As the field of genomics continues to grow, there is an overwhelming need for robust ways to assemble entire genomes, yet there is little guidance on what software best achieves this goal. An assessment of available genome assemblers is needed to improve assembly procedures. Genome assemblers output draft assemblies consisting of DNA scaffolds that are permeated with errors inherent to sequencing data, like wrong number of repeats, misjoins, inversions, missing sequencing, and extraneous sequences. These draft assemblies can then be input into an experimental assay called HiC that attempts to fix and orient the assemblies by taking into account chromatin contact interactions to elucidate genome architecture. The resulting genome is then more accurate and useful for research purposes. However, given the plethora of genome assembly and HiC software available, navigating each available software is time consuming and often requires extensive testing before actual use can begin. To improve this process, we aim to review several HiC scaffolding software tools for performance and quality, and fine tune their parameters for optimized outcome. This will allow us to suggest which software work best for a range of genomic data as well as provide summary statistics and accuracy measurements for each genome assembler tested. We tested several genome assemblers, including GRAAL and SALSA, which utilize HiC libraries to scaffold draft assemblies. After being produced and corrected through assemblers, these assembled genomes can then be used to advance research in functional genomics, genetic epidemiology, and many other fields reliant on genomic data. It is our goal to improve genome assembly procedures so that more researchers can use open source software to build and understand entire genomes and further the field of genomics.

Vagus nerve stimulation (VNS) involves the delivery of electrical stimuli to the vagus nerve, which is associated with several brain regions and functions. VNS is currently used as auxiliary treatment for some types of epilepsy and there is significant interest regarding its potential in the treatment of other illness as well as for cognitive augmentation and promoting plasticity. The goal of this project is to investigate the differences in ipsilateral and contralateral cortical responses to different vagus nerve stimulation protocols, as well as to compare the responses to invasive versus noninvasive VNS delivery methods. Data was collected from a non-human primate using dual electrodes (with epicortical and intracortical contacts) placed in the prefrontal, premotor, supplementary and primary motor and parietal cortical areas, as well as a nerve cuff placed on the left trunk of the vagus nerve. Data analysis focused on the cortical evoked potentials elicited by VNS and recorded at cortical sites both ipsilateral and contralateral to the VNS. VNS was delivered to both the auricular branch of the vagus nerve in the noninvasive protocol and through a nerve cuff on the left vagus nerve trunk for the invasive protocol. This work is related to a larger project focused on establishing protocols for noninvasive VNS to augment targeted neuroplasticity and enhance cognitive performance in normal nonhuman primates. This nonhuman primate study will be directly applicable to the development of noninvasive VNS technology that can be used to enhance neuroplasticity and cognitive performance in healthy adult humans.

For those interested in understanding more about fatal shootings of civilians by police, a natural place to start would be data on the events. Such data is, however, surprisingly hard to access. There are no official sources, but there are several crowd sourced data sets available online. These data sets each contain some common and some unique fields, and they are stored in very different formats. Hence, these data sets require preprocessing before the data can be used for other practical purposes like analysis and visualization. In this research, we focus on cleaning such heterogenous data sets using statistical methods with major focus on data matching. Data matching refers to matching data from different sources and recognizing the matching entities. It is an essential step in data cleaning that helps in resolving the conflicting information provided by the data sets---clerical errors, missing data, differently stored data, etc. Traditional statistical procedures and classification algorithms are based on supervised learning which requires training data. However, it is quite common that the training data is not available at hand and it might not be cost effective to prepare one from scratch. We explore a two step unsupervised learning algorithm that might achieve the same performance as the supervised counterparts. In the first step, it prepares a seed data set containing data points that are relatively extreme matches and non-matches based on discriminating fields. This prepared data set can be used as training data to train classical classification algorithms---k-nearest neighbors, Support Vector machines, Random Forests---which in turn can be used to classify the remaining data points. In the process, we examine missing data, geocode matching, and feature selection methods that increase classification accuracy by selecting the most discriminating fields. Finally, we examine the prospect of generalizing this approach to suit differently aligned data.

