The CS Research Seminar Talks are talks given by faculty members and undergraduate research assistants on a variety of topics at the cutting edge of computer science research. Talks happen on Fridays at 11:25am (Fall 2023) and are usually held in King 236. The format is a 40-45 minute research talk with 5-10 minutes reserved for questions. All CS students (and other interested students and faculty) are invited to attend. Email announcements about each seminar will be sent to the CS listserv.
Tue 9/12: Distinguished Lecture Series, Dr. Laurent Babout (Lodz University of Technology). King 259 11:30-12:30pm
Fri 9/15: Research Talk: What is Research? (With Mike Lam and Co.) King 259 11:25am-12:25pm Slides
Fri 9/22: Research Talk: Ben Saupp. King 236 11:25am-12:25pm
Thu 9/28: Distinguished Lecture Series, Dr. Everlyne Kimani (Toyota Research Institute). King 259 2:30-3:30pm.
Fri 10/27: Research Talk: Dr. Chris Mayfield. King 236 11:25am-12:25pm
Fri 12/1: Research Talk: Dr. Dee Weikle. King 236 11:25am-12:25pm
Fri 12/8: Research Talk: Dr. Siddharth Bhaskar. King 236 11:25am-12:25pm
Abstracts for the 2023 Fall Research Talks
* Sept. 22 Ben Saupp -- Praxly: An Online IDE for the Praxis CS Test Pseudocode
Abstract: CodeVA has a new professional development (PD) course designed to prepare Virginia high school Career Technical Education (CTE) teachers for the Praxis CS test. This test uses a pseudocode language that is different from traditional programming languages. In CodeVA’s course, teachers do not have an easy way to verify the correctness of pseudocode programs when learning and practicing. This talk will introduce Praxly, an online IDE we developed over the summer, for teachers to write, visualize, and run pseudocode. We will highlight Praxly’s implementation details, outline our current research goals, and discuss next steps for the project.
* Oct 7 Prof. John C. Bowers A proof of the Koebe-Andre'ev-Thurston theorem via flow from tangency packings
* Nov 4 Prof. Siddharth Bhaskar Four naive questions in the theory of algorithms
* Nov 11 Virginia Tech PhD Candidate and JMU CS Alumna Kylie Davidson Exploring Sensemaking Strategies in Immersive Space to Think
Abstracts for the 2022 Fall Research Talks
* Oct 7 Prof. John C. Bowers -- A proof of the Koebe-Andre'ev-Thurston theorem via flow from tangency packings
Abstract: The beginning of the study of circle packings is the circle packing theorem, which says that any triangulation of the sphere can be realized as a pattern of interior-disjoint disks on the sphere whose tangency graph is isomorphic to the triangulation. The now classic Koebe-Andre’ev-Thurston theorem extends the circle packing theorem to weighted triangulations whose weights code for overlap angles of up to pi/2 between disks and not just tangencies. Recently Connelly and Gortler gave a new proof of the Circle Packing Theorem, which is also constructive—the proof gives rise to a numerical algorithm for computing a circle packing for any given triangulation. In this talk we will give an overview of Connelly and Gortler’s result and describe a new proof of the full Koebe-Andre’ev-Thurston theorem that extends their approach to packings with overlaps up to pi/2 and gives rise to an algorithm for computing them. We will discuss the main hurdles to the generalization and the main ideas used to overcome them. We will also show demo code for computing the packings.
* Oct 7 Prof. Siddharth Bhaskar -- Four naive questions in the theory of algorithms
Abstract: In this talk I will raise four questions that you could, in theory, ask after taking a first data structures and algorithms course, viz.,
--Why are so many sorting algorithms quadratic time?
--Why are stacks and queues the “only” implementation of certain abstract data type (namely add, query a distinguished element (the “top”), delete the top, and an empty test) for collections of unstructured data?
--Why are there only about three different types of proofs of general lower bounds?
--What is an algorithm anyway?
To be clear, I don't know a definitive answer to any of these, but the questions seem to be deep and continue to inspire me. I shall discuss each question, illustrate what the underlying problem is, and say a few words about how each has inspired some specific research projects.
* Nov 9 PhD. candidate Kylie Davidson -- Exploring Sensemaking Strategies in Immersive Space to Think
Abstract: Existing research on immersive analytics to support the sensemaking process focuses on single-session sensemaking tasks. However, in the wild, sensemaking can take days or months to complete. In order to understand the full benefits of immersive analytic systems, we need to understand how immersive analytic systems provide flexibility for the dynamic nature of the sensemaking process. In our work, we build upon an existing immersive analytic system – Immersive Space to Think, to evaluate how immersive analytic systems can support sensemaking tasks over time. We conducted a user study with eight participants with three separate analysis sessions each. We found significant differences between analysis strategies between sessions one, two, and three, which suggest that immersive space to think can benefit analysts during multiple stages in the sensemaking process.
