The following research posters were presented as part of the Computing[MTU] Showcase. The poster session took place Tuesday, April 5, and competition awards were announced April 6. See a list of the winners on the ICC news blog.
Faculty Posters
"ROBOT101: A reflection on (over) a century of robots in imagination and practice"
R.U.R., the play by Karel Čapek that brought the word ROBOT to the world, was first performed in 1921. That makes 2022 the 101st anniversary. In honor of this highly influential work and the ideas it has sparked, we are organizing a series of events at Michigan Tech that we’re calling ROBOT101.
The central event in this series will be a production of R.U.R. by the Tech Theatre Company in October 2022. We are organizing a variety of events in Fall 2022 (both before and after the performances) that will engage students, faculty, and community members with the issues raised by the idea of robot. These include (but are not limited to): developments of robots in reality and the imagination (literature, film, etc.); the status of work and workers; contexts of technological innovation; perspectives on otherness and inclusion; and evolving concepts of sentience. In addition, we want to provide insights into the capabilities and challenges of current robotics technology.
This is a timely opportunity for discussion and collaboration. We wish to draw from the diverse pool of talent at Michigan Tech as well as invited experts from outside the University.
Graduate Student Posters
"Assessing Cognitive Empathy Elements within the Context of Diagnostic AI Chatbots"
Empathy is an important element for any social relationship and it is also very important in patient-physician communication for ensuring the quality of care. There are many aspects and dimensions of empathy applicable in such communication. As Artificial Intelligence is being heavily deployed in healthcare, it is critical that there is a shared understanding between patients and the AI systems if patients are directly interacting with those systems. But many of the emotional aspects of empathy may not be achievable by AI systems at present and cognitive empathy is the one that can genuinely be implemented through artificial intelligence in healthcare. We need a better understanding of the elements of cognitive empathy and how these elements can be utilized effectively. In this research, the goal was to investigate whether empathy elements actually make a difference to improve user perception of AI empathy. We developed a scale "AI Cognitive Empathy Scale (AICES)" for that purpose and conducted a study where the experimental condition had both emotional and cognitive empathy elements together. The AICES scale demonstrated reasonable consistency, reliability, and validity, and overall, empathy elements improve the perceived empathy concern within diagnostic AI chatbots.
"Keep your hands on the wheel: the effect of driver engagement strategy on change detection, mind wandering, and gaze behavior"
Advanced driver-assist systems (ADAS) have revolutionized traditional driving by enabling drivers to relinquish operational control of the vehicle to automation for part of the total drive. These features only work under certain pre-defined conditions and require drivers to be attentive of their surroundings. While the features are engaged, there is an increased risk associated with drivers losing awareness of their environment. Popular manufacturers like Tesla requires drivers to have their hands-on-the-wheel while Cadillac’s ADAS requires drivers to keep their eyes-on-the road. We utilized a low-fidelity simulation and eye tracking to examine the effects of hands-on-the wheel and eyes-on-the road driver engagement strategies on change detection, mind wandering, and gaze behavior in a semi-autonomous driving task.
Coronary artery aneurysms (CAA) describe local dilatations of the coronary artery that exceed 1.5 the neighboring artery diameter. The prevalence of CAA ranges from 0.3 to 5%. The pathogenesis of CAA is not well understood however, several factors come into play such as certain vasculitic and connective tissue diseases such as the Kawasaki disease (KD) . CAAs are mostly not accompanied by symptoms and their finding is usually incidental.
Literature has shown that the blood flow stagnation and sluggish flow are correlated with thrombosis and atherosclerosis. Several patient-specific computational fluid dynamics (CFD) and experimental investigations to analyze the flow and risk level of CAA cases based on hemodynamic indices were performed . However, the effect of different aneurysm shape indices and resulting flow parameters on local hemodynamics in CAAs is not systematically investigated yet.
The goal of this study was to investigate the effect of CAA shape indices on local hemodynamics using the response surface method (RSM) through considering KD applications at rest (normal conditions) and during exercise. CFD was performed on idealized controlled geometries where different aspect ratios were considered. Regression models of time averaged wall shear stress (TAWSS), oscillatory shear index (OSI), and relative residence time (RRT) were developed. These models can be used as valuable tools to obtain a real-time assessment of CAA local hemodynamics and resulting risk stratification based on fusiform aneurysm shape indices.
In this study, a parametric CFD quantification of CAA was performed, and regression models of 2 hemodynamic metrics under normal at rest conditions and under exercise conditions were developed using the response surface method. The results show that TAWSS generally increased under exercise conditions regardless of the geometry. It is observed that the small aneurysms (low Dmax/Dneck) experience very high RRT values in elevated heart rates, which is neglected as a risk case in Z-score criteria. These findings highlight the importance of computational tools to achieve a general understanding of the hemodynamics in CAA.