The Puget Sound is a complex estuarine system within the Salish Sea, fed by both high salinity water from the Pacific Ocean and freshwater from a number of rivers. The Snohomish River is the second largest input, transporting freshwater from the Skykomish and Snoqualmie rivers to Port Gardner Bay off the coast of Everett. At its mouth, the higher density salt water from the Puget Sound intrudes into the freshwater, forming a salt wedge that causes a highly stratified water column which rapidly changes with the tidal cycle. These mixing dynamics may be the primary driver of biological productivity in the estuarine system. To characterize the density profile relative to tidal patterns and season, current speed and velocity were recorded utilizing a Nortek ADCP. During normal outflow (3,500 f^3/s) of the Snohomish River, upward velocities were positive during flood tide and negative during ebb tide, following a typical salt-wedge trend. During a peak outflow event (56,000 f^3/s) on October 22nd, this tidal cycle dependence was disrupted and downwelling dominated in the form of negative vertical velocities throughout the water column. In addition, the magnitude of east/west velocities increased during peak outflow. This signaled a period of intense turbulence at the bottom of the water column. During the same time span, data concerning turbidity and chlorophyll levels were recorded utilizing a Seabird CTD. No clear trend was evident in chlorophyll levels at both normal and peak outflow timespans. Turbidity maintained high levels (35 NTU) following peak outflow, possibly reflecting the transportation of sediment from the river bed into the water column. This transportation would be caused by the turbulence reflected in the higher magnitude of the east/west velocity data.

Modes and forms of communication have evolved drastically over the last several decades, with the emergence of large social media platforms and an increased accessibility to information over the Internet. Traditionally, journalists played the role of both providing and curating information to the general public, but with these new forms of communication, the role of journalists is evolving too. We examine how this role is changing in the crisis context. Specifically, we look at the interactions between journalists, citizens and government information workers after the Oso Landslide that occurred on March 22 of 2014. We analyzed the tweets from this event to see the interaction between these groups at specific points in time as the crisis unfolded. From a collection of ~1 million tweets, we conducted a micro-analysis of two important time periods: 1) all tweets about the Slide in the first two hours after it occurred and 2) all tweets over the first three weeks that included the hashtag created by the lead response agency (530slide.) We combine ethnographic analyses, descriptive statistics, and network analyses to better understand how contemporary crisis information is shared among these groups as presented on Twitter. Preliminary results show that there were common content themes that each represent a specific ‘information need’: situational awareness, public safety, aid, and commentary. 

This research focuses on the mathematics of gerrymandering. Gerrymandering refers to how political parties draw district boundaries to give them better odds at receiving a majority during congressional elections. Every 10 years district boundaries in the United States are redrawn for congressional elections by the current majority party. With such change in political cartography, redrawing district boundaries not only affects who wins but also the people who live within them. Given the loosely defined criteria for redistricting, we seek to quantify the process of redrawing district lines to determine when this partisan process becomes an act of manipulation and exclusion. We will be use The Metropolis Hastings (M.H.) Algorithm to produce a random walk that has a stationary distribution that will be the probability distribution of all possible districting plans for a given U.S state. By collecting congressional maps and election results, we implemented the M.H. algorithm in python to produce sample spaces for Washington State and Iowa. We will simulate local and federal redistricting requirements using self-defined parameters which measure: compactness of districts, population division, division of counties, and minority voter populations. We simulated election results with respect to our sample districts and compared the simulated outcomes with actual district maps and their election results. This study helps to determine if current redrawn maps legitimately reflect the people they purport to represent or if they are manipulations and subversions of our democratic process.

Synthetic biology allows us to use an organism’s metabolic pathways to produce new, useful metabolites cost-effectively. Growing organisms in a culture allows the production of these metabolites on a commercially-viable scale. However, management of these cultures is time-consuming and labor-intensive, such as HPLC. Our project aims to provide a low-cost, automated system that analyzes the culture in real time at low cost. Our device, a turbidostat bioreactor called the Chromastat, uses images and other optical data to measure the current state of a culture. If these measurements show metabolite production outside specifications, the system modifies the culture’s metabolic expression by adding inducers. To make this system cost effective, we focused on using affordable materials, an open-source controller and analysis system, and visible color outputs instead of the traditional fluorescence. To test our system, we used the violacein metabolic pathway in yeast. By regulating gene expression with inducible promoters, this pathway yields up to four visually distinct pigments. In a production environment, these pigments could provide a proxy measurement to produce useful substances which are costlier to detect. Pigment production is measured by the popular low-cost and open-source Raspberry Pi computer running a program displayed in Java Swing. Optical sensors collect color and opacity information about the culture. Our software, installed on the Raspberry Pi, analyzes the color, and in response, the hardware introduces inducers to stabilize the culture’s production. Data recorded by the Pi over time reveals the relationship between gene expression and metabolite production rates. By combining biological, software, and hardware systems, our unique design can generate previously unavailable visual data in certain biosynthesis processes, such as those involved in antibiotic production or fermentation.

