Computing Research Symposium 2017 – Poster Presentations
Poster presenters at the 2017 CCC Symposium on Computing Research included early career faculty members, post-docs, and graduate students from many fields of computer science. There were a total of 47 poster presentations during the poster reception. The videos below are from the poster presenters who chose to record a short video clip of their presentation.
To see the full list of poster presenters at the Symposium and their abstracts, please see the CCC Symposium Poster Book.
Data-driven techniques typically present results driven by a majority of the instances considered. As such, they often fail to address the diverse and sometimes contradicting preferences of the minorities and may become barriers to equality. Minorities who differ may have these differences overlooked in deference to the majority. To address this for the popular problem of ranked retrieval, we propose a third-party service that enables on-the-fly re-ranking of results based on user needs and preferences.
Complex algorithmic techniques are being increasingly incorporated into user-facing interactive systems. These systems provide numerous capabilities. However, mismatches can occur between how such systems work and what users interpret them to mean. This poster describes means of incorporating human users into the design process to achieve better alignment between the lay interpretation and the technical functioning of these systems.
Dr. Blanco’s work enables computers to understand language as humans do. Specifically, he is interested in building computational models to infer spatial timelines, extract detailed interpersonal relationships, and reveal hidden interpretations form intricate phenomena such as negation and modality. His research targets implicit meanings that are intuitive to humans when interpreting language, and incorporates temporal inference and uncertainty to create meaning representations from language.
Omar Chowdhury
University of Iowa
Towards the Vision of a Compliant Internet Public-Key Infrastructure (PKI)
With the growing concerns of surveillance by resourceful adversaries, design and deployment of effective secure communication mechanisms have become extremely desirable. Although the core SSL/TLS protocol and its security parameter choices have undergone heavy scrutiny, such level of rigor is absent from the inspection of X.509 implementations, which can make the SSL/TLS vulnerable to impersonation attacks and/or may cause interoperability issues. The vision of this research is to develop a formally verified X.509 reference implementation and an effective mechanism for detecting noncompliance of existing X.509 implementations with the standard. Although conventional wisdom says that symbolic execution is not scalable for such a task, our technique dubbed SymCert has been proven to be extremely effective in exposing noncompliance in real implementations.
Video games have shown potential as a framework for humans and computers to work together on solving challenging problems. Games have engaged players in solving real-world problems from a number of domains, including biochemistry. However, designing such games remains a challenging problem. Using telemetry data from gameplay may provide a means to assist in making and automating game design decisions, thus improving the effectiveness of games as human computation systems.
Sauvik Das
Georgia Institute of Technology
Social Cybersecurity: Reshaping Security Through an Empirical Understanding of Human Social Behavior
Social influences strongly affect cybersecurity behaviors, and it is possible to encourage better cybersecurity behaviors by designing security systems that are more social. In support of this statement, I will report on three projects: an empirical analysis of how security tools diffuse through the social networks; an experiment people in which we show that social proof encourages good security behaviors; and, the design and evaluation of the first socially inclusive authentication system.
Kenneth Fletcher
University of Massachusetts Boston
A Method for Dealing with Data Sparsity and Cold-Start Limitations in Service Recommendation Using Personalized Preferences
Data sparsity and cold-start remains the main limitations in recommendation systems that employ collaborative filtering. Efforts to alleviate these limitations typically require additional user or item information such as social context of users and features of items, besides ratings that are usually available. This poster presents a method to resolve data sparsity and cold-start limitations using users’ personalized preferences on non-functional attributes, as additional information.
Vanessa Frias-Martinez
University of Maryland
Understanding Human Behavior and Resilience During Shocks in Smart and Connected Communities
In this poster, Dr. Frias-Martinez will present two research projects focused on understanding human behaviors during shocks using large-scale, geo-referenced data extracted from ubiquitous technologies. In the first project, she and her research team proposed a novel framework to analyze behavioral changes in human mobility during floods in Rwanda using Call Detail Records (CDRs) from a telecommunications company. For the second project, Dr. Frias-Martinez and her research team proposed a semi-automatic framework to extract and compare the digital communication footprints of citizens and governments during snowstorms on the US east coast using Twitter communications. The end objective of her research is to offer evidence-based information that can help decision makers change, adapt or enhance response policies during shocks.
