Despite being individually uncommon, rare diseases are a heterogeneous set of nearly 7,000 diseases estimated to affect 6-8% of the global population, making them collectively common. Rare diseases are infrequent and their determinants are usually unknown which negatively impact clinicians’ ability to make an accurate and timely diagnosis. With the inception of consortia such as the Undiagnosed Diseases Network (UDN) and ontologies such as the Human Phenotype Ontology (HPO) there is now a wealth of information on rare diseases. This rich information has enabled researchers to infer gene-disease relationships and clinicians to assign genetic diagnoses. However, rare diseases that are caused by unique or uncommon gene combinations remain unsolved. In this study, we address this problem by assuming that a disease can be described by a combination of clinical features or phenotypes and, thus, a combination of genes. By developing a pipeline to incorporate known gene-phenotype relationships from publicly available resources such as HPO and using various combinatorial search algorithms, we can prioritize the most likely sets of genes to result in a given set of phenotypes. Integration of our pipeline with individual patient data is expected to further reduce the search space and produce higher quality predictions. We anticipate the results from this study to advance the quality and accuracy of gene-disease relationship predictions for rare and undiagnosed diseases associated with one or more genes. Our pipeline could be used as a tool for the diagnosis and discovery of likely oligogenic, rare and undiagnosed diseases that are normally turned away from consortia like the UDN.

First-passage and last-passage algorithms are two diffusion Monte Carlo methods that we can use to obtain the charge density distribution on a conducting surface. Usually, we use first-passage algorithms to obtain the capacitance and the overall charge distribution of the arbitrary-shaped conductor. On the other hand, the last-passage algorithms calculate the charge density at a point on a conducting surface by initiating the random walk at that point. The last-passage algorithm utilizes the dipole Green’s function to expedite the diffusion process which starts from the point. Past results determine the initial sitting using the largest hemisphere around that point with a radius that is contained in the surface, which inefficiently starts the random walk if the point is near the edge of the surface. Here, we derive the dipole Green’s function to start the diffusion process from a point on the hemisphere off-centered from the original point using weighted sampling to further expedite the computing process.

Small-angle neutron scattering (SANS) is a powerful tool that provides access to molecular-level structure in a variety of material, such as soft matter, colloids, and biomolecules. Currently, researchers must follow a manual and inefficient procedure to reduce and analyze data due to inconsistent meta-data storage and varying reduction processes. Thus, our project is aimed to improve the automated reduction of SANS data and streamline data flow at the experiment site by the use of machine learning (ML) and data science techniques. We propose a machine learning classification algorithm, trained on a historical SANS dataset, to automatically sort raw data into six categories of scattering profiles and relevant background measurements required for reduction. The first stage of our research project uses a feature reduction model to create a subset of most significant and relevant meta-data in the historical scattering data set. After pre-processing, we then explore different ML models (e.g. neural networks) to sort the data in a comprehensive, accessible, and maintainable form that can be easily applied to sorting data directly from experiment sites. In addition, we incorporate a Postgres database to improve data management as well as an interactive front-end to enable user interaction. Currently, we are building meaningful and useful data processing and management methods by starting with a small SANS dataset from the Pozzo Research Group at the University of Washington. In the future, these methods will be expanded to larger historical SANS datasets to train our machine learning models. By developing this machine learning algorithm that can utilize experimental information and scattering patterns for auto-labeling of raw data and relevant meta-data, we will enable more efficient storage, reduction, and analysis of SANS data for researchers in the future.

Located in the Whidbey Basin of the northern reaches of Puget Sound, Possession Sound contains the Snohomish River estuary, encompassing a river system that is the third largest contributor of freshwater to the Puget Sound. Myself, and several of my fellow students have had the opportunity to collaborate with the University of Washington, the Washington State Department of Ecology, and a local environmental consultant -- Gravity Marine. Utilizing data collected by permanently moored Sea-Bird CTD probes, during research cruises, and by the United States Geological Survey (USGS), my research partner and I have created a model to facilitate an investigation of how both tides and the discharge of freshwater from the Snohomish River influence the water quality of Possession Sound. To better understand the complex patterns of the water entering Possession Sound from the Snohomish River, we analyzed the relationship between a continuous stream of water quality data (temperature, salinity and turbidity) from a Sea-Bird CTD probe at the mouth of the Snohomish River and continuous discharge data gathered by the USGS at a station 12 miles up the river. The distance between these two sites results in a delay between when river discharge data is recorded up river and when it influences the water quality at the river mouth. From the analysis of these two locations and with guidance from ourcollaborators as well as outside professionals, we used the statistical analysis language R to create a model that predicts the travel time of water from the USGS stream gage to Possession Sound. This model can be applied when considering the effect on the estuary of important factors from the river, such as nutrient loading; influxes of cold water, which promotes upwelling; and the river’s contribution of heavy metals and other pollutants.

