Symposium 2015

Kenson Cai 
Monitoring Water Dynamics using Remote Sensing Data 
Advisor: Vipin Kumar 
Home Institution: University of Minnesota - Twin Cities


Abstract: Freshwater, though vital to all life forms on Earth, is becoming increasingly scarce due to droughts, decreasing amounts of groundwater, and more. As a result, we want to create a global water monitoring system that can display information of inland water bodies over a regular period of time. This project looks at classification algorithms to classify every pixel (a rectangular area of 0.25 sq. km.) on the Earth as land or water given the remote sensing data obtained from satellites. In this way, changes in water body formations can be seen over time. We trained predictive learning tools such as support vector machines over training data before using them to analyze the rest of the data. Several problems arose in the forms of heterogeneity, lack of training instances, and noisy/missing values. Heterogeneity problems occur when land features look similar to water features in different parts of the Earth while cloud cover caused the missing values. We addressed these problems by designing algorithms specific to certain land/water features and that could generate accurate results from looking at already classified surrounding areas. The results were then stored, through a multi-step process, into a database that can be displayed on a publically accessible web interface created for increasing public awareness.

William Flotte†, Caleb Larson, Duc Tran 
Creating a multisensory presence in the virtual world: Toward virtual reality (VR) applications for the management of anxiety and pain.
Advisor: Daniel Keefe, Lana Yarosh 
Home Institution: †Brown University 

Abstract: We present an exploration of new possibilities for creating more effective, realistic VR applications for the management of anxiety and pain utilizing Bluetooth, haptic feedback, hand tracking and sound. It is now possible to create VR experiences using low cost and therefore widely accessible smart-phone based systems. Early versions of these low cost systems have focused on the visual component of the virtual world. In contrast, our goal is to create virtual worlds that engage several senses including visual, auditory, and tactile senses. The rationale is that the resulting widely accessible, multi-sensory experiences will increase realism, engagement, and presence within the virtual environments and thereby serve as a more effective distraction from persistent and chronic pain or anxiety. To accomplish this, we utilize the Bluetooth communication protocol to create cordless connections between a smart phone and a variety of custom-built haptic devices. The smartphone acts as a VR viewfinder and is programmed to display a stereoscopic 3D view of the virtual world, and the devices provide haptic feedback to the user. To date, we have built two-such haptic device prototypes and coupled them with virtual environments that we believe can eventually be utilized as part of pain and anxiety treatment protocols. The first environment is a soothing, sitting meditation in an ocean cove complete with sounds and visuals of waves gently rolling in. The custom haptic device vibrates under the participant†¢s feet in time with the visuals to create the feeling of the waves. The immersive environment helps the user focus on something other than their pain. The second environment is a digital animal companion that sits on the participant†¢s lap, designed to help reduce anxiety. The custom haptic device responds to the users†¢ petting gestures in real world by vibrating and in the virtual world by changing color and audibly purring. The key results of this work in progress include identifying the most appropriate technical mechanisms (hardware and software) to build the two prototypes and early demonstrations of the potential for these enriched, low-cost, mobile virtual experiences to assist in managing anxiety, stress or pain.


Alex Foley 
Avatar Embodiment and Animation in a Multi-User Virtual Reality Environment 
Advisor: Dr. Victoria Interrante 
Home Institution: Lawrence University 


Abstract: Creating and animating realistic avatars based on head tracking data is important for immersion in a MuVR (multi user virtual reality) environment using HMDs (head mounted displays). When multiple users participate in the same virtual environment, they must know the location of other users in both the real and virtual spaces and they must be able to interact. While a 3D cylinder can give a users†¢ position, a realistic avatar provides for more accurate representation of nonverbal behaviors such as nodding and seeing where a user looks. This presents two major technical challenges: creating accurate avatars from real people, and animating those avatars based solely on head tracking data. By utilizing Body{SNAP}, we are able to create a virtual avatar from Kinect scans of a person. Texture data is not preserved, but this method is able to quickly and accurately capture body shape and facial features while still providing an avatar that is robust and fully capable of animation. Once the avatar is created and imported it needs to be animated from the head-tracking data. While full body tracking is theoretically possible, it is cumbersome for the user and requires the use of additional tracking hardware which may not be readily available. Our technique seeks to infer reasonable full body tracking data solely from the head data and determine if the person is rotating, walking, or even stationary and only moving their head. This allows the avatar to walk, rotate, and freely move their head as the user does. During our initial lab tests, the results have been very positive, but more refinement is needed to better match the animation speed with the actual distance moved and increase the motions available to the avatar.