Rafael Frongillo
University of Colorado at Boulder
Information Elicitation: Crowdsourcing, Peer Grading, Machine Learning
Robust crowdsourcing mechanisms often use monetary contracts that incentivize someone to truthfully reveal their private information. Surprisingly, these contracts also show up in machine learning and statistics as loss functions, which guide the algorithm toward better predictions. Leveraging this connection, we combine ideas from economics and computer science to design new mechanisms for crowdsourcing and peer grading, but also new loss functions for machine learning.
Yufan Guo
IBM Almaden Research Center
Cognitive Review of Multimodal Clinical Data: Improving the Path from Diagnosis to Documentation
EMR systems are intended to improve care management and system analytics. However, the information stored in EMRs can be disorganized, incomplete or inconsistent creating problems at the patient and system level. We present a technology that enables reconciliation of inconsistencies between clinical diagnoses and administrative records by analyzing multimodal clinical data (HL7 and DICOM) in real time. This cognitive data review tool improves the path from diagnosis to documentation, facilitating accurate and timely clinical and administrative decision-making.
Christoffer Heckman
University of Colorado at Boulder
Rapid, Reliable Perception and Control for Autonomous Robots
To address the growing disconnect between probabilistic perception algorithms and deterministic control methods, the theory of dynamical systems is leveraged to develop rigorous, novel control approaches. This work draws on developments from other fields in robotics such as visual-based perception to tightly fuse complex models of the environment and their dynamics. Experimental platforms including autonomous vehicles are used to validate these methods in uncertain and dynamic environments.
Amir Houmansadr
University of Massachusetts Amherst
MassBrowser: Unblocking the Web for the Masses, by the Masses
The Internet plays a crucial role in today’s social and political movements-democracy and human rights throughout the world critically depend on preserving and bolstering the Internet’s openness. Consequently, repressive regimes and totalitarian censor their citizens’ access to the Internet using a wide range of technologies. This poster will introduce MassBrowser, a new system designed in our group to help censored users bypass censorship reliably.
Chien-Ming Huang
Johns Hopkins University
Amplifying Human Abilities through Human-Centered AI systems
In this poster, I will highlight my prior work demonstrating the importance of human-centered design in building effective AI systems that aim to support people in a socially intuitive manner. I will show how embodied AI systems—robots—can employ human-inspired interaction strategies to provide cognitive (learning gains), social (user experience), and task (team performance) benefits to their users in various contexts such as personalized tutoring and adaptive collaboration.
Eyes are a rich source of information about a human. They tell us what a person is interested in, what she finds important, and whether she is fearful or angry. My research aims to leverage eye-tracking technology to create empathetic artificial intelligence.
Zhe Jiang
University of Alabama
Utilizing Spatial Big Data from Intelligent Infrastructure to Enhance Situational Awareness for Disaster Management
Deadly flood events (e.g., Hurricane Harvey) are costing the U.S. billions of dollars each year. Accurately assessing disaster situation is still an unsolved issue. Spatial big data (e.g., high-resolution aerial imagery, volunteered geographic information on Twitter and Google Map) provides unprecedented opportunities to enhance the situational awareness for disaster response agencies. The poster will introduce research on utilizing spatial big data for flood extent mapping in disaster response.
Aron Laszka
University of Houston
Privacy-Preserving Energy Transactions (PETra): Providing Privacy, Safety, and Security in IoT-based Transactive Microgrids Using Blockchains
Power grids are undergoing major changes due to rapid growth in renewable energy and improvements in battery technology. Prompted by the increasing complexity of power systems, decentralized IoT solutions are emerging, which arrange local communities into transactive microgrids. However, providing security, safety, and privacy in such energy systems is challenging. We introduce PETra, a blockchain-based solution for transactive microgrids that enables consumers to trade energy without sacrificing their privacy and provides safety and security for the grid.
Traditional Internet communication leaves communicating parties vulnerable to persecution and disruption from online censorship. Although there are many systems in active use today that seek to resist censorship through anonymous, confidential communication (most notably Tor), they are currently rather brittle in the presence of a large censoring regime. This poster will present our work towards empowering users with greater control over where their packets don’t go. Rather than rely on inaccurate maps of the Internet, we use novel measurement techniques and universal constraints to provide provable guarantees. We have applied this insight to achieve provable avoidance of user-specified geographic regions, and to protect Tor against several challenging attacks from nation-state adversaries.