This presentation seeks to summarize a solution to helping the visually-impaired navigate new areas. While previous solutions have been relatively successful, many lacked two key features that we hope our solution addresses: being affordable and allowing customization towards those with compounding disabilities. Our solution consists of two main parts: (1) a user-interface created for Fusion 360, a popular 3D-modeling application, that is built upon an existing framework detailed in Hofmann (2018) called PARTs (Parameterized Abstractions of Reusable Things), and (2) an optimization algorithm to generate maps that are tailored for its users. Through PARTs, we developed different variations of modular pieces of map (e.g., roads, buildings, and sidewalks), which increases ease of customization. After the user specifies personal information and preferences through the PARTs UI—such as the width of their finger, their physical limitations, their understanding of braille, and their desired map features—the optimization algorithm will select the best combination of features from the PARTs database for that specific user. At the end of the process, users have a model of a tactile map in Fusion 360 which can be printed out with commercially-available 3D-printers. With 3D-printers becoming more affordable, this solution is significantly less cost prohibitive than other means of generating tactile maps, which required an initial investment upwards of a thousand dollars. Through user studies, we also test how blind users interpret these maps, which helps us guide design improvements in the future. In this presentation, we discuss the efficacy of our solution by comparing it to previous works and detail our plans to improve the system by making the PARTs user-interface more accessible and incorporating user feedback about the map itself. 

In computer graphics, a voxel (volume element) is a point in a 3D world coordinate system (i.e. the coordinate system of a virtual world). In games like Toca Blocks or Minecraft, voxels are used to store the texture of a particular terrain. Sometimes, voxels next to each other have the same texture. When voxels of homogeneous textures form polygons, rendering systems will optimize memory storage by storing the polygons' vertices rather than every single voxel in the polygon. The process of choosing polygons that cover the voxels is known as meshing. We refer to these polygons as quads and the collection of quads as a mesh. Current methods for polygon meshing require too much data storage or require a drastic change in the mesh after a small change in the world coordinate system. We propose the Greedy Face Meshing (GFM) Algorithm, a linear time algorithm for meshing voxels into quads. We prove that our algorithm is within a constant factor of the optimal solution (in terms of number of quads) and can update in constant time for a single-voxel change in the world coordinate system. We also show how the GFM Algorithm can be implemented using the Segment Tree data structure. Rendering systems can use the GFM algorithm to mesh polygons since its storage is no worse than any existing algorithm and its updates take constant time.

Trigger-action platforms enable end-users to program their smart homes using simple conditional rules of the form: if condition then action. Although these rules are easy to program, subtleties in their interpretation can cause users to make errors that have consequences ranging from incorrect and undesired functionality to security and privacy violations. Based on prior work, we enumerate a set of nine error classes that users can make, and we empirically study the relationship between these classes and the interface design of eight commercially available trigger-action platforms. Particularly, we examine whether each interface prevents (e.g., via good design) or allows each class of error. Based on this analysis, we develop a framework to classify errors and extract insights that lay a foundation for the design of future trigger-action programming interfaces where certain classes of errors can be mitigated by technical means or by alerting the user to the possibility of an error. For instance, we identify that an analysis of a dataset of functionally-similar trigger-action rules could be used to predict whether certain types of error patterns are about to occur. We believe that this work is a first step towards trigger-action interface designs that significantly mitigate user error.