Jeremy Hamning 
Evolution of the Influenza Virus 
Advisor: Daniel Boley 
Home Institution: Normandale Community College 


Abstract: In our research we expanded on previous work by exploring the evolution of the H1N1, H3N2, and type B influenza virus strains. Taking sequences from the online NCBI flu database and converting the sequences to binary using the technique described in [Lam et al., 2015] we were able to run a principal component analysis on the sequences. Graphing the first two principal components of the results we were able to see the path of the virus evolution. In addition, we also performed a pairwise Hamming distance calculation on the PB1, HA1, HA2, and NA proteins in each H3N2 strain. We used this to measure the rate at which these proteins evolve in relation to each other in order to see how much the outer proteins (HA1, HA2, NA) change as a response to influenza vaccines.

Corrie Moore†, Helen Dougherty‡ 
Haptic Perception and Visual Dominance
Advisor: Victoria Interrante 
Home Institution: †Rhodes College, ‡Grinnell College 


Abstract: Haptic perception and visual dominance are key factors in making virtual environments more immersive and useful, especially in professional and architectural settings. In order for virtual environments to be effective, the perceived scale of objects in the virtual world needs to be identical to the scale of objects in reality. We are conducting a user study to examine these questions. Participants will be immersed in a virtual environment in which they can see and feel a rectangular block. The study will involve three trials: one in which the user has no visual indication of hand position; one in which the user’s hand position will be relayed visually via avatar hands; and one in which the user will see a live see-through feed of their own hands. During each trial, the physical block will be of various lengths, and the virtual block will be larger, smaller, or true-to-size in relation to the physical block. The user will be asked to report information about the block and their haptic and visual experience during each of the three trials. By representing the user’s own hands in a more realistic manner through the video input, we are thinking that we will be able to get them to believe more strongly that what they are seeing is what is what they are feeling than if the hands were embodied as a generic avatar or not embodied at all. Other work that can stem from this experiment could involve studies focusing on the extent to which visual dominance could be exploited to alter the user’s perception of a physical object. 

Luka Milekic
Virtual Humans using Dynamic Obstacle Avoidance
Advisor: Victoria Interrante
Home Institution: Brandeis University


Abstract: The goal of this project was to create virtual humans who realistically wander around a virtual space in order to allow architects to see what their designs would look like when filled with people. There are two reasons for having a virtual space populated with virtual people. First is to get a better idea of scale of the architecture. Second is to make a space seem more "alive" and realistic. This second point is more than for aesthetics though, for many spaces, such as hospitals, it’s very important to see how people move around in the space, because places that are difficult for many people to get through quickly could be the difference between life and death for someone.

Allison Pine 
Predicting Protein Targets of Compounds by Combining Chemical Genomics and Blind Docking 
Advisor: Chad Myers
Home Institution: Tufts University


Abstract: Determination of compounds that target specific diseases or bind to known proteins before expensive clinical trials is a critical and complex problem. The common process of docking generally entails screening a large compound library for energetically favorable poses and scoring the results to determine compounds most likely to bind to a given target protein, and therefore, mostly likely to be effective drugs for the disease implicated by that target. This process, however, requires huge computational power to dock libraries of potentially millions of compounds against hundreds or thousands of possible target proteins as well as detailed information about where on the protein a compound will bind (the protein†¢s binding site). Yeast chemical genomic interaction studies of large compound libraries have identified and scored both compounds†¢ target gene predictions and predictions of what functional processes compounds are likely involved in. We propose a pipeline that combines these two complementary approaches by leveraging the chemical genomic pipeline to generate a list of candidate proteins for each compound of interest based on chemical-genetic interaction profiles. From these candidates, docking can then be used to further identify the correct binding protein. By incorporating chemical genomic functional data, this approach is less computationally intense and more selective than structure-based docking alone. Additionally, by evaluating the entire surface of each protein, or blind docking, this method requires no prior knowledge of the protein's binding site.