Lateral flow tests are assays conducted on porous media, commonly referred to as paper, that can detect and quantify analytes in a sample. They have the potential to return results within minutes and can be used in low-resource or non-medical settings. Protein adsorption is critical in these tests because protein interaction with a media can affect the delivery of reagents or choice of blocking agents to prevent nonspecific binding. A computational model of the fluid and adsorption physics may improve the development of paper-based assays. Yet-to-be-published data from the Yager group evaluates protein adsorption based on the deposition and analysis of protein spots on paper. The computational model used is derived from Darcy’s law, commonly used to model fluid flow in porous media and to calculate volumetric flow rates. This model makes two main assumptions: the porous media is fully saturated, and there is zero outflow on the boundaries of the media. These assumptions are not applicable because the protein spot dries as a result of evaporation, creating a non-zero outflow at the surface of the paper. The saturation level in the paper will also vary therefore Darcy’s law is not an adequate representation of the paper. My project aims to develop a more accurate computational model of fluid and protein absorption dynamic by modeling protein spots on paper. I incorporated an evaporation component that will allow the use of Richards equation to model fluid flow, which applies to unsaturated conditions. This is coupled with a protein transport and adsorption component to achieve a complete model. Although literature on evaporation from sand or soil is abundant, these results cannot be directly applied to evaporation from paper. This model will aid the diagnostic community in better understanding their assays and ultimately help improve the sensitivity and specificity of their assays.

One issue in many medical applications is detecting the presence of forced exhalations or breaths through audio data. For example, analysing sleep disorder breathing or analysing a person's lung function from audio of their exhalation first requires localization of a breath event in an audio file. Past work has focused on extensive feature extraction from sound and machine learning methods to detect these events. However, these black box methods require significant feature engineering dependent on the type of exhalation and environment the audio data is recorded to find the start and end of an exhalation event. We present a novel method for detecting the start and end of forced exhalations for spirometry, a medical test of lung function, in audio files. The algorithm requires an initial set of audio files containing exhalations to be manually labelled for locations where exhalations can be heard. Then, we build an envelope which encapsulates the shape of our exhalation signal. We take the cross-correlation of this initial envelope with vast amounts of unlabelled audio data containing exhalations, and use the estimated locations of exhalation events to recalculate the envelope. We repeat this process with the new envelopes until convergence. The converged envelope is then used to autocorrelate an audio signal to detect the start and end of exhalation. While preliminary results show that this method only achieves comparable performance to ad-hoc approaches currently used compared to human labels, it has the benefit of being simpler and more interpretable, enabling easier improvements and runtime of the algorithm.

Have you ever misunderstood someone because of their pronunciation? Although pronunciation is a vital part of foreign language “fluency,” it is often overlooked in classrooms in favor of vocabulary and grammar training. The sounds present in, and absent from, one’s mother tongue directly impact foreign language pronunciation. I propose that exposing language learners to phonemes that are not present in their mother tongue will help improve foreign language pronunciation. I examine the case of Korean language learners in particular. Korean learners were recorded saying a list of phrases in Korean before testing. Then, they listened to artificially slowed clips of native Korean speakers saying phrases with phonemes absent from the learners' native languages. After hearing the slowed clips, learners were then recording saying the same phrases as before. I expect that Korean language learners will show an improvement in pronunciation after listening to the slowed clips of native Korean speakers, especially for the phonemes that are not present in their native languages. The results of this research may be extended to learners of languages other than Korean, as well. This could hold implications for the future of language learning in our interconnected and increasingly globalized world.