We introduce Project Sidewalk, a new web-based tool that enables online crowdworkers to remotely label pedestrian-related accessibility problems by virtually walking through city streets in Google Street View. To train, engage, and sustain users, we apply basic game design principles such as interactive onboarding, mission-based tasks, and progress dashboards. In an 18-month deployment study, 797 online users contributed 205,385 labels and audited 2,941 miles of Washington DC streets. We compare behavioral and labeling quality differences between paid crowdworkers and volunteers, investigate the effects of label type, label severity, and majority vote on accuracy, and analyze common labeling errors. To complement these findings, we report on an interview study with three key stakeholder groups (N=14) soliciting reactions to our tool and methods. Our findings demonstrate the potential of virtually auditing urban accessibility and highlight tradeoffs between scalability and quality compared to traditional approaches.

Developmental and Epileptic Encephalopathies (DEEs) are a group of severe epilepsy disorders in children and infants characterized by prominent EEG (electroencephalography) abnormalities that disrupt brain function leading to cognitive decline. Identifying genetic causes of DEE is a key step to help researchers develop and personalize medical treatments for affected patients. Approximately 5% of DEE cases are caused by a copy number variation (CNV), where a region of DNA involving a disease gene has been duplicated or deleted. Historically, this type of mutation has been difficult to detect using sequence data. To address this, I have written a multi-step algorithm that analyzes smMIP (single molecule molecular inversion probe) targeted DNA resequencing data for known DEE genes to identify CNVs in patients’ DNA that are potentially disease causing. I have run this algorithm on the large collection of smMIP data for 1158 DEE patients available in the Mefford Lab and identified several potential CNVs. Among these, three CNVs ranging in size from 250,000-2,790,000 base pairs, each involving a DEE gene - GNB1, GRIA2, and UHRF1BP1L - were validated by a second method, array CGH, the current gold standard for CNV validation. To date, the validation rate of high-confidence CNV candidates is 50% or higher. Currently I am expanding the algorithm’s functionality to include the ability to selectively search for single-exon CNVs, which are as small as 500 base pairs, are more challenging to detect, have largely been missed by all CNV detection methods, but could still be pathogenic. To do this, I am leveraging the power of intersecting duplicate smMIP datasets to improve the sensitivity of single-exon CNV detection. As any disruption of a pathogenic DEE gene could be disease causing, inclusion of these smaller CNVs will increase our ability to solve DEE cases and improve patient care.

The environment in which a molecule resides can drastically affect both its reactivity and its spectroscopic properties. These effects can be clearly seen in phenomena such as solvochromism or increased reactivity in a protein active site. The most accurate models that can capture these effects are purely quantum mechanical, but they also have a steep computational scaling cost that prohibits their use on large molecular systems. In order to circumvent this, less expensive models, such as molecular mechanics, can be used for the environment while maintaining the accurate model for the system of interest. In recent years, development of polarizable molecular mechanics has enhanced the accuracy of this technique for time-independent systems, but a gap still exists for time-dependent systems. In particular, the frequency dependence of polarizability has yet to be addressed. This research uses frequency dependent polarizabilities to parameterize a novel molecular mechanics scheme. Specifically, a machine learning model is trained to predict the best locations to place isotropic polarizabilities in order to recover the molecular, frequency-dependent polarizability.

Current artificial neural networks (ANNs) use an archaic view of neurons based on oversimplification of their biological computations. This has allowed optimized computation through GPUs, leading to the widespread adoption of ANNs in deep learning but losing important biological features. In this research we create a new model for artificial neural networks that incorporates more realistic aspects of biological neural networks such as stochastic vesicle release in neuronal synapses and dendritic computation. It has been shown that animals learn models of the environment when introduced to a new situation, but this type of learning is often not incorporated into reinforcement learning models in AI. The goal is to have the new, biologically realistic ANN learn models of environments in simulation frameworks like OpenAI Gym and AI2Thor so that given past frames/images and an action taken by the agent/player the network can predict how the environment will react over time. We compare the performance of this network with that of traditional ANNs (e.g. recurrent neural networks with long-short term memory) to demonstrate the capabilities of the new network. Our results have implications for recent efforts to move toward biologically inspired models of learning in the fields of artificial intelligence and computer vision in robotics. We expect the model learning algorithms we present to more efficiently learn an environment and select actions to achieve arbitrary goals within that environment. This is different than traditional reinforcement learning models, which aim to complete a single goal and can take a long time to train. The novelty of this work is the increased biological realism without the computational complexity of simulating real neurons, the temporal aspect in neural processing in addition to the spatial aspect, and prediction based on actions instead of pure video prediction.