"Pay "Attention" to Adverse Weather: Weather-aware Attention-based Object Detection"
Despite the recent advances of deep neural networks, object detection for adverse weather remains challenging due to the poor perception of some sensors in adverse weather. Hence, instead of relying on one single sensor, multimodal fusion has been one promising approach to provide redundant detection information based on multiple sensors. However, most existing multimodal fusion research is ineffective in adjusting the focus of different sensors under varying detection environments, such as dynamic adverse weather conditions. Moreover, it is critical to simultaneously observe local and global information under complex weather conditions, which has been neglected in most early or late-stage multimodal fusion works. In view of these, this paper proposes a Global-Local Attention (GLA) framework to adaptively fuse the multi-modality sensing streams, i.e., camera, gated camera, and lidar data, at two fusion stages. Specifically, GLA integrates an early-stage fusion via a Local Attention network and a late-stage adaptive fusion via a Global Attention network to deal with both local and global information, which automatically allocates higher weights to the modality with better detection features at the late-stage fusion to cope with the specific weather condition adaptively. Experimental results demonstrate the superior performance of the proposed GLA compared with state-of-the-art fusion approaches under various adverse weather conditions, such as light fog, dense fog, and snow.
Mobile computing devices have been used broadly to store, manage and process critical data. To protect confidentiality of stored data, major mobile operating systems provide full disk encryption, which relies on traditional encryption and requires keeping the decryption keys secret. This however, may not be true as an active attacker may coerce victims for decryption keys. Plausibly deniable encryption (PDE) can defend against such a coercive attacker by disguising the secret keys with decoy keys. Leveraging concept of PDE, various PDE systems have been built for mobile devices. However, a practical PDE system is still missing which can be compatible with mainstream mobile devices and, meanwhile, remains secure when facing a strong multi-snapshot adversary. This work fills this gap by designing the first mobile PDE system against the multi-snapshot adversaries.
OS-level malware may compromise OS and obtain root privilege. Detecting this type of strong malware is challenging, since it can easily hide its intrusion behaviors or even subvert the malware detection software (or malware detector). Having observed that flash storage devices have been used broadly by computing devices today, we propose to move the malware detector to the flash translation layer (FTL), located inside a flash storage device. Due to physical isolation provided by the FTL, the OS-level malware can neither subvert our malware detector, nor hide its access behaviors from our malware detector.
"Surface Text Interaction System using Microphones"
Text entry is of necessity to most systems requires human interaction. However, virtual reality (VR) and augmented reality (AR) devices lack efficient and ergonomic ways to perform text entry. In this work, we propose a design of a text-entry approach using contact microphones, which is easy to deploy on flat surfaces. Then we plan an experiment to collect data which will be later used to evaluate our system.
"Optimizing an Ambiguous Keyboard for Location-Independent Text Entry"
In many situations, it may be impractical or impossible to enter text by selecting precise locations on a physical or touchscreen keyboard. We present an ambiguous keyboard with four character groups that has potential applications for both eyes-free and single-switch text entry, as well as text entry in virtual or augmented reality. We develop a procedure for optimizing these character groupings based on a disambiguation algorithm that leverages a long-span language model. We produce both alphabetically-constrained and unconstrained layouts in an offline optimization experiment and compare them in a longitudinal user study. Our results did not show a significant difference between the constrained and unconstrained layouts after four hours of practice. Participants were able to achieve an average entry rate of 13.0 words per minute with a 1.8% character error rate using a single hand and with no visual feedback.
"Perceived Expertise of Diverse Construction Professionals"
Authors: Josiane Isingizwe, PhD Student and Ricardo Eiris, PhD, Assistant Professor, in the Department of Civil Environmental, Geospatial Engineering, Michigan Technological University
In multidisciplinary industries such as construction, a variety of experts are required to complete a project. Because of the large scope and complexity of most construction projects, not a single individual can be the source of all needed expertise to achieve project goals and objectives. Experts are often perceived by others as individuals with superior proficiency in a particular knowledge domain. People may receive a variety of experts’ advice and recommendations from different sources, but it has been realized that the use of an advice is solely dependent on assessment of its quality. Researchers have found that when expertise is properly assessed within project teams, professionals can collaborate more effectively, productivity is increased significantly, and it builds a good reputation of the organization with the clients. Although people tend to assess expertise adequately, a variety of contextual and behavioral parameters have been found to undermine perceived expertise in individuals even among peers. For example, expertise perception might fail when team members have distinct characteristics (e.g., race, gender, accent.), especially when individuals are not conscious of how these differences affect their perception. Having access to a set of professionals that can be used consistently to test the factors that influence perceived expertise is always a challenge. This project evaluates the actual practice used by members of the construction teams to assess each other’s expertise. This study uses 360-degree panorama and virtual human’s techniques to create a virtual field trip experience that mimic real-world construction jobsites. This study will recruit undergraduate students in civil and construction students at MTU. Virtual humans of different race and gender will lead a virtual field tour and perform some domain-related tasks. Participants will explore developed platforms using the Eye Tracker Head Mounted Displays. Participants racial implicit bias and learning will be assessing through surveys and the perceived level of their expertise will be assessed by participants through semi-interviews. The purpose of the study is to explore how differences and similarities of construction teams influence the perceived expertise.