Ethics and diversity are critical components of engineering training and practice, but most undergraduate engineering programs do not address these issues in-depth. We describe the design and implementation of a novel curriculum that allows early engineering students to explore the interplay of diversity and ethics in an engineering context. Although upper-division bioethics courses may also address this relationship, we chose to implement this curriculum in a large, introductory course. We intend to provide an accessible opportunity for early engineering students, particularly underrepresented students interested in engineering, to engage with this material early and to impress upon a broad audience that diversity and ethics are critical topics in engineering. Engaging these students in meaningful discussion about the intersecting roles of diversity and ethics in bioengineering enables them to apply course competencies to their future engineering practice. Our learning objectives include: (1) Summarize key case studies regarding diversity-related ethics in STEM, (2) Identify how cultural concepts of race, gender, sexuality, and disability have shaped scientific thought (and vice versa) through history, (3) Critically evaluate literature regarding ethics and diversity in bioengineering, (4) Analyze how engineers handle implicit bias during research and design processes, and (5) Propose approaches to promote ethics and diversity in engineering practice. Course activities cover the importance of diversity and ethics competency in engineering; historic and current case studies of diversity-related ethical issues and how historical perceptions and contexts still influence modern scientific thinking and engineering design; advocacy and representation of minorities in engineering; evidence supporting the value of inclusive teaching and diverse teams; and best practices for advocacy and representation of diverse peoples in engineering. We assess the effectiveness of these teaching innovations through student surveys, student performance on assignments, and instructor observations. Additional supporting data is provided by excerpts of student work.

Coreference resolution, the process of identifying expressions that refer to the same entity in discourse, is an important task in natural language processing.  Abstract anaphora are distinct from other types of referring expressions because their antecedents are abstract entities, which can surface in non-nominal form. Abstract pronominal anaphora pose an interesting challenge for anaphora resolution due to the variation in the form and semantics of possible antecedents, which are associated with a small set of pronouns. The goals of this study are to investigate the distribution of abstract pronominal anaphora by analyzing language in use and to examine the role of abstract pronominal anaphora in resolution systems. I will investigate the distribution of the referential pronouns ‘it’, ‘this’ and ‘that’ by annotating and analyzing a corpus of five English registers: newspaper text, academic writing, fiction, spontaneous dialog and planned speech. I will construct these five subcorpora to include a sufficient amount of text for analysis. I will analyze the role of referring pronouns, anaphoric distance, and the structure and semantic category of abstract antecedents. I will compare the distribution of these features within each register and the corpus as a whole.

The US homeless population is disproportionately affected by alcohol use disorder (AUD). This disparity leads to greater health and behavioral problems which precipitates greater utilization and associated costs of publicly funded services, such as the emergency healthcare and criminal justice systems. To address AUD in this population, researchers worked with homeless individuals with AUD and agencies that serve them to develop the Harm Reduction Treatment for Alcohol (HaRT-A). HaRT-A participants evinced improved alcohol outcomes compared to treatment-as-usual control participants. The present study’s aim was to test a secondary research question: Does HaRT-A significantly reduce the high costs from emergency healthcare and criminal justice system utilization in people experiencing homelessness and AUD? Participants (N=167; 24% women) were recruited for the parent HaRT-A study. Cost data were collected from participants’ use of the King County Correctional Facility and Harborview Medical Center. Jail costs included the costs of booking, days spent in jail, and use of supplementary jail services. Emergency healthcare system costs included the costs of emergency department visits and inpatient hospital admissions. All costs were added up for the 24 weeks prior and subsequent to study participation. Mann-Whitney U tests were conducted to assess group differences on post-pre cost change scores. Analyses indicated no significant group differences on post-pre changes in costs associated with utilization of emergency healthcare and jail services (ps > .58). There are several possible explanations for the lack of significant group differences. First, the time frame was likely too short to detect differences in statistically rare outcomes. Second, HaRT-A had a statistically significant impact on proximal outcomes; however, the brief nature of the treatment (4 sessions) might have limited its ability to impact the more distal variables measured in this study. Finally, homelessness creates multiple challenges that may require more intensive and multilevel interventions to fully address.