Reinforcement Learning (RL) is a subcategory of machine learning, in which an agent (the decision maker) observes its environment and executes the best course of actions to maximize rewards. This is similar to teaching a pet to perform tricks using treats as positive reinforcement. Our research compares different RL methods on low-performance devices like a Raspberry Pi in real-time, real-world environments. RL has gained popularity recently with breakthroughs from DeepMind’s paper, Playing Atari with Deep Reinforcement Learning, where an agent learns to play Atari games from raw pixels and from DeepMind’s AlphaGo (DeepMind, https://deepmind.com/research/alphago) program that was the first computer program to beat a world champion Go player. RL projects like AlphaGo have utilized big data, powerful computing resources, and simulated environments that do not require real-time interaction to train the machine learning models. Our group compares the effectiveness of different RL methods on an accessible level of computing power on offline devices that an average consumer could acquire. The team constructs a physical environment for the robot to navigate, creates an OpenAI Gym environment that our agents will use to control the robot and get feedback from the environment. We train our agents using different RL methods to optimally navigate the environment and avoid collisions. We then compare the performance of the different methods in our physical real-time environment. Reinforcement Learning in small, offline devices could pave the way for a variety of devices that learn over time without being connected to a network. Imagine a small Mars rover that learns to navigate its environment efficiently over time.

Minimizing cognitive load is an integral part of human-centered design, where a more intuitive, easy to learn, and adaptive interface is desired. In this study, I implemented a program which was used to measure cognitive load using blink rates collected from pilots during takeoff and landing in a flight simulator. A GoPro camera was used to capture pilots’ blink rate. The program I wrote analyzed the GoPro video by first extracting the face of the pilot and then points on the face called facial landmarks. These facial landmarks are then rotated and scaled to standardize for analyses. The facial landmarks are fed into an Support Vector Machine (SVM), a type of machine learning model. To train the model I labeled around 60,000 frames that were sampled from the GoPro videos, as having pilot eyes “open” or “closed”. To predict blinks, the model is tasked with classifying each frame in a video as either eyes open or eyes closed. This data is then smoothed by a heuristic 20 frames. That is, if any two frames within a set of 20 frames is marked as closed, the entire series is marked as closed. The middle of each series of closed eye frames is then marked as the frame where the blink occurred. The output of the process is the video where each frame is marked as being the apex of a blink or not. I evaluated several machine learning models (random forests, gradient boosted decision trees, a multi-layered perceptron, and a support vector machine) and the support vector machine had the greatest precision and recall. This work is important because identifying peak workload tasks in aircraft cockpits contributes to the identification of areas that require better user interfaces and automation improvement to increase safety and efficiency of aircraft operations.

Medical imaging techniques such as X-ray, provide clinicians intensive information on the disease/condition of patients. However, clinicians have to look away from the subject to refer to medical images, thereby losing track of their work. Thus, clinicians usually study the images prior to surgery and limit reference time to images during surgery. Furthermore, unlike X-ray, novel imaging methods (such as optical ultrasound) are not taught in medical schools, so untrained clinicians face challenges in interpreting the images. These two limitations restrict the clinicians’ ability to fully utilize and adopt advanced medical imaging techniques. In this work, we explore the possibility of using Augmented and Virtual Reality (AR/VR) in the context of medical imaging. Prior applications of AR/VR technology in medicine have been limited to AR-aided training for medical students, telepresence for interaction, as well as remote therapy. We aim to use AR as a real-time diagnostic and therapeutic tool by augmenting the clinicians’ live view with various imaging modalities (such as X-ray, optical ultrasound, near-infrared). We hypothesize that providing these images in-context, and in some cases aligned with the subject, will improve the interpretation of images resulting in better guidance for diagnosis or surgery. To test this, we are creating an AR-based medical imaging/analysis application that uses techniques such as volumetric rendering and real-time image registration to augment the clinicians' view. Furthermore, clinicians can interact with the images by filtering, slicing, and reducing dimensionality, in order to better understand the images and thereby the underlying disease/condition.