"AudiWare: Finding the presence of malware in audio files"
Secure communication is a highly demandable feature for many fields. A user does not want to expose his data to the world except the receiver in this process. Data encryption is a well-known approach for efficiently hiding data by transferring it into another form. Researchers have developed several well-established encryption algorithms uncompromised in feasible time with state-of-the-art research in this field. Another approach to data hiding is to embed the data in another cover file so that no one is concerned. Currently, the second approach is getting popular among the attacker to hide malware and pass it to any system without a user's concern. In this paper, we hide malware in the audio file and detect its presence.
Virtual engineering relies heavily on augmented reality. They serve as the foundation for functional virtual prototyping, which allows engineers to examine the design, form, and functional behaviour of future goods in a virtual environment that is both immersive and interactive. The communication in product design and development is substantially improved when these technologies are used: It assists in the early detection and avoidance of design flaws, reduces the number of physical prototypes, and saves time and money for businesses. In many industrial applications, virtual reality and augmented reality are seen as valuable tools for improving and speeding up product and process development. However, many requirements remain unmet, leaving significant room for ongoing development and improvement of VR/AR-based tools and procedures.
"Habitat Heterogeneity Promotes Linked C and N Cycling in Streams: A Predictive Modeling Approach"
Heterogeneous habitats can have distinct physical and chemical characteristics, which may provide ideal microclimates for different biogeochemical processes. Additionally, products of reactions occurring in one habitat may provide limiting reactants for different processes in adjacent habitats. We ask: do streams with higher habitat diversity promote more C and N cycle processes, where the relative activity of denitrification, N fixation, gross primary production (GPP), and ecosystem respiration (ER) are interrelated? We measured denitrification (acetylene block) and N fixation (acetylene reduction) in 14 National Ecological Observatory Network (NEON) streams, scaling from chamber measurements to whole-stream rates using substrate cover. Preliminary results show on average N fixation (μ = 0.06, SD = 0.12 mg-N m -1 h-1) replaces about 2.5% of reactive N removed by denitrification (u = 1.97, SD = 2.87) on the reach scale. GPP and ER in the same streams was modeled from NEON data using a two-station Bayesian approach, and heterogeneity metrics (Simpson’s information index, evenness) were quantified from NEON habitat data. We will use predictive modeling (linear and nonlinear regression, trees and rule-based models) to determine if streambed heterogeneity, or other factors, predict the relative activity of denitrification, N fixation, GPP, and ER in streams. If the same environmental factor is highly important in determining two or more cycling processes, this suggests the factor strengthens links between the processes. Understanding why C and N cycling may be interrelated in streams provides a framework to investigate if denitrification and N2 fixation covary with GPP and ER across ecoregions.
"Adding elaboration to word game: Examining the effect of generated hint on learning"
Teachers have implemented some sort of word games in a variety of science-related classroom settings in order to have students learn technical vocabulary and to improve their scientific thinking. However, the majority of the word-game studies show positive findings only for the improvement in retaining memory of learned vocabulary. The literature suggests that in order to go beyond remembering word-definition pairs and to achieve higher-level of scientific thinking, learners need to elaborate on materials they are engaged in.
In this work, we describe a series of experiments that evaluate the effect of crossword coupled with an elaboration task. After learners learn a set of technical words from solving a crossword puzzle, they are to generate a new set of crossword hints to be associated with those words. The mentioned task supposedly enhances learning and retention of learned words by having learners summarize and then elaborate on the learned material. Total of 64 undergraduate students were recruited as participants in the aforementioned within-subject-design experiments. Results indicated that adding the elaboration task to the crossword task improve participant's learning and remembering scientific vocabulary. However, when control for time-on-task, the effectiveness of crossword coupled with elaboration task is not significantly higher than crossword alone. We also found that a factor that contributes to the quality of generated hints and, in turn, learning outcome is time and effort learner spends in the hint generation task. This work suggests some new best practices in an implementation of word game in learning technical vocabulary which require important but minor changes that have a broad impact on learning outcome.