In this poster, I will present two experiments integrating statistical models, visual analytics techniques, and user experiments to study the effectiveness of combining human- and machine-driven forecasting systems. These experiments are framed around the analysis of social media data for box office prediction problems and compare the prediction performance between a baseline model and users and between different settings of human collaborations. In another experiment using the same system, we analyzed the effects of human collaboration as teams, groups, and individuals. Our results indicate that a team’s performance is mediated by the team’s characteristics such as openness of individual members to others’ positions and the type of planning that goes into the team’s analysis.
Francisco Munoz-Arriola
University of Nebraska-Lincoln
Toward a Resilient Food-Energy-Water-Ecosystem Services Nexus: Analytics and Synthesis
To ensure long-term U.S. global competitiveness, it is imperative to build resilient infrastructure based on cyber-physical systems that ultimately enable sustainable food, water, energy and ecosystem services (FEWES). FEWES systems involve spatially and temporally varying data, present in multiple forms, formats and volumes, available at various exchange rates, and connected to different applications. To create a secure FEWES system we require efficient and effective transformations of multidimensional data into information based on novel theories, designs, and technologic developments.
This poster portrays the problem of fair statistical inference involving outcome variables. The issue of fairness arises where some covariates or treatments are “sensitive,” in the sense of having potential of creating discrimination. We believe the presence of discrimination can be formalized in a sensible way as the presence of an effect of a sensitive covariate on the outcome along certain causal pathways. A fair outcome model can then be learned by solving a constrained optimization problem.
We consider a helper-AI problem with two agents, human and AI that make joint decisions over a longer period of time. The human agent has an imperfect model of the world, so the AI agent needs to take into account possible imperfections of the human’s actions when calculating its optimal policy. We model this as a multi-view decision process, which we use to formally analyze the positive effect of the AI’s steering policies that can lead to a significant improvement of the agents’ utilities.
Privacy notice and choice are essential aspects of privacy and data protection regulation. Yet, today’s privacy notices and controls are surprisingly ineffective at informing users or allowing them to express choice. We analyze why existing mechanisms fail consumers, study how their usability and effectiveness can be increased, and design and evaluate usable privacy mechanisms that empower consumers to effectively manage their privacy online, with mobile technologies and the Internet of Things.
Muhammad Shahzad
North Carolina State University
WiFi Based Human Sensing to Enable Aging-in-Place for the Elderly
With the senior members of the “baby boomers” demographic cohort crossing the age of 70, the US has started to experience a considerable growth in its elderly population. The National Research Council’s report from the workshop on “Grand Challenges of Our Aging Society” has stressed the need to develop technologies that can track and monitor the activities of the elderly, and thus enable them to age in place. We present WiFi based human sensing systems that can unobtrusively monitor activities and gestures of the elderly and enable not only their continuous health monitoring but also gesture-based control of equipment in their smart homes.
Houbing Song
Embry-Riddle Aeronautical University
Intelligent Infrastructure for the Airport of Things Towards Smart & Connected Cities and Communities
There are airport capacity needs in the National Airspace System (NAS). Airport of Things (AoT) is the networking infrastructure for the landside airport assets, systems, and services. Airport of Dependable and Controllable Things will enable the full systems-of- systems optimization in the NAS and make the national airport system safe, efficient, and environmentally responsible.
Eirini Eleni Tsiropoulou
University of New Mexico
Socio-physical Coalition Formation in Smart IoT Applications
A socio-physical coalition formation process among smart Internet of Things (IoT) devices in an intelligent infrastructure environment is proposed and the problem of efficient resource management is studied. Realistic IoT applications, such as multi- purpose sensing in smart homes, smart lighting systems, etc. will be illustrated. The importance of IoT devices’ interest, social and physical ties for the robustness, viability and sustainability of smart IoT applications will be demonstrated.
Ranga Raju Vatsavai
North Carolina State University
Monitoring Critical Infrastructures with Spatiotemporal Edge Computing
In many real-world applications, such as, natural disasters, crop diseases and bioterrorism, traffic, human activity, and public place monitoring, near real-time extraction of knowledge from the sensor data streams is becoming critical. Recent advances in embedded supercomputers (e.g., Jetson TX1, size of a credit card, low power ~10W) are bringing computing closer to the sensors to enable real-time analytics and decision-making. We present the enabling technologies behind this revolution and showcase the utility in critical infrastructure monitoring.
Machine learning has been widely used to help computer systems to better interact with the physical world. Meanwhile, machine learning can also introduce new vulnerabilities. For example, changing a few target pixels in an image may cause a machine learning classifier to misclassify a “stop” sign, causing real troubles to systems like self-driving cars. Existing studies have looked into adversarial machine learning mainly in the digital domain. Our work seeks to understand the feasibility of adversarial attacks in the physical domain and potential ways of defense.
