GHC hosts a mind-expanding poster session for students, faculty members, and industry professionals. We are honored to be joined by these poster presenters, some of whom are also competing in the distinguished ACM Student Research Competition. Learn about the innovative technology the next generation is dreaming up.
|Poster Session 1
Wednesday, October 2
12 — 2:30 p.m.
|Poster Session 2
Wednesday, October 2
3 — 5:30 p.m.
|Poster Session 3
Thursday, October 3
10 a.m. — 12:30 p.m.
|Poster Session 4
Thursday, October 3
2 — 4:30 p.m.
|Poster Session 5
Friday, October 4
10 a.m. — 12:30 p.m.
Poster Session 1
Wednesday, October 2, 12 — 2:30 p.m.
Assessing the effects of gamification on computer science students: A gender study on performance
Presenter: Leila Zahedi, Florida International University
Despite recent improvements, we still see women’s underrepresentation in computing fields. By presenting challenging courses in a way that makes learning more enjoyable, students will be more likely to consider computing careers later in life. This quantitative study assesses if gamiﬁcation-using game elements-can improve students’ performance in a gamiﬁed online learning environment.
Broadening Participation: Inclusion of Female Computing Students in Competitive Programming
Presenter: Tamanna Motahar, North South University and University of Massachusetts - Amherst
This study describes an initiative for the inclusiveness of the female undergrad CS students into competitive programming in a developing country through “female only” environment. Many of the female CS students face visible and invisible barriers on their paths to comply with their computing career. This initiative was designed to focus on how they can overcome those barriers of participation.
Computer Science Mentors & Undergraduate Women Share Strategies for Academic Success & Persistence
Presenter: Breauna Spencer, University of California, and University of California, Irvine (UCI)
The study examined how faculty and graduate student mentors including undergraduate women enrolled in CS collectively articulated of the barriers and challenges women encounter in CS including ways to ameliorate these problems. Findings suggest that the mentors and students were equally aware of the challenges women faced in CS, and have crafted best practices and strategies to retain women in CS.
Influences of Friends and Family on Women’s Pursuit of Computing: A Sequential Explanatory Design
Presenter: Maral Kargarmoakhar, Florida International University
The participation of female students in computing fields remain low. This study examined the most important factors influencing female students computing occupation. The impact of family and friends on four different racial groups including White, Black, Hispanic, and Asian (n=1650) were examined. Results indicated friends had a positive significant role on White female students major decisions.
Closing the Signaling Gap
Presenter: Tessa Forshaw, New Sector Alliance and Stanford University
As society faces an uncertain future of work, workforce development needs a new paradigm. A preliminary trial of a skills visualization tool suggests that when participants in a workforce development program created a skills visualization map, the quantity and quality of skills used to self-describe increased, as does the likelihood of them being recommended for an internship.
Cybersecurity for the Internet of Things: Vulnerabilities of a smart home doorbell system
Presenter: Claudia Rojas, Marist College
The Internet of Things (IoT) refers to a network of devices, including consumer electronics, embedded with software, sensors, and network connectivity that enables them to collect and exchange data. We discuss security vulnerabilities identified from testing three common smart doorbell products using port scans, and man-in-the-middle attacks using a WiFi rogue hotspot.
Increasing student retention in STEM through academic communities that enable informal mentoring
Presenter: Sahithya Reddivari, Georgia State University - Perimeter College
The academic communities established at Perimeter College broaden participation, improve retention, and increase student persistence by enabling informal faculty and peer mentorship among students who are first generation, at-risk, and from minority populations. The qualitative data and findings reported in the poster will identify key mentorship facets achieved through these academic communities.
Back to top
Accelerating Autonomous Driving with Blockchain Technology
Presenter: Nagashree Tadapatri Gundu Rao, SAP
More than 1 billion cars travel the streets of the world today. While cars have made our lives easier and comfortable, they have also made our lives riskier. Autonomous vehicles bring structure to the chaos. With the advancements in automotive industry and technologies like Blockchain, self-driving cars are getting closer to becoming a viable method of transportation that will make driving safer.
A Cooperative Driving Performance Rating System for Defensive Reckless Driving Alerting
Presenter: Lan Zhang, University of Florida
Reckless driving severely threatens the safety of people, which accounts for around 33% of all fatalities in major vehicle accidents. Most existing efforts focus on the detection and adjustment of a vehicle’s own driving behaviors with limited effectiveness. We propose a defensive alerting system to automatically and intelligently detect and notify the threats posed by reckless vehicles.
Back to top
Customer Sentiment Analysis with Extracted Keywords Using Lexical Chains and SentiWords
Presenter: Chhavi Sharma, Microsoft
This paper uses customer reviews to extract keywords for a product and gives sentiment score using ‘SentiWords’ on top of extracted keywords. Reviews are scored using lexical chains, and WordNet provides synsets to identify relevant words. These keywords and sentiment score are useful for providing summary of customer sentiments without having to read the entire data and help in business analysis.
A Parallel Hierarchical Attention Network for Style Change Detection
Presenter: Marjan Hosseinia, University of Houston
We propose a parallel attention network for style change detection. Given a document, it verifies if it has at least one writing style change. Unlike the conventional recurrent networks that use word sequences to learn the language model, our model learns the hierarchical structure of English language using a pretrained statistical parser. It stays at the second rank of PAN 2018 with 82% accuracy.
Exploiting Textual, Visual and Product Features for Predicting the Likeability of Movies
Presenter: Mahsa Sahafaei, University of Houston
In this work, we propose a multimodal model to predict the likability of movies using textual, visual and product features. We capture different aspects of movies and use them in binary and multi-class classification and regression models to predict IMDB rating. We achieve 76% and 63% weighted F1-score for binary and multiclass classification, and 0.7 mean square error for the regression model.
The Fine Line between Linguistic Generalization and Failure in Seq2Seq-Attention Models
Presenter: Leena Shekhar, State University of New York - Stony Brook and NVDIA
We look at a Seq2Seq neural model’s ability to generalize on a simple symbol rewriting task with a clearly defined structure. We find that the model’s ability to generalize this structure beyond the training distribution depends greatly on the chosen random seed, even when performance on the standard test set is the same. Thus, showing their sensitivity and the need for non-standard test sets.
Isa: Intuit Smart Agent, A Neural-Based Agent-Assist Chatbot
Presenter: Ting-Yu Ko, Intuit
To assist customer support agents in promptly resolving a customer’s issue in the call centers, we have developed an agent-faced, neural-based conversational solution that employs BiLSTM with attention. The system can help agents provide a high-quality service by reducing customer wait time and applying the knowledge accumulated from historical customer interactions.
Sentiment Analysis of Twitter Users Towards #Charlottesville Unite the Right Violent Protest
Presenter: Vivian Guetler, West Virginia University and Syracuse University
Using NLP techniques, this study examines opinions and emotions from tweets regarding the white supremacists’ violent protest. A random sample of 200,000 tweets with the hashtag #Charlottesville were analyzed. Overall, the opinions were negative suggesting users were against the 2017 protest in Charlottesville, VA, yet some few positive opinions show there exists some supporters of the protests.
Classifying Sentiment of Tweets Based on External Factors
Presenter: Vicki Liu, Bloomberg
Emotions are grounded in contextual experience. While sentiment classifiers typically look at text to predict an author’s emotional state, factors occurring throughout the day, such as the weather or news, may prime one towards certain emotions. We explored five external factors’ impact on a user’s emotional state and found that extrinsic features can be a powerful predictor of Twitter sentiment.
Neural Machine Translation for Low-Resource Indian Languages
Presenter: Muskaan Singh, Thapar Institute of Engineering and Technology
The global Machine Translation market enables content to be available in all regional languages across the globe. As computational activities become more mainstream and the internet opens up the wider multilingual and global community, research and development in Machine Translation continue to grow at a rapid rate. In this work, Neural Machine Translation is developed and presented.
Multimodal Decisions for Conversational AI
Presenter: Malihe Alikhani, Rutgers University
We demonstrate a conversational system for food ordering that integrates natural language text with links to a visual interface touch. The system uses reinforcement learning to plan when to deploy text and touch, deriving a policy estimated to resolve ambiguity as effectively as possible as a function of the user’s likely goal and context.
Markovian Magic: A Suggestion Engine with Nothing to Hide
Presenter: Ruchi Asthana, IBM
Chatbots effectively engage users on webpages, but they lack guidance (provided by live agents) to push the customer journey. Leveraging Watson Assistant for intent classification and a Markov Model for intent/buyer stage prediction, we created a transparent suggestion engine (lacking hidden layers) to recommend questions to customers in a way that preserves this conversational-journey guidance.