"Examining counterfactual sensemaking of social judgments to improve algorithmic decisions"
Developers continually readjust their framing of the problem and potential solutions throughout the design process (Hoffman, Roessler, & Moon, 2004; Costanza-Chock, 2020). When encountering potential cybersecurity threats, users scrutinize the evidence and adjust their theories about its trustworthiness. Both involve sensemaking (Weick et al., 2005; Klein et al., 2007). This research explores how people will question their frames, assuming that questioning one’s frame is necessary in order to change it. In this study, we examined the role of different factors (e.g., mutability of information, ambiguity, ease of explaining) on alternative outcome likelihood judgments.
"Learning from Social Media: Using Posts to Develop Effective Training Programs on AI/ML Systems"
Decisions made by Artificial Intelligence/Machine Learning (AI/ML) systems affect our daily lives. Therefore, it's important to be able to predict, and even know whether, or when these systems might make a mistake.
Traditional training approaches show learners of these cognitively challenging systems examples of correct and incorrect predictions. However, these learners are not much more proficient in forecasting predictions when compared to no training at all. A possible approach to better train users in these complex systems is Cognitive Tutorials for AI (CTAI; Mueller, Tan, Linja et al., 2021), which is an experiential method used to teach conditions under which the AI/ML system will succeed or fail.
One specific CTAI technique that was proposed involved teaching simple rules that could be used to predict performance; this was referred to as Rule Learning. This technique aims to identify rules that can help the user learn when the AI/ML system succeeds, fails, the system’s boundary conditions, and what types of differences change the output of the AI system. To evaluate this method, we ran a series of experiments in which we compared different rule learning approaches to find the most effective way to train users on these AI/ML systems. Using the MNIST data set, this includes showing positive and negative examples in comparison to providing explicit descriptions of rules that can be used to predict the system’s output. Results suggest that although examples help people learn the rules (especially examples of errors), tutorials that provided explicit rule learning and provided direct example-based practice with feedback led people to best predict correct and incorrect classifications of an AI/ML system. The next step is developing these tutorials for image classifiers and autonomous driving systems.
"Active learning with binary feedback on multiclass problems"
An active learning approach is often used for multiclass classification problems, where predictions are made on new data and a human user is used to determine if the predictions are correct. Typical approaches may ask a human to select the correct class if the prediction is incorrect. This work attempts to use a binary feedback on the predicted classes to save time and allow maximal use of a negative prediction on a partly trained model.
"Connected Vehicle Field Study: Outcomes and Challenges"
Poor driver decision-making continues to be a challenge at Highway-Rail Grade Crossings (HRGC). One way to improve safety has been to introduce a new, in-vehicle warning system that communicates with the external HRGC warning systems. The system gives drivers different rail-crossing-related warnings (e.g., approaching crossing, train presence) depending on the vehicle location. In a rare field study, 15 experienced drivers drove a connected vehicle (Chevy Volt) and used the warning system on a 12-mile loop, then completed a semi-structured interview and usability survey. Results from the post-drive survey and interview are reported and provide a template for future usability assessments for field studies involving new technologies.
"A Qualitative Data Coding Tool: From Excel Sheet to Simple Web Based Platform"
The exponential growth of data in many research fields means that revolutionary measures are needed for data analysis. In the field of Human Factors, researchers are already well accustomed to different usability questionnaires, workload measures, and other quantitative techniques to give conclusions to their studies. With these quantitative techniques, researchers are also using different qualitative techniques like semi-structured interviews, case studies, etc. to get more in-depth insight into their studies. Qualitative analysis often requires close reading of text, reflecting on data, and writing down interpretations (Hsieh & Shannon, 2005) that may induce cognitive workload. So proper software is needed to make the process smoother. This poster will present such a platform for doing the analysis.
"A novel sequential method to train physics informed neural networks for Allen Cahn and Cahn Hilliard equations"
A physics-informed neural network (PINN) incorporates the physics of a system by satisfying its boundary value problem through a neural network's loss function. Recent studies have shown that the PINN approach can be used to approximate the map between the solution of a partial differential equation (PDE) and its spatio-temporal coordinates. However, we have observed that the PINN method is significantly inaccurate for strongly non-linear and higher-order time-varying partial differential equations such as Allen Cahn and Cahn Hilliard equations. Therefore, to overcome this problem, a novel PINN scheme is proposed that solves the PDE sequentially over successive time segments using a single neural network. The key idea in the new proposed scheme is that the same neural network is re-trained for solving the PDE over successive time segments while satisfying the already obtained solution for all previous time segments. Thus it is named as backward compatible PINN (bc-PINN). We illustrate the advantages of bc-PINN, by solving the Cahn Hilliard and Allen Cahn equations. Furthermore, we have introduced two new techniques to improve the proposed bc-PINN scheme. In the first technique, we have taken advantage of the initial condition of a time--segment to guide the neural network map closer to the true map over that segment. In the second technique, we have implemented a transfer learning approach to preserve the solution features learned while training the previous segment. We have demonstrated that these two techniques improve the accuracy and efficiency of the bc-PINN scheme significantly. The convergence has also been improved by using a phase space representation for higher-order PDEs.