The 2016 United States presidential election illustrated the pervasiveness of foreign information campaigns - state-backed operations aimed at influencing political discourse within a target nation state. Twitter has uncovered several such campaigns and has released datasets containing tweet activity of over 4,000 accounts identified as information operatives. The work presented here utilizes these datasets to study the interactions these campaigns have with science, scientific discourse, and the broader public understanding of science. A primary objective of these campaigns is to weaken a foreign adversary through information warfare. For example, one strategy might be to reduce vaccination rates. One way this is accomplished is by sparking debates in politically controversial scientific claims, such as the safety of vaccines, by referencing pseudo-scientific and predatory literature. To understand the extent of this strategy, we utilize document clustering to measure the concentration of engagement - total tweet frequency - around these controversial topics, comparing against a baseline random sampling of tweets from the Twitter real-time data stream. In addition to this, we perform exploratory analysis to detect references to both legitimate and illegitimate scientific articles, leveraging data from the Web of Science (WoS), a bibliographic database of more than 50 million papers across all disciplines since the mid-19th century. We expect the proportion of tweets mentioning controversial scientific topics to be higher in the Twitter data disclosure than in the random sampling of tweets. Of those tweets mentioning science related issues, we expect a small proportion of tweets to be direct references to scientific papers, which will likely be studies that are either controversial, disinformative, or directly related to foreign policy interests. Our goal for this study is to assess the strategic role that science disinformation plays in these campaigns and to contribute to the growing body of research aimed at detecting and preventing misinformation.

Base J is a glycosylated nucleobase found in Trypanosomatids, a family of single celled parasites causing Sleeping Sickness, Chagas Disease, and Leishmaniasis. This modified base is thought to play an important role in transcriptional termination for these organisms. Current methods of analyzing Base J rely solely on chromatin immunoprecipitation experiments, which provide low resolution information pertaining to Base J positions. While previous studies have shown that SMRT-seq interpulse duration (IPD) is associated with the position of Base J, we still lack methods to use this information to produce a genome wide, nucleotide-level map of Base J. Here we explore analytical approaches such as dimensionality reduction, machine learning, and signal processing to determine patterns of IPD and DNA enrichment which correspond to Base J across the entire Leishmania tarentolae genome. We show that many simplified approaches such as peak calling, and dimensionality reduction do not contain enough information to accurately classify Base J. We also utilize signal decomposition with Fourier transforms, machine learning clustering and regression methods to provide a more complex treatment of the data. Our findings are an important step in producing an algorithmic approach to identifying precise locations of Base J and can yield insight into transcriptional regulation in Trypanosomatids.

Labeling objects with DNA-based tags can provide a secure, difficult to fake identifier that is particularly useful for objects of high value or those that cannot be physically tagged. In this problem setup, a tag is a bit string, where each bit represents the presence or absence of a DNA strand containing a particular barcode. Our goal is to consistently and accurately identify the tag. These DNA barcodes were designed for use on a MinION nanopore sequencer, which outputs a time series signal corresponding to the DNA sequence. Ideally, each barcode should generate a dissimilar signal, which makes it easier to distinguish from other barcodes. We designed 96 barcodes that are signal orthogonal (i.e the signal output from the MinION was as dissimilar as possible), and detected them using signal processing algorithms. Using this system, I created an error analysis pipeline to ensure that we can identify tags both quickly and accurately. In order to optimize the time it takes to identify a tag, it was important to minimize the number of sequencing reads we needed to observe on the MinION, without sacrificing accuracy. I found that using a subset of the reads produced approximately the same error rate as a full run. Therefore, we can run the MinION for a shorter amount of time, and still identify tags at a similar error rate compared to a longer runtime.

Contextual word representations derived from large-scale neural language models are successful across a diverse set of natural language processing (NLP) tasks, suggesting that they encode useful and transferable features of language. To shed light on the linguistic knowledge they capture, we study the representations produced by several recent pretrained contextualizers (variants of ELMo, the OpenAI transformer LM, and BERT) with a suite of sixteen diverse probing tasks. We find that linear models trained on top of frozen contextual representations are competitive with state-of-the-art task-specific models in many cases, but fail on tasks requiring fine-grained linguistic knowledge (e.g., conjunct identification). To investigate the transferability of contextual word representations, we quantify differences in the transferability of individual layers within contextualizers, especially between recurrent neural networks (RNNs) and transformers. For instance, higher layers of RNNs are more task-specific, while transformer layers do not exhibit the same monotonic trend. In addition, to better understand what makes contextual word representations transferable, we compare language model pretraining with eleven supervised pretraining tasks. For any given task, pretraining on a closely related task yields better performance than language model pretraining (which is better on average) when the pretraining dataset is fixed. However, language model pretraining on more data gives the best results.