Why is AI 'a sea of dudes?' Using data science and NLP methods to understand gender imbalance
Presenter: Ramona-Gabriela Comanescu, University of Cambridge and University of Edinburgh
We employ topic modeling to capture how the shift in the field of Computation Linguistics affects the gender gap and contrast this with earlier findings. Our analysis suggests that there are subtle ways in which gender differences can occur in scholarly authorship and practitioners should be aware of the dangers.
Images Complement Text in Extracting Possessions from Social Media
Presenter: Dhivya Infant Chinnappa, University of North Texas
This work determines if authors of tweets possess the objects they tweet about. My research shows that both humans and neural networks benefit from images in addition to text. I also introduce a simple yet effective strategy to incorporate visual information, where I semantically represent tags identified in an image to a neural network. Experimental results show this novel strategy is beneficial.
Predicting Phases and Goal Attributes in Health Coaching Dialogues
Presenter: Itika Gupta, University of Illinois - Chicago
In this paper, we present our ongoing work towards building a virtual health coach. We will first discuss the health coaching corpus we collected, schemas for annotating health goals and coaching stages/phases, and inter-annotator agreement results. We then use these annotations to train supervised models for predicting coaching phases and goal attributes in these dialogues.
Automatic Tagging of Stack Overflow Questions Using Word Embeddings and Deep Learning
Presenter: Hannah Senediak, Youngstown State University
Question-and-answer websites like Stack Overflow require users to attach up to five tags when they submit a question. However, users may assign tags that are not relevant to the question. A better approach would be to recommend to users. The goal of this project is to combine newly developed natural language representations with deep learning algorithms to improve the prediction of tags.
Back to top
Is Dieting Really Stressful?: Sentiment Analysis On WeightLoss-Related Online Posts
Presenter: Jihyeon Kim, Ewha Womans University
A number of communities and apps for weight management are limited to logging meals or workouts without considering emotions, which is known to have a strong impact on health. We conducted a RNN-based sentiment analysis on weight loss-related online posts. We built a SVM model and show that the accuracy of predicting weight loss is higher if sentiment analysis results of the post is considered.
Translating neural signals to text using a Brain-Machine Interface
Presenter: Janaki Sheth, UCLA
Brain-Computer Interfaces help patients with neurodegenerative diseases communicate by producing text or speech using brain signals. However, existing systems restrict patient conversations to a limited pool of words. Our method of using novel input features and a deep-neural-network model eliminates this constraint and to our knowledge, is the first such generalization of a BCI to natural speech.
Social Context Middleware for Mental Health Crises: Examining Risky Behaviors in Military Veterans
Presenter: Nadiyah Johnson, Marquette University
Many veterans undergo difficulties when reintegrating into civilian society. Post-Traumatic Stress Disorder (PTSD) due to combat and related exposures affects 15-20% of veterans. In this paper, we document the design and implementation of a smartphone-based HCI social context middleware that coordinates the collection of data potentially relevant for monitoring high-risk behavior in veterans.
An Eyes and Hands Model: Extending Visual and Motor Modules of Cognitive Architectures
Presenter: Farnaz Tehranchi, Pennsylvanian State University
A form of Artificial Intelligence is simulating human intelligence. Cognitive architectures provide a Unified Theory of Cognition for developing and simulating cognition and human behavior but are not interactive cognitive models. I propose a new tool, JSegMan, to facilitate cognitive models to interact with the world and extend visual and motor modules of cognitive architectures.
AI vs. Human Intelligence: Privacy Implications of Assistive Tools for Visually Impaired People
Presenter: Taslima Akter, Indiana University, Bloomington
Camera-based assistive technologies (AT) have the potential to empower people with visual impairments (VIPs) to obtain more independence. VIPs are adopting artificial intelligence (AI) and human intelligence (HI) based AT to overcome accessibility barriers. We focus on the privacy concerns experienced by VIPs while using AT and report their preferences on AI vs. HI AT in different contexts.
Assisting Neurodivergent Individuals in Forming Lasting Relationships Through A Digital Medium
Presenter: Elliot Fox, Western Washington University
Among those who identify with Autism Spectrum Disorder, there is a noticeable trend of adults living their lives in isolation. The research problem we addressed is how to design a communication platform that those with autism can use to create and maintain relationships. Our research is a human-centric study based on the analysis of papers and user interviews, followed by prototype revisions.
Chat-Box: Mood Analyzer for Individuals with Social Interaction Disabilities
Presenter: Bineeta Gupta, Arizona State University
Perception and understanding of social cues is a fundamental communicative skill which gets hampered by hearing and cognitive disorders. Understanding informal words and sarcastic intent is difficult for individuals with social interaction disabilities which can lead to isolation. Addressing sentiments of the slang words and detecting sarcasm is necessary to create an emotional cue analysis tool.
GitRDone: Using Nudge Theory and Positive Reinforcement to Improve Git Usage Practices
Presenter: Brittany DePoi, Cigna
GitRDoneBot is a product that does not attempt to automate tasks away from engineers who use git, but rather the feedback so over time they will implicitly strengthen practices rather than being explicitly told what to do. Gathering good git practice from published work we aim to improve git usage and understanding through the application of nudge theory and positive reinforcement in real-time.
Countdown to Adulthood: How Parents and Teens Navigate the Concerns and Promises of Smartphones
Presenter: Phoebe Chua, University of California and University of California, Irvine (UCI)
Current parental control apps assume that parents monitor teens’ phone use to prevent smartphone addiction. Through interviewing parent-teen dyads, we find that parents view that the phone can promote children’s success in life. They then use various tactics to help teens cultivate appropriate relationships with the phone. Therefore, we encourage designers to consider these broader perspectives.
Back to top
Poster Session 2
Wednesday, October 2, 3 — 5:30 p.m.
CNN+LSTM model for classification of Twitter messages
Presenter: Alisa Gazizullina, Innopolis University
Tweets majorly containing words of idiosyncratic nature are difficult to analyze. In order to prevent loss of the information introduced by context and out-of-vocabulary words appearing in the training set, we design models capturing the specific nature of Twitter posts. Our hybrid CNN+RNN model trained on fastText embeddings outperforms current state-of-the-art models predicting message deletion.
Reducing tag identification time in a molecular tagging system
Presenter: Aishwarya Mandyam, University of Washington, Seattle - Computer Science and Engineering
A molecular tagging system with DNA based tags can provide a secure, difficult to fake identifier. An optimal system minimizes the time needed to accurately identify DNA molbits, which are unique DNA sequences. I created an error analysis pipeline that quantifies the minimum subset of reads on a MinION Nanopore sequencer necessary to produce a similar identification error rate as the entire set.
Human Protein Atlas Image Classification
Presenter: Akshita Bhagia, University of Massachusetts, Amherst
This task aims to assign protein organelle localization labels to protein samples. The ResNet34 model has been fine-tuned using different learning rates for different parts of the network. Focal loss is used as the loss function. Output is the set of target labels, and different thresholding schemes are evaluated for obtaining labels from class scores. Performance is evaluated on macro F1-score.
Parallel Implementation of an Astronomical Algorithm for a Hybrid Computing Environment with OpenACC
Presenter: Ana Luisa Solórzano, Universidade Federal de Santa Maria
Astronomy deals with large datasets that need to be manipulated and analyzed to study cosmic evolution. Friends-of-Friends (FoF) is an algorithm to process this data. The existing serial and parallel FoF versions run in CPU and do not reveal significant performance improvements, exceeds memory usage or use complex structures. In this work we present a simple parallel FoF for GPU using OpenACC.
Private Cooperative Control
Presenter: Andreea Alexandru, University of Pennsylvania
Present-day technologies rely on distributed control, which reduces the load on a single central party, but demands communication and data sharing between the participating agents. However, agents might not trust their peers with their private local data. I propose a cryptographic scheme that ensures the private computation of each agent’s control decisions, without leaking their neighbors’ data.
Towards Human-like Grasping and Manipulation in Assisted Grasps Through Accurate System Modelingl
Presenter: Paria Esmatloo, University of Texas at Austin
Despite the potential and complexity of current assistive hand devices, their grasping and manipulation abilities are far inferior than humans due to the elementary control methods used. I aim to enable accurate control of fingertip position and forces by developing models of the subsystems and the interactions among them, enabling the users to achieve accurate grasping and dexterous manipulation.
A cost-effective framework for IoT-based autonomous ecological sensing
Presenter: Nwamaka Okafor, University College Dublin and Federal Polytechnic Nekede, Owerri, Nigeria
Our world is currently being threatened by increased level of Green House Gases (GHGs) production. To meet new targets for global GHG levels, it is key to monitor GHG production. Current monitoring technologies such as Internet of Things present high cost with no formalized architecture. This paper details a cost-effective framework for the collection, curation, and processing of environmental data.
Opening the Black Box: A Tool Allowing Advisors to Engage in Knowledge Discovery
Presenter: Nasheen Nur, University of North Carolina, Charlotte
We are using a human-centered design approach to engage student advisors in the development of an interactive knowledge discovery tool for a better understanding of student success and risk. It allows the advisors to interact with the data used in an analytic model, the results of the analysis, and the story of individual students. This research is supported by NSF under award number 1820862.
A Refined Deep Learning Model for Optimal Semantic Tiny Object Detection and Segmentation
Presenter: Maryamossadat Aghili, Florida International University
Tiny object detection has a wide application in various object recognition problems such as traffic sign detection, counting the number of blooming flowers in orchards, counting the number of people present in a densely populated picture, detecting electrical components on an electrical board, detecting small insects in occluded complex arable or outdoor lands and many more.
How Can We Detect Impact of Documentary Films from Text? Using Social Computing to Assess Impact
Presenter: Rezvaneh Rezapour, University of Illinois at Urbana-Champaign
We present a study where we analyzed user-generated reviews to detect and predict various types of impact of documentaries on people. After developing a classification schema (change versus reinforcement of cognition, behavior, emotion) and annotating the corpus, we used methods from machine learning and natural language processing to predict impact, achieving accuracy rates of 81% (F1).
How Widely Can Student Performance Prediction Models Generalize?
Presenter: Niki Gitinabard, North Carolina State University
Blended courses are getting popular in secondary education and record a rich amount of data on students’ study habits. Prior research has shown that these habits are correlated with students’ performance, but predictive models for blended courses are still limited. In this work we evaluate predictive models and their application for early and cross class predictions to help at-risk students early.
Early Detection of Cyberbullying Events in Online Social Media
Presenter: Niloofar Safi Samghabadi, University of Houston
In this research study, we apply the early text categorization strategies to find cyberbullying incidents in social media. We first present a new corpus specialized for early cyberbullying detection, then propose our preliminary approach using lexical and semantic features to design a system that monitors the stream of users’ data and detects cyberbullying events as early as possible.
Automated Classification of Media Content for Violence using Convolutional Neural Networks
Presenter: Anusha Munjuluri, University of California, Berkeley
Children are consuming media at an increasing pace and it is extremely hard for parents & teachers to identify if a particular piece of media (books/movies/songs) is appropriate for their child’s consumption based on their age. This project applies a variety of Natural Language Processing & Deep Learning models to automatically tag media content for violence/age-appropriateness with 82% accuracy.
Exploring Health Challenges and Solution Approaches among the Female Rohingya Refugees in Bangladesh
Presenter: Kimia Zaman, North South University
Rohingya refugees have faced various problems during their transition to Bangladesh. A close study with this community along with corresponding doctors opened up challenges impacting the community in their physical health as well as mental conditions. Ensuring the best support for this community for physical and mental wellbeing, We have proposed a solution that can support health workers.
Secure Key Storage in Hardware
Presenter: Shahrzad Keshavarz, University of Massachusetts at Amherst
Saving today’s hardware systems against attackers can be a huge concern considering the budget spent on the design and the sensitive information they may contain. We propose a methodology for secure key storage in hardware that applies threshold-defined behavior to memory cells. Our method achieves a high degree of protection against reverse engineering without compromising the key reliability.
Towards Dynamic Vehicular Clouds
Presenter: Aida Ghazizadeh, Old Dominion University
Motivated by the success of cloud computing, vehicular cloud (VC) was introduced as a group of vehicles whose corporate computing, sensing, communication, and physical resources can be coordinated and dynamically allocated to authorized users. Our main contribution is to design a dynamic VC model involving vehicles on a highway and offer an easy-to-compute approximation of job completion time.
On Hierarchical Data Integrity for IoT Devices in Connected Health Applications
Presenter: Maryam Karimi, University of Pittsburgh
Health tracking devices are used to supplement or replace expensive health monitorings. In recent proposals, patients own their medical data and doctors access it. Patients rely on clouds to maintain data and are responsible for the integrity of retrieved data, from multiple points. We present a simple approach to verify if the data, sent by health monitoring devices to the cloud, remain unchanged.
Quantifying the Sentiment of Online Drug Reviews
Presenter: Gabrielle Gurdin, Virginia Commonwealth University
Patient-generated online drug reviews offer informative data for the medical community. Natural Language Processing methods for determining these reviews’ sentiment could be used for postmarketing surveillance of drugs. We present a dataset of 123,152 WebMD.com reviews, and evaluate the effectiveness of five supervised learning models for the sentiment analysis of this dataset.
Visualizing Parallel Concepts: Guided Scheduling, Work Sharing, Work Stealing
Presenter: Sarah Hendriksen, Calvin College
As of CS2013, parallel computing is now part of the CS core curriculum. Parallel computing concepts are often hard to understand because they are abstract and hard to visualize. ParallelAR is a teaching tool that uses either an augmented reality or desktop version to help students visualize aspects of parallel computing using an office space analogy.
Private Connections: Unique Privacy Provisions for a Neurodiverse Community
Presenter: Phil Fox, Western Washington University
Western Washington University’s “Connection” dating and friendship-facilitating mobile application is tailored towards a Neurodiverse (incl. ASD, Asperger’s, more) community. Our platform is inclusive by-design, and we find that special privacy considerations welcome a maximal amount of users across our target audience. We propose particular privacy feature implementations based on our findings.
Interactive Communicative Technology demonstrating Cognition in Non-speaking Autistic Individuals
Presenter: Jhillika Kumar, Georgia Institute of Technology
This study leverages breakthroughs in neuroscience, observations from occupational therapy sessions, and eye tracking technology to demonstrate cognitive competence with an aim of restoring communicative abilities in non-speaking individuals with autism. By monitoring cognitive arousal, the study provides a quantitative examination of therapeutic techniques to improve communicative speech.
Predicting Smoking Urges from Biometric Markers
Presenter: Forest Sweeney, Western Washington University
Tobacco addiction is one of the most challenging behavioral health problems. We present findings from a study where we collected 398.2 hours of physiological data from regular smokers in the field (N=5) using commercial wearable devices. We developed a statistical model capable of inferring lapse vulnerability from physiological and contextual signals collected from the natural environment.
Clustering Heterogeneous Autism Spectrum Disorder Data
Presenter: Mariem Boujelbene, University of Louisville
Autism spectrum disorder is a developmental disorder that affects communication and behavior. The goal is to mine homogeneous groups of patients from a heterogeneous set of medical records including ADOS data, behavioral data, and fMRI data. In addition, we design a framework that adds explainability to clustering algorithms in a way that assists the user to understand the predictions.
Predicting Suicidal Ideation via Reddit Posts
Presenter: Uma Bhattacharyya, University of Illinois at Urbana-Champaign
Suicide rates are on the rise and pose a societal concern. This study uses text mining to analyze such postings on social media sites to determine if suicidal intention can be predicted. The study finds strong relationship between certain sentimental words and suicidal intentions. Application of the findings may support early intervention to prevent suicidal actions.
Back to top
Correlated Sensor-based Data Fusion in Mobile Crowdsensing
Presenter: Samia Tasnim, Florida International University
With the increase in smartphone usage, a new mode of data collection named mobile crowdsensing (MCS) has emerged. However, its open structure allows malicious entities to interrupt a system by reporting erroneous data. The goal of this research is to show that the introduction of a reputation system in the correlated sensor-based data fusion will enhance the overall quality of the sensed data.
Exploring Ensemble Learning for Anomaly Detection on Smart Home IoT System
Presenter: Zhaochen Gu, University of North Texas
The massive scale of Internet of Things (IoT) device deployment enables access to a wide gamut of data generated by these devices. Anomalies may occur where the devices fail to self-diagnose themselves. We aim to build a model from heterogeneous data and differentiate good situations from the bad. Our solution to anomaly detection of smart home devices is our aggregate model learning system.