"Programming by Voice"
For programmers with motor impairments, programming by voice can be a promising alternative to typing. In interviews with motor-impaired programmers, we found most would prefer speaking code in a natural manner rather than adhering to a strict grammar as current systems require. To see how programmers speak code, we had programmers dictate a line of code that was either highlighted or missing. We found they spoke highlighted lines faster than missing lines. They often skipped punctuation in both conditions. We found two commercial speech recognizers had a high error rate on our spoken code collection. Adapting the recognizer's language model on our spoken code transcripts cut errors in half.
"Machine learning based prediction of the electronic structure of quasi-one-dimensional materials under strain"
We present a machine learning based model that can predict the electronic structure of quasi–one–dimensional materials while they are subjected to deformation modes such as torsion and extension/compression. The technique described here applies to important classes of materials systems such as nanotubes, nanoribbons, nanowires, miscellaneous chiral structures and nano–assemblies, for all of which, tuning the interplay of mechanical deformations and electronic fields— i.e., strain engineering — is an active area of investigation in the literature. Our model incorporates global structural symmetries and atomic relaxation effects, benefits from the use of helical coordinates to specify the electronic fields, and makes use of a specialized data generation process that solves the symmetry-adapted equations of Kohn-Sham Density Functional Theory in these coordinates.
Using armchair single wall carbon nanotubes as a prototypical example, we demonstrate the use of the model to predict the fields associated with the ground state electron density and the nuclear pseudocharges, when three parameters — namely, the radius of the nanotube, its axial stretch, and the twist per unit length — are specified as inputs. Other electronic properties of interest, including the ground state electronic free energy, can be evaluated from these predicted fields with low-overhead post-processing, typically to chemical accuracy. Additionally, we show how the nuclear coordinates can be reliably determined from the predicted pseudocharge field using a clustering based technique. Remarkably, only about 120 data points are found to be enough to predict the three dimensional electronic fields accurately, which we ascribe to the constraints imposed by symmetry in the problem setup, the use of low-discrepancy sequences for sampling, and efficient representation of the intrinsic low-dimensional features of the electronic fields. We comment on the interpretability of our machine learning model and anticipate that our framework will find utility in the automated discovery of low–dimensional materials, as well as the multi-scale modeling of such systems.
"Teaching with Modeling and Simulations"
We have been working on different interactive simulation tools for educational use and this year we are working to design a tool that would allow teachers to design their own virtual models and simulations. Using models and interactive simulations helps develop pattern recognition and abstraction skills necessary in computational thinking. Academics like Jeannette Wing, David Weintrop, and Barbara Sabitzer have researched and wrote about the necessity of computational thinking being taught and used in schools.
This poster will have information about past simulation tools that were made and what they aimed to teach. It will also contain information on why modeling and simulations are beneficial teaching tools for students of all ages and how they can be used. Lastly, this poster will explain the work and goal of a current project to design a virtual tool that will give a teacher full control of a simulation design, how involved the students are with building visual models and concept maps, and how students can interact with the simulation.
"O-GlcNAcylation (O-GlcNAc) Site Prediction Using Deep Learning Methods"
Protein post-translational modifications (PTMs) are the addition of various functional groups to amino acid residues following protein biosynthesis. The process of converting DNA into RNA is called Transcription where information coded in the sequence of base pairs in DNA is passed to molecules of RNA. After transcription, the translation process takes place where RNA is further converted into protein sequences in the form of amino acids. Protein post-translational modifications (PTMs) are modifications in chemical composition that have a key role in functional proteomics. The study of proteins and their PTMs’ is important while dealing with heart disease, cancer, neurodegenerative diseases, and diabetes. O-GlcNAcylation (O-GlcNAc) is a kind of post-translational modification found on serine and threonine residues defined by a β-glycosidic bond between the side-chain hydroxyl and N-acetylglucosamine. The prediction of an O-GlcNAc site in a given protein sequence is important because abnormal O-GlcNAcylation has the potential of causing cancer and various neurodegenerative diseases. In this work, different deep learning based methods are employed in order to build a robust predictive model to predict the occurrence of abnormal O-GlcNAcylation in a given protein sequence.