Automatic fake news detection is a challenging problem in deception detection, and it has tremendous real-world political and social impacts. However, statistical approaches to combating fake news has been dramatically limited by the lack of labeled benchmark datasets. In this paper, we present LIAR: a new, publicly available dataset for fake news detection. We collected a decades-worth of 12.8K manually labeled short statements in various contexts from POLITIFACT.COM, which provides detailed analysis report and links to source documents for each case. We have designed a novel, hybrid convolutional neural network to integrate metadata with text that can improve a text-only deep learning model.
While many sophisticated models are developed for visual information processing, very few pay attention to their usability in the presence of data quality degradations. Most successful models are trained and evaluated on high-quality visual datasets. On the other hand, the data source often cannot be assured of sufficiently high quality in practical scenarios. Quality factors, such as occlusion, motion blur, missing data and bad weather conditions, are also ubiquitous in the wild. The seminar will present a comprehensive and in-depth review, on the recent advances in the robust sensing, processing and understanding of low-quality visual data, using deep learning methods. I will further demonstrate how our proposed approach largely improves a broad range of real-world applications, such as traffic monitoring, security surveillance, and video communication.
Gloria Washington
Howard University
Affective Technologies for Improving the Lives of Persons with Chronic Disease
Behavioral and physical characteristics can be used as input to technologies that monitor stress, anxiety, and/or depression in humans. These technologies are commonly found in academic laboratories and not where we need them: mobile devices. This work seeks to improve the lives of sickle cell disease sufferers through motivational, encouraging technologies that also provide dietary tips. Results from this work will help to alleviate the emotional toll for managing and handling chronic disease.
Da Yan
University of Alabama at Birmingham
Big Data Frameworks: Bridging High Performance of HPC Community with Programming Friendliness of Data Science Community
The poster will introduce the latest big data analytics frameworks being developed in my group. They include vertex-centric model for data-intensive graph analytics, subgraph-centric model for compute-intensive graph mining, and matrix-based model for data analytics and machine learning. These frameworks expose user-friendly API to data analysts and guarantee the scalability and efficiency of the underlying execution engine.
Kwang Soo Yang
Florida Atlantic University
Spatial Network Big Data for Transportation Resource Planning: Challenges, Approaches, and Opportunities
Increasingly, Spatial Network Big Data (SNBD) is of a size, variety, or update rate that exceeds the capacity of commonly-used spatial computing technologies to learn, manage, and process with reasonable effort. Examples of SNBD include temporally detailed road maps that provide a driver’s speed every minute for every road-segment, GPS trace data from cell phones, and engine measurements of fuel consumption, greenhouse gas emissions, etc. However, there are many practical challenges today because the methods, models and algorithms currently used for query processing do not scale and/or perform well for large volumes of SNBD. We are currently investigating three research problems within SNBD data processing: 1) storage scheme for SNBD, 2) design of scalable algorithms for transportation resource planning, and 3) unified data model for SNBD.
Yezhou Yang
Arizona State University
Computer Vision-based Quantitative Assessment of Motor Function in Stroke Patients
Objective quantification of movement characteristics of Upper Extremities is a critically necessary step to capture the key characteristics of patient’s sensorimotor impairment reliably. Recently, significant effort has been devoted to developing technology and methods for measuring hand and arm kinematics and kinetics. However, these methods are typically expensive, need significant time to set up and are designed for specific tasks that are unfamiliar to clinicians. To overcome these drawbacks, we explore a novel solution to quantitative assessment of human UE function based on state-of-the-art Computer Vision techniques. This approach combines the use of existing, standardized clinical assessment tools with advances in human movement tracking using low-cost cameras, to measure the motion of the arm and hand, as well as their interaction with the environment.
Identifying the footprint of urban events such as crowd gathering and traffic congestion is crucial to improving transport efficiency and mitigating public safety risks. The poster presents our recent research on urban event analytics, i.e., early detection and forecasting of urban gathering events through mining spatio-temporal big data (e.g., vehicle GPS trajectory). Case studies and experiments on real data are presented to validate the effectiveness and scalability of the techniques.
These were recorded at the Computing Community Consortium (CCC) Symposium on Computing Research, which was supported by the National Science Foundation under Grant No. 1136993. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation
