations for many types of geospatial objects. We propose two new spatio-temporal clustering algorithms, called Spatio-Temporal Shared Nearest Neighbor (ST-SNN), and Spatio-Temporal Separated Shared Nearest Neighbor (ST-SEP-SNN) to cluster overlapping polygons that can change their locations, sizes and shapes over time. The effectiveness of our algorithms is tested in a real-world case study involving ozone pollution events in the Houston-Galveston-Brazoria (HGB) area.

  • Cheng Wang - “High-Performance Parallel Sparse FFT Algorithms for Multicore CPUs and GPUs”

    Fast Fourier Transform (FFT) is a widely used numerical algorithm in scientific and engineering applications. When the input signal is sparse, i.e., most of the Fourier Coefficients are zero or negligibly small, FFT is clearly inefficient. In this paper, we present a high-performance parallel sparse FFT algorithm to compute the Fourier Transform on multicore CPUs and GPUs. Our sparse FFT is over 10x faster than state-of-the-art FFT libraries.

  • Rengan Xu - “Reduction Operations in Parallel Loops for GPGPUs”

    In this poster, we present the design and parallelization of reduction operations in parallel loops for GPGPU accelerators using OpenACC programming model. The algorithms are implemented in OpenUH compiler. Comparing to other two commercial OpenACC compilers, the result demonstrates better robustness and competitive performance than others.

  • Olga Datskova - “Cloud Computing: Tackling Key Challenges”

    The fad of Cloud Computing (CC) is over. CC technology has matured to the point of being ready for considerable financial investment and research interest. While migration to cloud is seen as cost-effective and a promising computing model for companies and users, it is still riddled with infrastructure, security and service provision issues. The purpose of this work is to provide an overview of CC and its challenges. We aim to summarize possible solutions for the major concerns and propose an API framework architecture that addresses these challenges.