Back to top
Avoiding Energy Hole and Coverage Hole Problem in Wireless Sensor Networks Using Matrix Completion
Presenter: Ridhi LNU, Expedia and Indira Gandhi Delhi Technical University for Women
The aim is to avoid energy and coverage holes using correlation and matrix completion in a network. Nodes deployed in a corona-like structure; coalitions are made using correlation among these nodes. The data of coalitions sent to the sink is processed using matrix completion to retrieve data of all nodes at the sink; that reduced the number of messages sent to sink; thus conserving energy.
A Novel Timestamping Mechanism for Clouds and Its Application on Available Bandwidth Estimation
Presenter: Phuong Ha, University of Nebraska Lincoln
The packet time information at receivers carries useful network information. It is challenging to accurately measure this information in a cloud network due to various factors such as virtual machine (VM) scheduling. We propose a novel packet timestamping mechanism and demonstrate its application using an improved available bandwidth estimation tool.
Back to top
Poster Session 3
Thursday, October 3, 10 a.m. — 12:30 p.m.
Hybrid Parallelization of Particle in Cell Monte Carlo Collision (PIC-MCC) algorithm
Presenter: Unnati Parekh, Dhirubhai Ambani Institute of Information and Communication Technology
Kinetic simulation of plasma with PIC method involves numerical constraints on simulation parameters and time-scale making it computationally prohibitive on CPUs for large problem sizes. We illustrate hybrid parallelization of PIC code and compare the performance with shared-memory based parallel code. Our technique scales efficiently with increasing problem size and number of cores in cluster.
Transforming Query Sequences for High-Throughput B+ Tree Processing on Many-Core Processors
Presenter: Ruiqin Tian, College of William and Mary
With modern CPUs, it is possible to process larger batches in parallel in latch-free B+ tree query processing model without extra delays. With increased batch size, there will be more optimization opportunities exposed beyond parallelism, especially under highly skewed query distributions, such as avoiding redundant and unnecessary queries. We propose QTrans to efficiently solve this problem.
Scalable Algorithmic Methods for Large-scale Graph Mining
Presenter: Naw Safrin Sattar, University of New Orleans
Graph mining is of great importance in solving real-world problems in many application domains. Community detection is a computationally challenging problem, solved efficiently with Louvain Algorithm.We design parallel algorithms using OpenMP, MPI,use dynamic load-balancing to achieve around 12-fold speedup.We also present a brief overview of our work on temporal graphs.
Back to top
Predictive Modeling of RNAs for Permanent Shunt Placement in Pediatric Intraventricular Hemorrhage
Presenter: Komal Agrawal, Translational Genomics Research Institute
Intraventricular hemorrhage (IVH) is a complication in premature infants. Babies that develop hydrocephalus as a result of IVH often require a permanent ventriculoperitoneal shunt. Here, we will identify RNAs correlated with shunt placement and develop a predictive model based on RNA expression. Our predictive model could potentially assist clinical decision-making for patients with IVH.
Using Virtual Reality to Examine the Anxiety-Related Response
Presenter: Rachneet Kaur, University of Illinois at Urbana Champaign
We examine a novel virtual reality (VR) based experimental setup to study the anxiety-related responses in balance-demanding target-reaching leaning. An EEG and EKG data collection system is established to analyze the real-time responses from subject’s brain and heart. The Brain-computer interface system integrating VR and online anxiety feedback provides the technology to study task performance.
Learning models for writing better doctor prescriptions
Presenter: Tingting Xu, Boston University
We develop a data-driven treatment recommendation framework to predict physicians’ prescription effects, learn and improve the doctors’ prescription policy by optimizing over the estimated policy model. With the recommended prescription policy, diabetic patients see significant blood glucose reduction on average and obtain a better therapeutic effect than the state-of-art deterministic algorithms.
Spatiotemporal Pattern Synthesis Using Machine Learning and Optimization
Presenter: Noushin Mehdipour, Boston University
A novel approach is proposed to synthesize spatiotemporal patterns in stem cells population. Combining feature engineering, machine learning, optimization and statistical analysis, a hierarchical framework is designed to automatically identify optimal conditions which lead to desired spatiotemporal patterns in stem cells, advancing emergent behaviors in multicellular systems and tissue engineering.
Virtual Letdown: A Virtual Reality App for Breastfeeding Moms
Presenter: Pinar Yanardag, Bogazici University
The global market for VR in healthcare is expected to reach 3.8 billion by 2020. Created at MIT Media Lab, award-winning Virtual Letdown is an immersive virtual reality app that help moms pump milk more efficiently, and enjoyably. Specifically targeted for moms who pump at work, pre-term moms who need to establish & maintain their milk supply while their babies are in NICU!
Back to top
Constructing an Ecologically-Valid Formal Markovian Model of Human Activities
Presenter: Beiyu Lin, Washington State University
While pervasive computing technologies becoming mainstream for observing human behavior, few formal models have been theorized of behavior based on collected sensor data. We construct Markov models that express the nature of human behavior captured by smart home sensor data. We will evaluate if behavior becomes less Markovian and more random as individuals experience cognitive decline.
Single molecule subcellular analysis of RNA distribution in three dimensions using open source tools
Presenter: Pearl Ryder, Emory University
Cells are multifunctional units of life that coordinate tasks by generating intracellular domains of organelles and macromolecules, such as proteins and RNA. We developed an open-source approach to precisely analyze the distribution of individual RNAs relative to centrosomes, organelles involved in cell proliferation. Our approach can be generalized for spatial analysis of other microscopy images.
Hate Lingo: A Psycholinguistic Analysis of Hate Speech in Social Media
Presenter: Dana Nguyen, University of California, Santa Barbara
We focus on a largely neglected but crucial aspect of hate speech – its target: either directed towards a specific person or entity, or generalized towards a group of people sharing a common characteristic. Our work provides a data-driven analysis of the nuances of online-hate speech that enables not only a deepened understanding of hate speech and its social implications, but also its detection.
Narrative Threads Extracted Through Topic Modeling and Sentiment Analysis
Presenter: Chloe Quinto, Stevens Institute
EVE Online is a space-based multiplayer online game that fosters complex online communities. Understanding narratives that unfold across complex online communities is crucial to be able to interact successfully with the customers. By extracting, analyzing, and visualizing topic evolutions across time within the forums of EVE Online community, narrative threads can be seen from a new perspective.
Stock Price Prediction using Convolutional Neural Network
Presenter: Huayan Zhong, Bloomberg LP
Modeling the stock market is challenging due to its complexity and volatility. Our novel approach to this treats the daily stock market as an image inside which stocks are grouped based on similar historical price moves. We apply a 2D convolutional neural network to the daily market images to predict future stock returns. Our method outperforms a random forest regression baseline.
ADQuaTe: An Automated Data Quality Test Approach for Constraint Discovery and Fault Detection
Presenter: Hajar Homayouni, Colorado State University
Data quality tests validate the data to detect violations of constraints. Domain experts grapple with the issues related to the capturing of all the important constraints and checking that they are satisfied. ADQuaTe is data quality test approach that uses machine learning techniques to discover constraints from data, detect records that violate the constraints, and explains the violations.
BERE: Bayesian Quality-Estimator for Reproducible Biological Experiments
Presenter: Shuowei Li, University of Washington
We propose a Bayesian inference framework (BERE) that enables accurate and efﬁcient root cause analysis for reproducible biological experiments. Given the experimental record, BERE automatically infers the most probable source of variability in the experiment, and formulate the scheduling of the experiment as an optimal experiment design problem.
EnhancedMonitoring: A Large Scale Monitoring System
Presenter: Urvashi Senha, Microsoft
Extracted information from semi-structured emails is displayed in Outlook. EnhancedMonitoring calculates the precision metric of these extraction models and reports precision issues to ensure quality user experience. It scales despite increasing traffic volume and adheres to privacy regulations. Heuristic methods and Anomaly Detection are combined to detect incorrect extractions more efficiently.
DeepAuth: A Framework for Continuous User Re-authentication in Mobile Apps
Presenter: Sara Amini, University of Illinois - Chicago
In this work, we leverage currently available built-in motion sensors in smartphones to learn users’ behavioral characteristics while interacting with the mobile device to provide an implicit re-authentication mechanism that enables a frictionless and secure user experience in the application. We present DeepAuth as a generic framework for re-authenticating users in a mobile app.