"Improving Interdisciplinary Ideation with the Facilitation of Future Failures"
One aspect of software development is anticipating potential risks, which can be challenging during the development of novel sociotechnical systems with various human and physical components. Our interdisciplinary team, involving researchers from Michigan Tech’s Computer Science and Cognitive and Learning Sciences departments, is developing an innovative software system and device with the goal of empowering people to receive technology help in a remote setting. We used the premortem method early in our system design process to identify and mitigate possible future risks in this previously unexplored space. The premortem method centralizes failure across a range of system uses to facilitate collaboration. Our team brainstormed failure scenarios and ways to eliminate, mitigate, or monitor the risks of those failures. We have found the premortem method valuable in enriching interdisciplinary team communication to recognize and mitigate previously unanticipated risks.
"Medical Image Segmentation Using U-Net Based State-of-the-art Deep Learning Techniques"
Recent advances in Deep Learning (DL) have been contributing to diverse modalities of health care. Biomedical image segmentation is an important piece in quantitative analysis and clinical diagnosis. U-net is one of the core DL architectures, mainly developed for biomedical image segmentation, and has been effectively utilized for many image modalities, such as CT scans, MRI, X-rays, and microscopy. The singular modular/structural design of U-net allows it to be optimized by the integration of various novel techniques to produce state-of-the-art deterministic and predictive results. This research explores recent developments in U-Net based DL architecture. A broad-spectrum comparison of the results and performance of these models will be presented by using standard evaluation metrics on popular benchmarking datasets.
"Categorization of Martian Landslides from Satellite Imagery Using Vision Transformer"
In this study, we use satellite images from the Mars Reconnaissance Orbiter (MRO) Context Camera (CTX) imagery of Vallis Marinaries region in Mars to locate landslides. We have created a dataset that has the images of rock avalanches, debris flows, and slumps. We have implemented a vision transformer based classification algorithm on the ataset. Vision transformer adopts the mechanism of self-attention and maps a sequence of image patches to the semantic label, so that the model can differentiate all >the three landslides from each other. Vision transformer model is achieving an accuracy of 79.5% for classification in our study.
People are good at accurately judging distances in the real world. Virtual reality (VR) systems attempt to accurately recreate environments in a computer simulation. However, many studies have found that distances are underestimated in VR. Distance judgments are important for numerous applications including training, entertainment, prototyping, and education. If people misperceive distances, they might be trained incorrectly or be unable to perform tasks as well as they could in the real world. "Direct blind walking" is the most common distance judgment technique used by researchers. This technique asks participants to view a target and try to walk to it while blindfolded. Since participants may be uneasy walking blindfolded, experimenters traditionally practice blind walking prior to the experiment. Little research has examined how this pre-experiment blind walking might impact subsequent distance judgments. We varied the amount of pre-experiment blind walking (very little walking vs. approximately four minutes of walking) and found that more pre-experiment walking led to less distance underestimation. Since our study found that pre-experiment walking can influence experiments, it is important that papers report the details of their pre-experiment procedures. Although there are many papers examining distance judgments, few report these details. We propose further studies to quantify how different amounts of pre-experiment walking might influence distance judgments in a VR environment. We hope that this work helps inform pre-experiment procedures in future studies and provides insights into how people make distance judgments.
"Explanation and Use of Uncertainty Quantified by Bayesian Neural Network Classifiers for Breast Histopathology Images"
Despite the promise of Convolutional neural network (CNN) based classification models for histopathological images, it is infeasible to quantify its uncertainties. Moreover, CNNs may suffer from overfitting when the data is biased. We show that Bayesian--CNN can overcome these limitations by regularizing automatically and by quantifying the uncertainty. We have developed a novel technique to utilize the uncertainties provided by the Bayesian--CNN that significantly improves the performance on a large fraction of the test data (about 6% improvement in accuracy on 77% of test data). Further, we provide a novel explanation for the uncertainty by projecting the data into a low dimensional space through a nonlinear dimensionality reduction technique. This dimensionality reduction enables interpretation of the test data through visualization and reveals the structure of the data in a low dimensional feature space. We show that the Bayesian-CNN can perform much better than the state-of-the-art transfer learning CNN (TL-CNN) by reducing the false negative and false positive by 11% and 7.7% respectively for the present data set. It achieves this performance with only 1.86 million parameters as compared to 134.33 million for TL-CNN. Besides, we modify the Bayesian--CNN by introducing a stochastic adaptive activation function. The modified Bayesian--CNN performs slightly better than Bayesian--CNN on all performance metrics and significantly reduces the number of false negatives and false positives (3% reduction for both). We also show that these results are statistically significant by performing McNemar's statistical significance test. This work shows the advantages of Bayesian-CNN against the state-of-the-art, explains and utilizes the uncertainties for histopathological images. It should find applications in various medical image classifications.
Objective: This study aims to evaluate the impact of catheter ablation for atrial fibrillation (AF) on left atrial (LA) flow dynamics and geometrical changes.