Pregnant women often get their bellies touched without their permission by people ranging from family members, to coworkers, to strangers. While usually well-intentioned, people who touch pregnant bellies are unaware that their actions may be unwanted and invasive. Previous works, such as Asta Roseway’s Printing Dress, have shown that wearable computing can bring attention to social issues by provoking discourse. Those works lead us to our wearable design that highlights the issue of personal space for pregnant women. We implemented a maternity shirt that reacts to inappropriate touches upon contact in various ways, such as displaying messages, vibrating, and setting off an alarm. We plan to deploy this shirt on experimental participants and gather qualitative data in addition to the quantitative data collected through the shirt about when and where touches occur. Moreover, findings from the experiment will provide insight into pregnant women’s experiences, and how to design wearables that benefit to pregnant women. Our future goal is to have the shirt deter the toucher as they approach the belly. Through this interactive design piece we hope to raise awareness and provoke conversation around themes of consent, inappropriate touching, and women's bodies.

Developmental Epileptic Encephalopathy (DEE) is a severe form of epilepsy in children and infants characterized by prominent EEG (electroencephalography) abnormalities disrupting brain function. Identifying the genetic causes of DEE will help researchers develop medical treatments for affected patients. Approximately 5% of DEE cases involve a copy number variation (CNV), where a region of DNA has been duplicated or deleted in a gene, causing the disease. Historically, this type of mutation has been difficult to detect using sequence data. To analyze the large amount of sequence data available in the Mefford Lab, I am creating an algorithm that will analyze smMIP (single molecule molecular inversion probe) data to identify CNVs in patient DNA that are likely causal for DEE. The algorithm is a two-step process. First, z-scores are assigned to the number of smMIP reads at each target site of each patient sample to numerically assess how many standard deviations it is from the mean. Next, regions of contiguous significant z-scores will be extracted as areas of top interest to verify as being a CNV. I will run the completed algorithm with smMIP data from the Mefford Lab's unsolved DEE cases. Since CNVs have historically been difficult to detect, it is unclear what percent of these unsolved cases will contain a CNV. It might be that, due to previous lack of detection, CNVs play a more significant role in DEE than previously thought. The opposite might also be true if a large majority of unresolved cases are due to other types of mutations, e.g. point mutations. Regardless, the algorithm will shed light on new causative CNVs and allow the Mefford Lab to solve more patient cases. Additionally, a successful CNV calling algorithm can be shared with other labs to aid them in their research.

The first-year population, with students transitioning from high school to college, is full of novel experiences. These experiences, be they academic, social, etc., influence the well-being of the student. The purpose of the study is to understand the first-year experience in a data-driven manner to create a cornerstone for institutional change. Although replicating a previously piloted study at Carnegie Mellon University in Spring 2017, this study extends focus to investigating stressful life events of first-years. Over the 2018 Winter and Spring quarters, a participant sample size of around 200 UW first-year undergraduate students (around 50% pre-/engineers) contribute data to the study. Data consists of both active and passive collection methods: participants actively respond to Qualtrics surveys, with questions investigating stressors, and passively via our phone software and FitBits. The two quarters of this study will yield tangible data that can be analyzed to advance the science of behavioral health and trauma. This data is unique in providing a picture of student behavioral changes after stressful experiences. Additionally, this advancement in science can become a basis for institutional change for student wellness. As the first-year population is only a slice of the overall student experience, we hope to expand this study further by encompassing the graduate student experience and the full four-year college experience in addition to a larger sample size.

Over the past few years, the presence of smart homes has increased rapidly. Previous research has observed that people make errors in smart home programming with their mental models differing from their actual programming implementation which can cause serious security and safety concerns. The most common errors include missing half-rules and the incorrect use of conjunctions in triggers. For example, if users desire to program a rule for turning on the light when they are at home, they might program a rule such as “when I come home, turn on the lights”, forgetting to program the rule for “when I leave home, turn off the lights”. Our research focuses on reducing user errors in end-user programming by using programming by demonstration. Our hypothesis is that when users physically demonstrate a trigger-action rule, they are less likely to commit the trigger-action programming errors explained above. We compare the number and types of errors made within the IFTTT (if-this-then-that) interface that does not implement programming by demonstration and two interfaces that we built that implement programming by demonstration: a web interface and an augmented reality based interface within the Microsoft HoloLens. If our hypothesis holds true, we might observe that programming by demonstration significantly reduces programming errors in end-user programming in the case of both missing half rules and incorrect trigger conjunctions. We postulate that there might not be a difference in the case of error reduction between both the web and augmented reality interface; however, the augmented reality interface might provide for a broader range of programming by demonstration interfaces such as those related to temperature, humidity etc. Moreover, if our hypothesis holds, we might further suggest that smart home providers incorporate programming by demonstration models in their programming processes for end users within smart homes.