Improving Student Motivation through Competitive Active Learning
Presenter: Manika Kapoor, IBM
In this research, we describe the CLP system and present the results of a set of analyses aimed at gauging the impact of competitive Active Learning activities using the CLP system on student motivation, engagement, and performance. Results indicate that competitive active learning is beneficial in this setting, leading to active student participation and improved motivation.
Data-Driven Reserve Prices for Social Advertising Auctions at LinkedIn
Presenter: Lijun Peng, LinkedIn
Online advertising auctions constitute an important source of revenue for LinkedIn. We study the problem of setting the optimal reserve price in a Generalized Second Price auction, guided by auction theory with suitable adaptations to social advertising. We demonstrate through field experiments the effectiveness of this reserve price mechanism to increase revenue and improve advertiser experience.
Back to top
Augmenting Visual SLAM with WiFi Sensing For Indoor Applications
Presenter: Zakieh Sadat Hashemifar, State University of New York at Buffalo
A key enabling technology for spatial applications is understanding of the device’s location and the map of environment, referred as Simultaneous Localization and Mapping (SLAM). Visual SLAM algorithms suffer from perceptual aliasing and high computational cost. Since WiFi routers are ubiquitous in urban environments, we utilized the WiFi received signal strength to alleviate those challenges.
Learning Decentralized Controllers for Robot Swarms with Graph Neural Networks
Presenter: Ekaterina Tolstaya, University of Pennsylvania
We consider the problem of distributed control in large networks of mobile robots with limited communication. By extending aggregation graph neural networks to time varying signals and networks, we learn a common local controller which exploits information from distant teammates using only local communication. We demonstrate the performance of the decentralized flocking controller in simulation.
Computational Theory of Robust Localization Verifiability in the Presence of Outlier Measurements
Presenter: Mahroo Bahreinian, Boston University
The relative measurements are not accurate, can have noise and outliers. We are concerned whether an L1-norm robust optimization can recover a solution identical to the ground truth. We prove the verifiability of a problem depends on the topology of graph, the sign of outliers, and the edge support of outliers; it is independent of ground truth locations of the nodes and the scale of the outliers.
Back to top
Catbot: Automated Triage Bot for Incident Management
Presenter: Kanika Sabharwal, Bloomberg
In this poster, we highlight how an automated bot can be a great solution for incident management in a company and a better solution than traditional incident management paradigms. To support our claim, we introduce Catbot, the incident management bot for our team in Bloomberg, its architecture, and how using it has had a big impact towards making incident management more effective in our team.
Website Performance: Cache can cost
Presenter: Kavita Asiwal, Intuit
Using a browser’s ability to cache resources eliminates the time to download resources and allows an application to load faster. Most engineers consider ways of using browser cache when designing a resource deployment strategy. This poster will highlight the cost of loading resources from a disk cache, the conditions when cache costs are high, and ways to reduce this cost.
Taming Web Views in the Detection of Android Privacy Leaks
Presenter: Xue Qin, University of Texas, San Antonio
Engineering Future AI Research: Designing Sonnet 2
Presenter: Tamara Norman, DeepMind
Scientific research is often constrained by the limits of engineering, and designing systems which are efficient yet flexible for research is a trade-off Software Engineers are required to balance. Sonnet is a neural network library built on top of TensorFlow, designed in collaboration with researchers at DeepMind, has been instrumental in accelerating DeepMind research.
Back to top
Is Mobile App Tracking Really Anonymous and Aggregative?
Presenter: Xueling Zhang, University of Texas, San Antonio
Mobile app companies utilize third-party tracking service to learn their app users’ behavior. An app developer could set user attribute on tracking. This paper studies what attribute did app developers choose for tracking, whether they choose personally identifiable information (PII) such as social security number or email. We performed a study on 1,031 apps and found 125 use unencrypted PII.
Ancile: Attack Surface Reduction Through Application Specialization
Presenter: Priyam Biswas, Purdue University
Control-Flow Hijacking is the most common attack vector against C/C++ programs. Automatically specializing code for particular usage scenario is a promising new technique for software security. I present Ancile which implements required functionality analysis, achieves significantly smaller subset of functions by specializing the code, and substantially raising the bar for CFH attacks.
Truth Inference for Crowdsourcing under Data Poisoning Attacks
Presenter: Farnaz Tahmasebian, Emory University
Crowdsourcing provides a cost effective solution for obtaining service from a large group of users. A key component of these applications is truth inference which aims to derive the true answer for a given task. A challenge presented to crowdsourcing is data poisoning attacks. We propose a comprehensive data poisoning attack for truth inference in crowdsourcing and evaluate these methods.
Visual Cryptography based Authentication Scheme for Medical Images over Cloud
Presenter: Sonal Kukreja, Thapar Institute of Engineering and Technology
We propose an authentication scheme based on Visual Cryptography for medical images stored over cloud to handle data breaches caused due to various cyber-attacks. The novelty of this scheme is that it embeds watermark into the medical image without modifying any pixel, no complex computations required for extraction of watermark, and high robustness against various image processing attacks.
Zero Trust ‐ Infrastructure to Application
Presenter: Shamna Mattammel, eBay
As platform security engineers, we wanted to share how we designed our next generation security products which protects eBay starting from provisioning of a VM to the application to application flow. By sharing our journey we hope to help others achieve a security-first infrastructure and give users the trust they deserve.
Back to top
Poster Session 4
Thursday, October 3, 2—4:30 p.m.
An End-To-End Learning for Autonomous Driving: Steering Angle Prediction and Lane Keeping
Presenter: Yaqin Wang, Indiana University - Purdue University
The end-to-end approach is one of the frequently used approaches for the autonomous driving system. In this study, we adopt the end-to-end approach to train a deep convolutional neural network to steer a car in the Udacity simulator. Our proposed model map raw pixels from cameras of three different angles and generate steering commands for autonomous driving.
Ambient Assisted Dementia Care using Smart Home featuring Activity Recognition and Decision Support
Presenter: Gayathri K.S, Sri Venkateswara College of Engineering
Smart home, a ubiquitous environment aims to offer Ambient Assisted Living. The activity modeling framework integrates ambient intelligence into the home environment. This research proposes an assistive dementia care system through smart home that offers functional assistance to the dement occupant during critical situations without the help of caretaker.
Visual Tools for AI Creators To Understand and Diagnose AI Systems
Presenter: Arunima Chaudhary, IBM Research
Building AI applications is not only tricky to execute but the lack of diagnostic tooling to make their behavior transparent and explainable for developers and designers keep them from creating more robust solutions. We built and present two visual tools to help the AI creators to understand and diagnose AI systems.
Temporally Evolving Community Detection and Prediction in Content-Centric Networks
Presenter: Ana Appel, IBM Research
We consider the problem of combining link, content, and temporal analysis for community detection and prediction in evolving networks. Most of the work deals with static networks. Incorporating dynamic changes in the communities into the analysis can also provide useful insights. Chimera is a shared factorization model that can simultaneously account for graph links, content, and temporal analysis.
To spend resources or to generate them? A different perspective in Egocentric Activity Recognition
Presenter: Sheila Maricela Pinto Caceres, University of Sydney
Creating robust methods to identify a person activity is the foundation towards technological wearable assistance. Most of current research has focused on improving accuracy by using resource intensive techniques neglecting the constraints of wearable technology. This work aims to reduce the battery consumption by generating visual information instead of making intensive use of the camera.
Online Learning for Multi-skill Orchestration
Presenter: Sohini Upadhyay, IBM Research
Orchestrating multiple dialog agents, or skills, to create a unified system is an open problem in AI. Using online learning for orchestration offers several advantages over supervised learning. We provide a template for integrating online learning into orchestration with contextual bandits and propose a new bandit algorithm that navigates a feature retrieval challenge in orchestration.
Customized Spell Checker Using Multinomial Bayes Classifier
Presenter: Sheryl Zhang, Bloomberg
There is an increasing trend to use deep learning in real-world applications. However, sometimes a simple probabilistic classifier can be very powerful. Here we demonstrate how we utilized a classic multinomial Bayes classifier to perform customized spell checking on Bloomberg’s text datasets and the enhancements to make our approach more precise and tailored to our need.
Model-based Actor-Critic with Mutli-agent Assistance
Presenter: Haoran Wei, University of Delaware
Reinforcement Learning (RL) can solve complex decision-making problems, but the convergence requires a large number of real-environment interactions. Simulating on a high-fidelity environment model can help to speed up a single RL agent. In this paper, we scale up RL acceleration to domains where there is no high-fidelity simulator available and real environment interactions are costly.