Methods: This exploratory study included computational flow simulations from 10 patients who underwent catheter ablation for AF. Complete cardiac cycle dataset was simulated before and after ablation using computational fluid dynamics. The study main endpoints were the changes in LA volume, LA velocity, LA wall shear stress (WSS), circulation (Γ), vorticity, pulmonary vein (PV) ostia area, and LA vortices before and after ablation.
Results: There was an average decrease in LA volume (11.58±15.17%), and PV ostia area (16.6±21.41%) after ablation. A non-uniform trend of velocity and WSS changes were observed after ablation. Compared with pre-ablation, 4 patients exhibited lower velocities, WSS distributions and a decreased Γ (-21.4±10.6%) while 6 developed higher velocities and WSS distributions. These geometrical changes dictated different flow mixing in the LA and distinct vortex patterns, characterized by different spinning velocities, vorticities, and rotational directions. Regions with q-criterion>0 were found to be dominant in the LA indicating prevalent rotational vortex structures.
Conclusion: Catheter ablation for AF induced different geometrical changes on the LA and the PVs therefore influencing flow mixing and vortex patterns in the LA, in addition to overall velocity and WSS distribution. Further exploration of the impact of catheter ablation on intracardiac flow dynamics is warranted to discern patterns that may correlate with clinical outcomes.
Tremendous impact of the Coronavirus Disease 2019 (COVID-19) pandemic is unprecedented for human health and many aspects of our daily lives. A critical step in the fight against COVID-19 is effective diagnosing of patients, and the popular approach is to utilize chest x-ray images for decision making and/or based on patients' symptoms. In this paper, we explore machine learning models, convolutional neural network-based architectures, for automatically detecting COVID-19 from X-ray images as well as clinical symptoms. The datasets used in this study are obtained from Kaggle databases and the Israeli Ministry of Health publicly released data. From our experimental results, we found that the support vector machine classifier performed best for predicting COVID-19 from symptoms, and the feature named known contact with an infected individual is found to be the most important one. Clinical symptoms such as sore throat and headache showed some degree of significance for contributing model’s accuracy. The proposed 2D CNN architecture is found to be the most accurate for predicting COVID-19 from chest X-ray images. The outcome of this study will be a good reference for the medical research community to accelerate the development of practical AI solutions for COVID-19 detection and treatment.
"Interpretable Machine Learning Model for the Deformation of Multiwalled Carbon Nanotubes Under Torsion and Bending"
"Comparing a Pixhawk Quadcopter to a GoPiGo with Regards to Primary and Secondary Education"
The rise of drones has brought the ability to bring hands-on STEM experience to students through the implementation of drones into their curriculum. This poster looks at the benefits and drawbacks of using a Raspberry Pi controlled Pixhawk quadcopter as compared to a Raspberry GoPiGo. The quadcopter with a Pixhawk flight controller has many sensors such as a camera, GPS, compass, etc. and is able to fly due to having four propellers. The GoPiGo is a ground robot on four wheels with a distance sensor controlled using only a Raspberry Pi and electronics. We look at the pricing, assembly, usability for new and experienced learners, implementation into the curriculum, and available features while taking into account an institution's resources.
Undergraduate Student Posters
"Creativity and Stem Education: Does video game play improve creative thinking?"
Many high schoolers have played 10,000+ hours of video games by the time they start college (McGonigal, 2011). Does what they learn help STEM problem-solving skills? Participants were 159 students (70% gamers) who were randomly assigned to one of three game conditions: control (no game-watch a video), control game (Snood), or a perspective-shifting game (Roller Coaster Tycoon). Following this task, participants completed a standard creativity measure (alt-use task, How many different ways can you use a shoe?). A 2 x 3 ANOVA on the creativity measures revealed that while there were no differences in the number or variety of ideas, there was a statistically significant difference in the originality of ideas. Overall, people playing the perspective- shifting game generated more original ideas (M=8.19) than those in the control game (M=5.63) or video control condition (M=5.4). These results extend prior work and suggest that not all game play warms up creative thinking, but that the type of game play matters. Implications for 21st -century education are discussed.
"Supply Chain Simulator"
My team and I designed a small game to demonstrate the supply chain of lumber in the UP. Our poster will show how a computer game is designed from start to finish. We will show sketches of initial UI design and gameplay. We will also show how we researched the local supply chain of lumber distribution, environmental impacts, and cost.
"Image Steganography: Unknown Information"
Modern Image Steganography is incredibly powerful in hiding large amounts of information, discreetly into image. This process does not increase the file size of the image, rather it distorts the image in discrete ways such that to the human eye and many programs are unable to tell that information has been hidden. This information can then later be extracted, and used by another. This means that anything that can be converted to bits can be transmitted through images covertly. As such, because malware can be transmitted this way, it is incredibly important to be able to determine if an image contains hidden information and if that information is malicious.