Causal inference is a large and long-standing field that attempts to answer a very fundamental question: does X cause Y? This question arises in many fields such as healthcare, genomics, biology, and econometrics. The complex systems analyzed in these fields often have many interacting components mathematically represented as nodes and edges (connectivity) of a network. Network inference is the study of the time-dependent behavior of these nodes to reverse-engineer the network connectivity. The theory of causal inference has many contending definitions of causality such as state-of-the-art techniques Granger Causality (GC) and Convergent Cross Mapping (CCM). These algorithms are computationally tractable and easy to use but require strong mathematical assumptions. We simulate networks of harmonic and Kuramoto oscillators and attempt to reconstruct their ground-truth network structure using observations of oscillator displacements over time. Our analysis investigates the performance of inference methods on Erdos-Renyi and scale-free random graphs. We show that the GC and CCM inference methods systematically fail to determine network structure by returning overly sparse or dense connectivity results. These findings challenge the applications of such top-down inference approaches to physical and biological systems. Using a few basic assumptions, we demonstrate how networked systems of coupled oscillators can be successfully reconstructed if perturbations of the system are allowed. We propose a Perturbation Causal Inference (PCI) algorithm that uses systematic perturbations and tracks how perturbation cascades spread through the network. Using changepoint detection, correlation, and windowed variance statistics, we predict causal relationships between nodes in the graph. Our analysis shows that PCI works at scale and efficiently returns high accuracy reconstructions of large networks with varying coupling strengths and connectivity structures. We conclude by proposing future applications of perturbation inference methods into neuroscience and make a connection with Hebbian learning rules.

Many people in today’s society do not devote enough time to enjoy the food they are eating, consuming their food with their attention devoted to phones or television screens, and in some instances even finishing their food without realizing how much they have eaten. This is especially prevalent in people who regularly eat alone, falsely believing that multitasking will increase their productivity. In this design research project, we investigate the problem of people’s lack of focus while eating with the goal of using our design insights to develop technological solutions to help people be intentional, mindful, and present when they eat. Our work is grounded in the premise that mindful eating provides health benefits and is a gentle and approachable way to introduce mindfulness into people’s lives. Our design process follows the established practice of exploration, divergent ideation, and then convergence, with an added reflection component. Specifically, we first explore the problem space by gaining a deep understanding of the relevant literature on mindfulness, multitasking, and the interactions between meals and technology, followed by conducting several user interviews about the problem. These interviews are with college students and other individuals who tend to eat alone in their homes. We then use divergent thinking to ideate many possible, diverse solutions, for example, a phone application, use of smart home technology, a standalone physical object. From there, we converge on a few candidate solutions and analyze the implications—positive and negative— of each. Design reflections consider not only the design outcomes but also our design process and key points in that design process. Our work contributes to a growing body of design research concerned with Slow Design, distraction, and mindfulness.

The Yerkes-Dodson law suggests that performance is optimal when physiological arousal is moderate. Prior research has shown that arousal induced by listening to music will affect task performance on a cognitive task. The goal of this study was to test whether different music tempos would affect arousal levels and performance on a cognitive task. We recruited 16 UW undergraduates and exposed them to different tempos of the same song while performing a math test. We tested their accuracy, completion time, and recorded their heart rate. Results suggest that accuracy was higher under moderate music tempo than slow but there was no difference between moderate and fast. Interestingly enough there was no difference in completion time or physiological arousal. Although we failed to establish an effect of arousal and completion time the fact that participants did better on accuracy when music tempo was moderate is in line with what we predicted. Music tempo could still be influential to performance and future research can be done to understand these effects by operationalizing performance differently. Future directions about ways to improve the study will be discussed especially whether music/music tempo affect people subjective arousal and physical arousal differently.