Jonathan Castellanos-Gomez†, Nasar Roble‡ 
Peer-to-Peer Framework Based On Mobile Distributed Engine 
Advisor: Chandra Abhishek, Jon Weissman 
Home Institution: †Normandale Community College, ‡Saint Paul College

 
Abstract: Smartphones have become pervasive in modern day, and while the technological specifications improve yearly, they continue to experience notable resource constraints and limited battery capacity. Recent wearable smart technology such as Google Glass, though less ubiquitous, share equivalent drawbacks in their resources limitations. This project addresses the issue by creating a peer-to-peer framework that allows users to request pre-cached geo-locations from nearby mobile devices to provide quick, efficient, and fault-tolerant sharing of route information to nearby locations of interest. This framework is built upon an existing three-tiered architecture and middleware that encompasses Amazon EC2 Cloud, Android mobile device, and Google Glass. The Android device and Google Glass relay their geo-coordinates to the cloud, which will create a geo-spacial model containing directions to locations that the user may be interested in. By offloading heavy computations from the end-devices to the cloud power dissipation can be significantly reduced, as well achieving a reduction in latency by pre-fetching and pre-caching results from the cloud onto the mobile device. We enhance this architecture by implementing a peer-to-peer framework that allows sharing of pre-cached data through Wi-Fi Direct. Rather than making a request to the cloud, the user can query whether the desired resource can be obtained from a nearby peer. With the abundance of smart devices presents in public areas, reliable communication and data transfer can be achieved even in the absence of network connectivity. Our evaluation demonstrates the effectiveness of the architecture and middleware in reducing delay in resource retrieval and improving battery longevity on both Android device and Google Glass.

Tyler J. Skluzacek
Multi-Tiered Storage for HDFS using Tiera
Advisor: Abhishek Chandra 
Home Institution: Macalester College 


Abstract: The Hadoop Distributed File System (HDFS) is a distributed file system for storing large data sets used by different computing frameworks for big data computing and key-value stores. Traditionally, HDFS has run on commodity hardware native disks, but the advent of new storage devices has motivated the need to run HDFS on different storage devices, including solid state drives (SSDs). The diversity of applications using the storage devices warrants the implementation of unique storage policies derived from the requirements of said applications. Presently, more systems are moving onto the cloud and attempting to utilize new cloud technologies provided by each cloud provider. Hence, it has become of prime importance to provide a standard and easily implementable solution to provide multi-tiered cloud storage for HDFS. In this paper, we first illuminate the need for multi-tiered storage for HDFS by surveying different Hadoop workload studies and Hadoop systems. Then we present a novel solution for multi-tiered storage in HDFS using TieraFUSE. TieraFUSE provides first-class support for encapsulated tiered cloud storage and ease of programmatically, while providing storage policies for these tiers. In the end, we support our design by giving some experimental results for a few example workloads on both single-node and multinode clusters.

Andrey Smirnov
Correlation of centrality measures with the node’s influence in the disease spread model
Advisor: Daniel Boley 
Home Institution: University of Massachusetts Amherst 


Abstract: Centrality indices are important measures of degree involvement in the graph structure. In the past, numerous centrality indices has been developed to identify important properties of the graph nodes that contribute to the cohesiveness of the network. However, studies have shown that centrality indices are application-specific and some centrality indices are more suitable in certain areas than others [1]. In this study, we analyze how well certain centrality measures correlate with a node’s prominence in terms of node’s influence in the epidemiology model. In particular we examine such standard measures of centrality as betweenness, degree, closeness, and current flow centralities. Furthermore, we expand on the very recent work and evaluate the avoidance-hitting time pivotality metric [2] for application of the epidemiology model using the same analysis. We estimate the contribution of a node to disease spread by simulating the compartmental epidemiology model with and without that node and calculating disease extinction time and the number of diseased agents. We then find the correlation of these results with the centrality measures of a node. Finally, we run our test on both the synthetic examples of a network as well as real life ones such as a subset of a social network from Facebook.

Henry Ward† , Varun Mangalick 
Obesity, Geography and the Comparative Diversities of the Human Gut Microbiome 
Advisor: Dan Knights 
Home Institution: †Lawrence University 


Abstract: In the past two decades, much research has gone into establishing a link between obesity and altered expression of the human gut microbiome (Turnbaugh et al., 2009), as well as into exploring the hygiene hypothesis linking decreased hygiene to improved defense against allergic illness (Schaub, Launer, & von Mutius, 2006). However, relatively few large studies based on fecal sample sequence reads utilize sample BMI measurements as crucial metadata, or measure potential differences in the human microbiome based on geography. Using existing data (Yatsunenko et al., 2012), we performed a range of existing QIIME tests (Caporaso et al., 2010) as well as novel tests, which measure unique vs. shared OTUs between samples, to compare the diversities of the Venezuelan, Malawian and American obese and lean microbiomes. We provide more evidence for the hygiene hypothesis, showing that while all infants possess highly individual microbiomes, adult microbiomes are far less individualistic. Moreover, American adults share less in common with each other’s microbiomes than adults in the non-Westernized countries. This effect is not driven by obesity. However, American obese adults share more with each other’s microbiomes than lean adults.