At a time of heightened global attention to the threat of misinformation, disinformation, political propaganda, and the role of technology in facilitating their spread, this research addresses the media ecosystem encompassing the Syrian Civil War. We examine the competing narratives surrounding the role of Syria Civil Defence, a volunteer humanitarian organization popularly known as the White Helmets, working in war-torn Syria. Mainstream media articles sympathetic of the White Helmets and their efforts brought global awareness to the plight of Syrian people, especially those resisting the Assad government. However, this narrative was not aligned with other views of the complex geopolitical landscape of the Syrian conflict, and in response counter-narratives challenging the White Helmets emerged and spread to other online communities. Using a mixed-method approach based on seed data collected from Twitter, and then extending out to the websites cited in that data, we examine content sharing practices within news articles across distinct media domains that functioned to construct, shape, and propagate these narratives. We articulate a predominantly alternative media “echo-system” of websites that repeatedly share content about the White Helmets. Among other findings, our work reveals a small set of websites and authors generating content that is spread across diverse sites, drawing audiences from distinct communities into a shared narrative. This analysis also reveals the integration of government-funded media and geopolitical think tanks as source content for anti-White Helmets narratives. More broadly, the analysis demonstrates the role of alternative newswire-like services in providing content for alternative media websites. Though additional work is needed to understand these patterns over time and across topics, we provide insight into the dynamics of this multi-layered media ecosystem.

An important and under-explored component of information exchange and diffusion on social media involves permutations—i.e. how information changes, or more accurately, is changed as it flows. In this research paper, we aim to identify the various rumors and narratives that exist within a conversation surrounding Omran Daqneesh, the ‘Aleppo Boy’, on Twitter, and to investigate the actions that users take to propagate a certain permutations of these narratives. Before we can see how narratives propagate, we must first seek to understand these narratives at various levels of analysis, representing both the individual and society more generally. To this end, we examine narratives within the conversation at various levels: the micro, meso, and macro, representing individual actions and rumors, context-specific narratives, and a higher-level beliefs and worldviews. In this research we adopt a mixed methods approach. We quantitatively analyze our data to gain insights into the key events and influential actors in the space, and qualitatively code tweets to identify 1) rumor permutations that contribute to larger narratives surrounding the Syrian conflict; and 2) specific actions that help us to understand how individuals interacting within this information space. We aim to build a deeper understanding of how rumors are spread, how disinformation effects the online information space, and to develop a robust coding scheme that is transferable and can be applied to other rumors outside of the Syrian context in the future.

Computer science and neurobiology have an undeniable entanglement with one another in the 21st century with the advent of advanced artificial intelligence and attempts to understand the black-box model of a brain. However, most neurobiology research has revolved around understanding existing behaviors in large animals, with commonly researched organisms including zebrafish and mice. Introducing novel behaviors to animals in a controlled environment is undeniably attractive in modern neurobiology. With the use of mass data-collection in the online Aquarium lab environment, this study shows the potential to model and produce novel behaviors in Hydra vulgaris (freshwater anemones with simple nervous systems). We aim to create a high-throughput, data-oriented, model for gene-based behavior analysis and editing in neurobiology research. By creating a single controlled environment to perform all research jobs on mass-produced H. vulgaris strains with clockwork husbandry, extraneous factors affecting the organisms are significantly reduced. By taking data on the success of every modification of these Hydra, we can model the effects of specific factors on these animals and progress toward creating a model organism for neurobiology research akin to knockout mice in medicinal genetics. With these methods we have been able to produce a significantly more efficient process of gene editing in Hydra and with several different kinds of DNA integrants. Over the long term, the wide availability of the online Aquarium lab can yield the discovery of unknown trends in genetic knockouts and the discovery of novel behaviors through analysis of all data collected in this controlled environment. Here we highlight successes thus far and future developments necessary to make this system a viable model for gene-based neural engineering over this century.