Have you ever had to motivate the team? What was the result? The English language Wikipedia is notable for its large number of articles. However, 288 other active language editions of Wikipedia have also developed through the intricate interactions of contributing editors. While the editor interactions in the English Wikipedia have been researched extensively, these other language editions remain understudied. To understand how editors currently come to consensus in article building in the French language, a team of researchers has leveraged an existing English framework that depicts how power and policies play a role in mass collaboration. Using this English language framework, we are using qualitative coding methods to build a unique model of the editor interactions on the French language Wikipedia. The results of this study will help contribute to a deeper understanding of how a framework in a different language edition of Wikipedia differs from the English. Our preliminary results show that policy plays a large role in justifying editor decisions for the edits they make on various articles. Furthermore, our research findings have expanded our knowledge of the issues surrounding replication of an English framework in a different language platform.

Cancer is increasingly a global health issue. Access to radiation therapy, the primary method of cancer treatment, is limited in low- and middle-income countries due to difficulty translating technologies in resource-limited settings. We are developing a new generation of cost effective IMRT (intensity modulated radiotherapy) devices for low- and middle-income countries. The current IMRT is performed using linear accelerators with multi-leaf collimators (MLC) which is very expensive and challenging to maintain in low- and middle-income countries. As a team, we are replacing the current technology with an alternative one using physical compensators which have lower commitments in terms of cost, overhead, and complexity. During my time working in the Imaging Lab, I have implemented all of the custom scripting protocols in a Python script (IronPython Interpreter) which is connected to the treatment planning system (RayStation - RaySearch Laboratories). Treatment planning systems are at the heart of radiation therapy systems and the key to improved patient outcomes. My script has enabled the creation of a compensator map and contour from the fluence data of the beamset. To test validity and correctness of the equations derived - I created a treatment plan based on the CT image for a head and neck cancer patient. The preliminary results of my script on this treatment plan indicate that compensators with ray lines divergent with the radiation source can be constructed for use in the treatment planning system. My contribution satisfies a key requirement of the project. A prototype of the compensator will be developed in commercial and partner sites in India, validating the script with a large number of patients.

This research is aimed at improving the functionality and use of multi-robot control through several different aspects. One important aspect of this project is collision avoidance between multiple ASEA Brown Boveri (ABB) industrial robots. The active collision avoidance software is an efficient tool that detects potential collisions between multiple robots working in collaboration and the surrounding cell. The user can specify a safety net around each robot, which notifies the user if one or more robots are on a collision course. Each robot’s position is graphically displayed to the screen with respect to the user-defined world coordinate system, complete with logs of tool center point positions and time stamps, so collisions can be reproduced. This software is intended to be used in conjunction with automated scheduling as an external safety checker. Another important aspect of this research is the end effector, an attachment that is secured onto the end of the ABB robot and used to drill a series of holes in a given surface. The end effector has gone through numerous design improvements through which it became smaller and more cost efficient. The latest design incorporates the addition of a raspberry pi, a small single-board computer, which serves as a failure test. The pi is programmed to generate random failures which notifies the user to take the robot out of the running program. This system will eventually be used to detect real robot failures and alert the user of the malfunction. The end effector also has a separate attachment which is used to hold a camera in order to incorporate machine vision into the multi-robot system. Adding machine vision helps robot position calibration, along with detecting foreign objects other than the moving robots in the robot cell. This addition also improves security and liability of the system.

Dynamical systems are ubiquitous in science and engineering. Inferring the mathematical laws that govern dynamical systems typically requires a 'scientist-in-the-loop' to guide the discovery process, via their expert knowledge and intuition about the system. Getting computers to perform this task automatically, without the guidance of a domain expert in the loop, is a grand challenge in the field of data science. A number of algorithms have been developed to infer such laws. One leading algorithm is Sparse Identification of Nonlinear Dynamics (SINDy), which applies simple linear regression coupled with sparsification to optimize a model over a large library of candidate functions. This algorithm is purely data-driven and makes no use of information that may be known previously about the dynamical system, such as symmetries and conservation laws. In this work, we develop a framework for incorporating and enforcing symmetries and conservation laws in SINDy so that the inferred models are consistent with prior domain knowledge. We analytically show how to propagate symmetries and conservation laws through the SINDy function library, and from this analytically derive linear constraints on the resultant linear regression. These constraints can be incorporated into the regression problem using options available in standard quadratic optimization packages. We implement this method and show that it provides improved accuracy and robustness on the task of inferring several canonical dynamical systems.