Using Brain MRI Images to predict BMI, Memory, and Age.
Presenter: Chhavi Yadav, New York University
A lot of work has been done on applying deep learning to medical problems.In this paper,we use Brain MRI images to predict memory score, body mass index and age of a person with simple 3D neural network architectures.We model the problem as single & multitask predictions. We compare the performance of the two approaches. We also use visualizations to find potential biomarkers for these parameters.
SparkleCalibration: Glitter Imaging for Single-Image Camera Calibration
Presenter: Maya Shende, The George Washington University
We are working towards a system for camera calibration with a single image. That single image is captured of a sheet of glitter illuminated by a single light source. Here we describe our methodology for calibrating a camera based on the sparkles it sees from this glitter, as well as fully characterizing the glitter (solving for the position and surface normal of every piece of glitter).
Behavioral Planning for Automated Vehicles using Deep Reinforcement Learning
Presenter: Meha Kaushik, Microsoft
Reinforcement Learning enables us to train agents that can perform tasks typically humans have been known to do well on. We show how to use Deep Reinforcement Learning with an imposed curriculum to generate human-like driving behaviors like lanekeeping or overtaking without any data. This can be used for developing intelligent simulators for advanced autonomous driving research.
Computer vision tools for evaluating visual focus of attention during infancy
Presenter: Qazaleh Mirsharif, CrowdFlower
Infants actively use visual cues to interact with the world prior to acquiring verbal skills. Correctly understanding such cues is integral to the childhood development. We present computer vision tools which helps in understanding visual focus of attention in infants. We discuss how such tools help developmental scientist to explore new patterns in visual attention development during infancy.
Back to top
Chemical Cyber-attack Awareness
Presenter: Nia Blake, Winston-Salem State University
The American public never receives information about the vast majority of cyberattacks against U.S. industrial systems that control the operations of refineries, pipelines power plants, chemical plants, and other factories. These incidents are classified by the FBI and other agencies, and kept secret by U.S. companies unless a breach involves the theft of personal identifying information.
A Machine Learning Approach to Sleep Quality Prediction for Inpatient Rehabilitation
Presenter: Allison Fellger, Gonzaga University
Actigraph watches are a tool through which we can view the minute-by-minute sleep activity of unhealthy populations with minimum intrusion. While this data can be quite noisy, machine learning can help us analyze these irregular patterns. Through a machine learning approach, we predicted nighttime sleep duration for a group of post-stroke and post-traumatic brain injury patients.
Algorithms and Tools for Protein Variant Library Design
Presenter: Sarah Almeda, The College of New Jersey
When ordering oligonucleotides for a protein variant library, biologists must take many factors into account, from how the pieces are assembled to the codons and nucleotides used in the design. This project involved developing algorithms that optimize these variables to minimize cost, as well as a web-tool and an online database to make our work accessible to the greater scientific community.
Why Do Instagram Users Create Multiple Accounts?
Presenter: Liza Sivriver, Washington University in St. Louis
Increasingly, Instagram users maintain both public- and private-facing accounts, known as rinstas and finstas, respectively. This poster explores users’ goals in maintaining multiple accounts and how accounts interact with one another. I developed a system to compare and analyze differences between users’ rinstas and finstas with respect to content, network structure, and social metrics.
Evaluating the Perceptual Validity of Distance Metrics in Character Animation
Presenter: Katherine Gemperline, Xavier University
In character animation, it is important for many applications to know how similar two motions are. For this, the Root Mean Square Error is a commonly used distance metric. However, it does not always correctly map to our perception of similarities. In my research, I check several types of motion errors with the same similarity values. I investigate how these motions are perceived to help develop more perceptually valid distance metrics in the future.
Multi-SpooNN: A Lightweight Network for Multiple Object Detection
Presenter: Madelyn Gatchel, Davidson College
Many autonomous robots rely on efficient neural networks (NN). SpooNN is a lightweight convolutional NN for object detection on FPGAs; however, the network only supports single object detection without classification. In this project, we extend the network capability to detect and classify multiple objects and then evaluate network performance on various datasets appropriate for autonomous robots.
Physiological Data Analysis for Stress Prediction and Intervention
Presenter: Annika Sougstad, Massachusetts Institute of Technology
Given its significant negative health outcomes, minimizing stress is a high priority. This project created a software toolkit to analyze physiological data collected from college students via a wearable technology during induced stressful events in a lab session. The software interprets the data to maximize the accuracy and reliability of detecting physiological response to stressors.
Believe it or Not: The Human Perception of Altered Media
Presenter: Nneka Udeagbala, Stevens Institute of Technology
Deep fakes introduce a new level of complexity by creating distrust towards video evidence of events. Our research discusses the utilization of social media and news sources for media authentication, lessening the impact of published altered videos. User studies were used to investigate human perception limitations and dependency on media sources for detecting altered media.
Accessible CSS: Accessible Comprehensive Validation Tool for Screen Reader User
Presenter: Haoran Wen, Rutgers University
As the Internet and web-enabled technologies become ubiquitous and there is greater need for web-related jobs, there is a lack of diversity and representation by persons with disabilities. One factor contributing to this is that the production of web technologies presents various accessibility barriers for individuals that are blind or low vision. CSS and visual styling are areas of particular stumbling blocks that lacks easy, accessible, and comprehensive tools for nonvisual CSS validation. CSS is a core language and component of the web used to describe visual representations and due to the visual nature of CSS nonvisual developers struggle to with its use: often time relying on sighted third party member to assist with validating their CSS. In order to striving for a more diverse participation, better accessibility support, and greater independence of blind or low vision web developers we evaluated existing CSS tool to aim for creating a accessible CSS validation tool that would allow blind and low vision web developers to build, test, and produce websites and web applications with greater confidence and independence.
Teaching Intro Programming in VR
Presenter: Tajah Warren, Rhodes College
Having the correct mental model for assessing C code can boost students confidence and retain interest. We examine the traditional education methods of teaching students memory allocation and compare them to the new method that we developed using VR. We examine the traditional teaching methods and mental models to understand the disconnect in student ability and understanding of C-language code.
Teaching Intro Programming in VR
Presenter: Briana Williams, Morgan State University
Misunderstanding C code may cause students to leave CS. Our VR program assists students in using the correct mental model. Teachers can implement this VR program into their traditional teaching. We observe students’ interactions and compare the results to traditional teaching methods. This program should increase students’ confidence in their programming skills and maintain interest in the field.
No one reads the instructions: Designing an educational programming game to facilitate learning
Presenter: Jennifer Echavarria, Sewanee: The University of Sewanee
Through iterative design and playtesting, we found that the design of game instructions has a profound impact on students’ ability to play and learn from an educational programming game. We have designed and developed Resource Rush, a farming-based game to teach introductory programming. To win, players must plant, water, and harvest materials to solve each level of the game. Players are offered a choice of using keys and menus or writing short programs to achieve game goals. Since the tasks are repetitive, the game is designed to naturally incentivize programming. Our initial design included typical text-based instructions, containing the game objective and a listing of the game controls. However, very few of our initial playtesters attempted to use programs to achieve game goals, and expressed a lot of confusion about what to do or where to find game resources. Anecdotally, we observed that almost no one read the instructions, and if they did, they did not find the information they needed. Through an iterative series of four design revisions and playtests, we evolved our instructions to enable students to learn the game mechanics and objectives during gameplay. Overall, our lessons learned were that game instructions should be just-enough and just-in-time, and should demonstrate example incomplete or incorrect solutions rather than requiring students to build programs from scratch.
Explaining Fashion Recommendations with AI-based Justifications
Presenter: Lindsay Huang, University of Southern California
Consumers can experience choice paralysis when shopping for fashion online: on sites like Amazon, a single query often returns hundreds of products. This poster presents a strategy for generating AI-based justifications that explain fashion recommendations. I will present the results of a user study that measures how these types of justifications influence trust perception and buying decisions.
Synthesizing robot behavior for caregiver use in personalized neurorehabilitation for MCI patients
Presenter: Emma Peterson, University of Minnesota, Twin Cities
Robotic assistance in healthcare is becoming more widespread, but users may not have the programming knowledge to tailor a robot’s behavior to a task. We introduce a novel program synthesis approach that allows end users, e.g. clinicians or caregivers, to program an assistive robot’s behavior to support patients with mild cognitive impairment, allowing for personalized neurorehabilitation.