Mobile and worn sensors have been increasingly used in order to gather data in long term behavioral studies, especially as the cost has decreased for such sensors. These kinds of studies provide increased information that can be used in order to understand behaviors and to elicit changes in them. Such studies have been conducted at CMU, MIT, and Dartmouth. UW is conducting a similar study that uses mobile phone sensors via a custom app, surveys via Qualtrics, and FitBit sensors in order to collect relevant behavioral data that may be related to academic performance in college. Because such studies rely on participants following the protocol and actively participating in the study, it is important to create and modify tools that allow for analysis of participant compliance, such as the rate of survey completion and amount of sensor data. These tools allow researchers to keep track of participants and work on collecting more high quality data for later analysis. There is little information about the design of compliance tools for similar studies. In this work, the focus is on modifying an existing dashboard from a similar study at CMU. Development will be focused on adding additional functionality based on the needs of the research staff. In particular, work will begin by developing filtering functionality for non-compliant participants and creating useful views for researchers in terms of contacting participants around possible broken sensors or missing survey data. Usability of the tool will be analyzed for future improvement. The hope is that this tool can be used to gather better quality data in this UW study and can shape future dashboards for similar experiments that gather behavioral information.

Deep learning has pushed the boundaries of natural language processing, and neural network architectures have achieved state of the art results in tasks such as automatic machine translation, question answering, and language modeling, among others. The backbone of many of these models is the recurrent neural network (RNN), which serves as a general-purpose module for encoding sequences of text as a vector of real numbers. RNNs are ubiquitous in recent natural language processing research; in fact, it is often difficult to find work that doesn't use them! Despite their popularity, RNNs are often treated as black box text encoders, and little is understood about why they are so effective and how to improve them. Motivated by synthetic data experiments and observations from real-world natural language tasks, we introduce several new variants of RNNs and measure their performance on several tasks. Improving RNNs would subsequently improve the quality of modern language technologies as a whole.

During the Space Race, NASA calculated lunar rocket trajectories by using Euler’s Method to estimate the differential equation obtained by numerically integrating Newton’s Laws of Motion. At the time, computing power was scarce, costly, and primitive. The objective of this project is to develop an interactive MATLAB application that leverages modern computing architecture to accurately and efficiently approximate lunar rocket trajectories based on multiple variables relating to a rocket's journey to the Moon using the Runge-Kutta method. The Runge-Kutta method is a more accurate and sophisticated version of Euler’s method as it uses a multi-stage algorithm that involves slope calculations at discrete time values. This project will bring valuable input to the feasibility of using such numerical methods in approximating lunar trajectories to the industry. The rocket emulated in the program will be based on NASA's Saturn V rocket. Our application will simulate three phases of travel to the moon: achieving Earth orbit, lunar transfer orbit, and lunar orbit. It will consider variables such as time, distance, velocity, acceleration, position, rocket thrust, exhaust velocity, mass flow rate, and aerodynamic drag. The application will allow users to adjust various parameters in real time and see results graphically.

The ultimate goal of robotics is to be able to interact with robots as one would with another person; hand in hand with this goal lies the ability for robots to understand natural language and carry out tasks in practical environments. Let us take for example, if a human were to command a robot to “pick up a can of coke inside a cardboard box on a table”. The robot should be able to extract relational information such as a can of coke is inside a cardboard box and a cardboard box is on a table. If “coke” is unknown, the robot should be able to ask clarifying questions to ascertain the qualitative aspects that define it. So, robots should be able to reason in the presence of uncertainties during object manipulation in the real world. We use a Fuzzy Markov Logic Network (Fuzzy-MLN) to learn and predict object classifications given input from multiple sensory modalities. Inputs from vision (RBG-D), haptics, and audio are used along with a prior knowledge base to identify objects in the robot’s environment and annotate various attributes which are added to the MLN. Audio input is processed for natural language which can provide information to contextualize the robot’s environment or give instructions to/answer queries from the robot in order to learn new relations, update Fuzzy-MLN weights, and improve annotator accuracy.