"Assessing the Effectiveness of the XAI Discovery Platform and Visual Explanations on User Understanding of AI Systems"
Artificial Intelligence (AI) and Machine Learning (ML) are playing an increasingly important role in our lives. Although it has the potential to enhance and improve our lives, there is a risk that it will also reinforce existing inequalities, develop biased algorithms, and prevent access to resources in unjust ways. To improve transparency, fairness, and justice in how these systems are used, the field of eXplainable AI (XAI) has emerged, which aims to explain AI systems to various stakeholders. One subfield of XAI is the implementation of visual explanations; in this project, the effectiveness of visual explanations will be investigated in relation to the MNIST dataset and the use of the XAI discovery platform, which was designed to allow users to explore an AI system and construct explanations regarding the system. The eventual implementation will include a user test on the system to determine if visual explanations are useful to users.
"Universal Sensor Description Schema: An extensible metalanguage to support heterogenous, evolving sensor data"
Collecting and processing underwater sensor data is a critical need for U.S. Navy operations, but the heterogeneity of sensor data representations presents a challenge for complete and accurate use of these data. The Universal Sensor Description Schema (USDS) project seeks to design, evaluate, and deploy a unified, extensible metalanguage for supporting legacy and future sensor data across multiple programming languages and environments. Michigan Tech is collaborating with Applied Research in Acoustics LLC to develop a robust programming environment for development of data-intensive applications.
With the desire to process sensor data across multiple languages comes the need for a library of mathematical operations within the USDS language. The implementation of such operations will allow for unified data collection and processing within USDS, with the ability to efficiently deploy to multiple programming languages and environments. Michigan Tech Physics student Anthony Palmer is leading the effort to implement and validate this library.
In order for the use of USDS to be a positive experience for software engineers, the USDS team has developed a command line utility tool, consisting of automated processes that compile USDS source code and provide unit testing and USDS validation. The tool acts as a compiler for the USDS language by employing the internal programs. Michigan Tech Computer Science student Elijah Cobb is leading the effort to develop this utility tool.
"Results of typing on three different angled virtual keyboards using the HoloLens Version 2, a Mixed Reality Device"
This project signifies an important breakthrough in virtual typing technologies and how they may revolutionize typing in the future. The experiment we ran uses the Microsoft HoloLens Version 2 mixed reality device, which mix reality with tangible virtual items that can be manipulated. This study aims to explore a comfortable and efficient way to type on virtual keyboards in three extremes. The investigation we launched laid in three main typing conditions: a horizontal keyboard placed on a table, a vertical keyboard placed on a wall, and a midair keyboard placed in space. Each of our participants experience all three conditions in a predetermined order by their participant number. Our participants begin by typing a calibration word which sets an appropriate height and depth for the horizontal and vertical conditions; it’s then followed by two practice sentences and twelve testable sentences for each condition. The test is run in a quiet room that has a single table pushed against a wall with a single QR code placed upon it. The QR code will be used in all three conditions as a spatial anchor for our keyboards to attach to. The results from our several pilot runs show an affinity for typing on the midair floating keyboard, followed with the vertical keyboard, and ending with the horizontal keyboard. This study is ongoing and has some evidence to support the most comfortable and efficient keyboard, but further research is necessary to progress this field and find the best keyboard that provides the user with the most comfort and efficiency.
"Illuminated Devices: A Sociotechnical System to Broaden Access to Digital Assistance"
Continuous instruction and mentoring is an essential but underemphasized requirement for narrowing the Digital Divide. Digital newcomers and other learners gain confidence and competence from personal, situated, interactive sessions with human tutors. Libraries and other community centers can provide means of access to this learning, but not all those in need have regular access to the physical locations of these institutions. Our goal is to reach learners in a ubiquitous fashion, whenever and wherever they need help. Our project Illuminated Devices seeks to make the personal, interactive nature of a community-based tutoring program available anywhere, by connecting learners to tutors directly through common digital devices. The Illuminated system is a sociotechnical framework, blending digital technology and human interaction. It comprises a portal application that allows an immediate face-to-face connection between the user and tutor, and a sociotechnical network of tutors and learning resources.
"Shakudo: Learning the Language of Logic in a Scaffolded, Interactive Environment"
Using the formal language of logic is a key mathematical skill for students in computer science and other disciplines. As they learn this language, students need feedback on the syntactic and semantic correctness of the expressions they construct — the kind of feedback they get when learning programming languages. To address this need, we present Shakudo: an alternative, graphical programming language interface to the Alloy modeling language, along the lines of block-based introductory languages like Scratch. Through the interface, students using Shakudo generate logical expressions and then check the accuracy of these expressions through Alloy’s analysis engine. Shakudo provides scaffolding for the students that allows them to focus on exercises appropriate to their learning level.