Cumulative Adversity: The Contribution of Correlated Stressors on Mental Health in College Students
Presenter: Olivia Figueira, Santa Clara University
Adversity comes in many forms, but the accumulation of these adversities and relationship among them have significant effects on mental health. My research examines the impact of these stressors in a sample of college students leveraging actively reported data from surveys. I aim to find the correlations within and between several stressors using statistical data analysis and visualization.
A Dynamic Conflict-Based Search Algorithm for Multi-Agent Multi-Task Planning
Presenter: Hannah Lee, Colorado School of Mines
In this work, we present a method for finding an optimal plan for a set of unordered tasks using a multi-robot team. An optimal plan is generated using a conflict-based search that modifies invalid plans to solve robot allocation conflicts. The search is extended to dynamically replan as tasks and robot availability changes using a dynamic programming table that implements Dijkstra’s algorithm.
Using motion planning to make inferences about drug molecule pathways in bound and unbound proteins
Presenter: Abigail Ren, Vassar College
My research project uses motion planning to study how pathway accessibility changes between bound and unbound proteins. The problem is modeled by using the protein as an environment and the drug molecule as a robot moving through the free space of the protein. We gathered data to show that the accessibility of the protein’s binding site decreases after a drug molecule binds to the protein.
Assessing Protein-Ligand Binding Accessibility Using Motion Planning and Energy Metrics
Presenter: Bonnie Wang, Columbia University
My project uses motion planning algorithms to evaluate and understand the accessibility of a drug molecule (ligand) to a binding site within a protein. To better predict the ligand’s path to the site, we bias our planning strategy towards lower energy pathways by annotating our protein model with biometrics (energy), resulting in a more informed and accurate model of the accessibility routes.
A Robot Motion Planning Algorithm that Finds Safer Paths Faster
Presenter: Regina Rex, University of Wisconsin-Superior
In robotics, motion planning is typically used to help robots navigate the environment without colliding with obstacles. We present a planning algorithm which uses clearance to help robots exploit their environment properties, navigate more efficiently, and select safer routes. Our method is faster and performs fewer collision detection calls compared to existing methods.
Robot Perception of Human Group Motion Forecasting
Presenter: Joymaneet Kaur, University of California, Berkeley
As robots enter human-centered environments, they are expected to work alongside groups of people and require an understanding of social dynamics. Robots can learn this using group motion forecasting. Therefore, the objective of this project is to design an algorithm that enables robots to predict the future motion trajectory of groups of people. As such, robots will be able to employ safer motion planning and work seamlessly among people.
Poster Session 5
Friday, October 4, 10—12:30 p.m.
External Consistent Transactions Using Logical Clocks
Presenter: Masoomeh Javidi Kishi, Lehigh University
We present SSS, a scalable transactional key-value store deploying a novel distributed concurrency control that provides external consistency and never aborts read-only transactions due to concurrency. SSS uses a combination of vector clocks and a new technique, called snapshot-queuing. Results show significant speedup over state-of-the-art competitors in read-dominated workloads.
Detecting latent bugs in Partially Synchronous Distributed Systems Using SMT Solvers
Presenter: Vidhya Tekken Valapil, Michigan State University
In this poster, we will discuss about the problem of detecting latent concurrency bugs in partially synchronized distributed systems. Partially synchronized distributed systems do not have a specific total order of events in the system. So evaluating all possible serializations of events in the system to guarantee correctness is a challenging task and we propose to use SMT solvers to handle this.
Learningbot: Using AWS Lambda to Facilitate Intra-Company Peer Learning Through a Slackbot
Presenter: Nicole Kowtko, Tribe Dynamics
Within our company, there was no system in place for voicing learning needs. Though our peers’ skill sets are diverse and individuals are open to helping each other learn a skill, it’s hard for an individual to keep track of who knows what in the organization, creating a barrier to entry. Our primary learning channel was Wednesday presentations on any topic, but the topics were chosen by whoever volunteered to lead the learning instead of those who wanted or needed to learn a particular skill.
Back to top
High-level Functional Hardware Design
Presenter: Mahshid Shahmohammadian, Drexel University
Low-level languages for circuit descriptions, like VHDL, easily lead to errors in large designs. Much higher-level tools like Simulink yield fewer errors but lack modularity and efficiency. We show how language features of Haskell, like high-order functions, parametric polymorphism, and type classes, can provide a high-level design experience that reduces errors without sacrificing efficiency.
Energy Efficient Desktop Computers
Presenter: Jennifer Alphonse, Intel Corporation
Over the last two decades, the number of computer systems has exponentially increased. The number of mobile users has quintupled since the early 2000s. With this increase is the increased demand for power — a valuable commodity. This poster shows how software and hardware optimizations are used to design green, energy efficient systems that meet the aggressive low power requirements of desktop computers.
Back to top
Mechanics-Aware Modeling of Cloth Appearance
Presenter: Zahra Montazeri, University of California, Irvine
Fabrics are essential in our daily lives so designing and modeling them virtually is important. Capturing high-detailed appearance is challenging while ignoring the fiber-level details yield to a fake look for cloth. We introduced a new-of-a-kind technique to model cloth under physical forces. Our model accurately reproduce the change of appearance when fabrics interact with the environment.
Automating Quality Control at Disney: A Novel Aspect Ratio Defect Detection Tool
Presenter: Anna Wolak, The Walt Disney Company
Before films are released to the public, they must be inspected for defects. Currently, quality control teams must perform the majority of these examinations manually, which is not only time-intensive but also prone to human error. At Walt Disney Studios, we have developed an automated tool for aspect ratio shift detection that outperforms state-of-the-art, out-of-the- box software applications.
Back to top
Ensemble Approach: Detecting Fake URLs on Social Media
Presenter: Sawinder Kaur, Thapar Institute of Engineering and Technology
People get trapped through enormous online content. Many fake URLs exist. The idea of this paper is to evaluate ensemble models to build a predictive model to detect real-time fake URLs using best ensemble classifier. In this paper, it has been observed that Extra Trees classifier helps to build the best prediction model for fake URLs detection with 96.7% accuracy with minimum log loss of 2.11.
A Blockchain Based Architecture to Enable Secure Data Acquisition for Clinical Data Analytics
Presenter: Atigallage Rupasinghe, Monash University
Data acquisition for clinical analytics has become challenging as data custodians are being reluctant to disseminate data to external entities due to privacy and security concerns. As the ownership of medical records lies with patients, a patient-driven collaborative approach leveraging blockchain technology and smart contracts will be explored in this study to resolve this challenge.
Clickstream Pattern Analysis And Prediction using Machine Learning
Presenter: Neha Kumari, Intuit
Clickstream analysis is key to finding patterns like user drop-off and anomalies. Predicting drop-off can improve the customer conversion and retention. Poster will cover the process and algorithms for clickstream data analysis, clustering, anomaly detection, and machine learning. This poster will demonstrate the process and algorithm used for drop-off prediction using a machine learning algorithm.
Back to top
Glass Weave Skew in High Speed Channels
Using glass weave in Printed Circuit Boards creates an inhomogeneous environment for positive and negative sides of a differential pair. This effect creates skew causing a 56G PAM4 channel to operate very poorly. It is shown that for a 548 mil differential trace on a 2-ply 1067 glass, skew is about 1.2psec. We propose two methods to control glass weave skew. We verify these methods by simulations.
Back to top
The Power of Quantum Chemistry in Semiconductor Fabrication
Presenter: Nita Chandrasekhar, Intel Corporation
Hardware fabrication faces challenges of atomically precise control and limited data for novel materials. Quantum chemistry, driven by advances in computing, is finding prominence in solving critical challenges in chip making. We focus on real-life applications of Density Functional Theory (DFT) and present a few examples of DFT in predicting the chemical behavior and properties of novel molecules.
Back to top
Cellular Feature Difference Analysis of Astrocytomas Using Histopathology Images
Presenter: Mousumi Roy, Stony Brook University, State University of New York
Cellular phenotypic features are the basis of pathologic diagnosis and are related to underlying molecular profiles. Due to a large number of cells makes it challenging to quantitatively compute and compare features of cells with distinct molecular signatures. We propose an efficient analysis framework based on segmentation, feature computation, data pruning, dimensionality reduction, and clustering.
A Novel MRI Classifier of Arteriolar Sclerosis in Aging: Predicting Pathology and Cognitive Decline
Presenter: Nazanin Makkinejad, Illinois Institute of Technology
Arteriolar sclerosis is a type of small vessel disease, and has been associated with higher risk of dementia. Diagnosis of arteriolar sclerosis is only possible at autopsy. In this work, an MRI-based classifier of arteriolar sclerosis was developed by training a classifier on ex-vivo MRI and pathology data. Then, it was translated to in-vivo and tested its association with cognitive decline.
Error-Robust Multi-View Clustering
Presenter: Mehrnaz Najafi, University of Illinois, Chicago
Data may come from multiple sources, known as multi-view data. Multi-view clustering finds better clusters by exploiting information from multiple views rather than one view. Due to system errors, each view may be erroneous. Existing multi-view clustering methods cannot handle all error types such as noise and outlier. We propose a novel Markov method for Error-Robust Multi-View Clustering.
Online Learning in Presence of Concept Drift: a Diverse Ensemble based Streaming Approach
Presenter: Kanu Goel, Thapar Institute of Engineering and Technology
In machine learning underlying data distributions tend to change with time called as concept drift. Accurate labeling in case of supervised learning algorithms is essential to build consistent ensembles models. We propose diverse ensemble based concept drift handling approach which handles major types of drift patterns in evolving streams and thereby motivating need for adaptive systems.
Adversarial Multi-Label and Semi-supervised Classification
Presenter: Sima Behpour, University of Pennsylvania
We propose an approach for structured prediction problems, mainly multi-label classification and semi-supervised classification, and provide better performance and meaningful loss bound. It poses the learning task as a minimax game between predictor and “label approximator” based on minimum cost graph cuts. It always provides meaningful bounds on the Hamming loss unlike maximum margin methods.
You are How You Click: Model Based Clustering on Customer Click Paths via Mixture Markov Model
Presenter: Ruilin Zhong, Amazon
Traditional clustering algorithms on click paths usually ignores sequential order, or with results hard to interpret in production environment. We introduce Mixture Markov Model, a probabilistic model-based algorithm to cluster click paths with sequential order preserved, a parameter initialization method to stabilize the output, and a ranking method to intuitively describe results per cluster.
Chord Recognition in Symbolic Music Using Semi-Markov Conditional Random Fields
Presenter: Kristen Masada, Ohio University
We train a semi-Markov Conditional Random Field (semi-CRF) model to segment music into a sequence of chord spans tagged with chord labels. This approach enables the use of segment-level features that capture the extent to which events in a segment of music are compatible with a candidate chord label. Semi-CRF outperforms existing automatic harmonic analysis systems on classical and pop music.
Mitigating Adversarial Deep Learning Attacks in Architecture Mismatch Settings
Presenter: Rehana Mahfuz, Purdue University
The vulnerability of deep learning models to adversarial attacks poses a serious threat. Currently established defenses mitigate attacks generated based on the same architecture, but fail when the attacker behaves unexpectedly. This work proposes a unique defense to combat an architecture mismatch attack by training a denoising autoencoder with perturbations based on a variety of architectures.
Change-point Detection in Time Series using Deep Learning
Presenter: Tahiya Chowdhury, Rutgers University
The problem of finding abrupt changes in time series is known as change-point detection. In this work, we propose a self-supervised, autoencoder based approach for learning features in time series and detect change-points. By comparing the results of our approach on a wide range of real-world datasets, we demonstrate that our approach performs significantly well in predicting expert-annotations.
Deep Learning Framework for Joint POI discovery & Scene Classification of ground level imagery
Presenter: Seema Chouhan, Oak Ridge National Laboratory
We propose a Deep Learning framework that focuses on the utilization of geotagged ground-level imagery for the purpose of scene classification and accurate identification of Points of Interest (POIs) categories (e.g. restaurants, hotels, and schools, etc.) so as to augment efforts in improving location intelligence, such as context-aware POI mapping and for improving land use classification.
Investigating the effect of real-world events on Twitter discussions
Presenter: Lida Safarnejad, UNCC
The rich interaction among Twitter users is a valuable resource to study public opinions regarding trending topics. We present EventPeriscope, a pipeline that utilizes signal processing, text mining, and NLP techniques to discover the effect of a real-world event on the dynamics of Twitter discussions. This pipeline can also identify additional closely-related events that have invoked discussions.
Back to top
Preventing Vision loss of Diabetic Patients: Diabetic Retinopathy Assessment System
Presenter: Jannatul Ferdause Tumpa, Marquette University
Patients suffering from chronic diabetes are at higher risk of developing Diabetic retinopathy (DR), a retinal disease that can eventually cause vision loss. The goal of this ongoing research is to develop a system for automated grading of retinal fundus images to make regular eye screenings efficient and cost-effective, and thus helps to identify the disease early and prevent vision loss.
Women’s Health Application: An Approach to Holistic Healthcare Using a Cross-Platform Framework
Presenter: Zyrene Adao, Cigna Software Innovations
The Women’s Health Application by Cigna is a personalized tool designed to give women a view of their holistic health. Using a cross-platform framework, Flutter, this application was made in hopes of satisfying the needs of both Android and iOS users, while still delivering a consistent user experience. If so, Flutter will create an opportunity to develop more accessible technical solutions.
Back to top
Guaranteeing Security and Provenance with Blockchain
Presenter: Lilian Kasem, Microsoft
Everyone is talking about blockchain, but how do companies actually use it to solve real world problems? In the manufacturing industry, where buyers can purchase schematics on an online marketplace, how can a buyer be sure that what they’ve purchased is what they received? To solve this challenge, we built a blockchain solution that enables buyers to securely request, transfer, and verify IP.
Open Source Rockstar Program
Presenter: Aliza Carpio, Intuit
Open sourcing a project is step one in making an impact in the community. Follow it with an end-to-end program that highlights the value of your OS project so you can increase followers and contributors, which adds even more value to your solution. Learn about the “Open Source Rockstar” program, a holistic plan to connect with peers, increase contribution and “own” your piece of the OS community.
Back to top
Integrating SPDK with Oracle RDBMS
Presenter: Zahra Khatami, Oracle
The goal of integrating SPDK with Oracle DB is to build highly available and scalable applications to serve millions I/Os at low latencies. However, there are few challenges: (1) Multi-process Oracle DB quickly exhausts PCIe hardware queues. (2) DPDK memory model conflicts with Oracle’s shared memory model. In this work, the proposed Oracle Dispatcher and OraEnv handle these challenges.
Back to top
Performance guarantees for bandit algorithms with graphs
Presenter: Drishti Wali, Cornell University
We study the stochastic multi-armed bandit problem with the graph-based feedback structure where we show that the two most prominent stochastic bandit algorithms, Thompson Sampling and UCB, achieve regret guarantees that combine the graph structure and the gaps between the means of the arm distributions. Surprisingly this holds despite the fact that they do not explicitly use the graph structure.
Reachability Query in Spatiotemporal Contact Networks
Presenter: Zohreh Raghebi, University of Colorado Denver
With the rapid development of location sensors, it is now possible to study how items spread across populations of moving objects. Two objects are considered in-contact while they are sufficiently close to each other. This network of objects called contact network. We model evolving network of contacts as spatiotemporal graph and study reachability query through the population of moving objects.
Constructing a Time of Flight Algorithm for Feature Reconstruction through Ultrasonic Scanning
Presenter: Judith Wang, University of Southern California
This project uses ultrasonic testing to detect flaws and make dimensional measurements from a 3D-printed product. Ultrasonic testing uses high frequency sound energy and many functional apparatuses. Through the development of this methodology and its corresponding algorithm, surfaces that are neither reachable or visible to the eye can be reconstructed — improving nondestructive quality testing.
Back to top
Hack the Status Quo: How to Organize an Inclusive Co-Ed Hackathon
Presenter: Biya Kazmi, McMaster University
Accessibility, inclusion and diversity should be factors woven into the fabric of every Hackathon. These events can transcend the current climate of gender gaps and inaccessibility to show the world what the tech industry may look like if we were to challenge systemic inequalities in our community. We will discuss three simple rules any Hackathon organization can use to implement these changes.
When Life Gives Us Lemons… Supporting Careers of Parents of Special Needs Children
Presenter: Vaishnavi Krishnan, Google
There is a growing workforce of “exceptional” parents: those of us who have children with extra physical/mental needs, the children who are truly exceptional, but are also exceptions to the norms of the world. Please stop by to discover how you can manage your career as an “exceptional” parent and how teams can be allies to build a truly diverse and inclusive workplace!
Back to top