Faculty Research Talks Overview
* Feb 11 Prof. Siddharth Bhaskar -- Recursion, iteration, and program schematology
* Feb 25 Prof. Chris Mayfield -- POGIL in CS1: Evidence for Student Learning and Belonging
* Mar 4 Prof. Mike Lam -- Recent Work in Floating-Point Program Analysis
* Mar 25 Prof. Kevin Molloy -- Advances in Protein Characterizations
* Apr 1 Prof. Chris Johnson -- Programming language for Parametric 2D design
* Apr 8 Prof. David Bernstein -- Real-Time Pricing of Taxi Services
* Apr 15 Prof. Jason Forysth -- Designing Wearable Systems for Movement Correction in Physical Therapy Applications
* Apr 29 Prof. Kirsten Richards -- TBD
Abstracts for the 2022 Spring Research Talks
* Feb 11 Prof. Siddharth Bhaskar -- Recursion, iteration, and program schematology
Abstract: We can distinguish between logical and nonlogical syntax of a programming language: the former being “pure programming” constructs (like `if-then-else' statements) whose semantics is independent of the data type in question, and the latter being “primitive operations” whose semantics depends on the data type in question. This allows us to talk about programs which are logically equivalent, in the same sense that the logical formulas $(\forall x) \neg A(x)$ and $ \neg (\exists x) A(x)$ are equivalent—their denotation is the same no matter the interpretation of $A$.
It is folklore that every recursive procedure can be transformed into an iterative one, whose only recursive construct is a “while” loop. (Indeed, this is basically what every compiler does in transforming a high-level language to machine code.) However, very early on (the 60's!) in the history of programming languages, it was realized that there were general recursive programs not logically equivalent to any iterative program, and therefore that recursion elimination relied on semantic properties of the data in question (e.g., closure under taking lists).
In this talk, I will discuss the general problem of recursion vs. iteration, the separation I mentioned above, and my small contribution to this story, which has to do with the intensional (i.e., complexity-theoretic) distinction between the two paradigms. This talk is classical in two senses: one, it's concerned with problems which were thought about early in the history of programming languages, and two, it's personally classical in the sense that it harkens back to the stuff I was thinking about in grad school.
* Feb 25 Prof. Chris Mayfield -- POGIL in CS1: Evidence for Student Learning and Belonging
Abstract: For the past ten years, computer science instructors have adopted Process Oriented Guided Inquiry Learning (POGIL). Other STEM disciplines have shown conclusively that POGIL impacts student learning and knowledge retention. However, most research about POGIL in computer science has focused on perceptions and experiences, not learning outcomes. In this study, we examined the influence of POGIL on student learning in CS1. We collected data from all sections of CS1 at the same institution. Four of the faculty implemented POGIL, and three taught with other active methods. The learning data included pre and post assessments, midterm and final exams, and a retention test at the beginning of the next course. Students also completed three surveys about their prior programming experience, sense of belonging, and perceptions of teamwork. We used multiple regression to analyze the relationship between the survey data and learning outcomes. Our results show that students in the POGIL sections outperformed students in the other sections. POGIL students scored higher on the post-test, and a higher proportion of them met the grade requirement to progress to the next course. After the five-week winter break, POGIL students had higher and more consistent scores on the retention test. These results provide evidence that POGIL can be very effective as an instructional technique in computer science.
* Mar 4 Prof. Mike Lam -- Recent Work in Floating-Point Program Analysis
Abstract: There has been a sharp uptick in interest in floating-point precision analysis over the past decade, primarily in the context of high-performance computing (HPC) but also more recently in machine learning (ML). As HPC moves into the exascale era and ML increasingly dominates scientific workloads, it's increasingly important to understand where our algorithms are sensitive to small numerical insensitivities and where we need more or less precision. This presentation will explore recent work on the adaptation of existing analysis techniques (dynamic instrumentation and Monte Carlo Arithmetic) to Python programs. Historically, Python was of little interest to the HPC community because of the performance hit associated with being an interpreted language, but lately the language has become more prevalent both in HPC and across many areas of computer science (this was part of the motivation for the recent switch in CS 149). In this talk, I will present the work done by others and describe my recent attempts to reproduce and compare their results to previous work, and I will also discuss possible student projects for future extensions of the work.
* Mar 25 Prof. Kevin Molloy -- Advances in Protein Characterizations
Abstract: The study of proteins and the relationship between their sequence, structure and function has been ongoing for decades. While some of this research is performed in a biochemical laboratory setting, the processes employed can be very time consuming and expensive. To augment these techniques, structural biologists rely and need computational approaches. Many of the algorithmic challenges with modeling protein systems arise from the fact that these systems are typically modeled in very high dimensional settings.
This talk will provide a brief overview of proteins and their composition to enable reviewing how these systems are modeled computationally. A review of Google's DeepMind group's recent work on protein structure prediction, and how their approach has made tremendous progress in this field. Next, we will review work by our group to better understand anti-microbial peptides and building a universal structural alphabet. Finally, we will discuss work being performed to structural group ensembles of proteins, and specifically the virus that causes COVID-19 (and its variants).
* Apr 1 Prof. Chris Johnson -- Programming language for Parametric 2D design
Abstract: What if instead of printing “Hello, world!” to a terminal, your first program produced a tangible object? Dr. Johnson has been trying to make that a reality with Twoville, a programming language in which users code two-dimensional objects and fabricate them using a vinyl or laser cutter.
Twoville is meant to be a sandbox for young people to explore mathematics and computer science — and for Dr. Johnson to explore programming language features that facilitate parametric design. One big takeaway from his exploration is that not everything is an algorithm. Some designs need a more human touch. That's why Twoville supports bidirectional editing. A user can edit the design either through code or through mouse manipulation, with any changes made via the mouse propagated back to the code. Dr. Johnson will be joined by his family, who will demonstrate some of the objects they've made.
* Apr 8 Prof. David Bernstein -- Real-Time Pricing of Taxi Services
Abstract: Historically, taxi pricing has been tightly regulated, with limits placed on the number of vehicles and prices set by the government. The advent of alternatives to traditional taxis, like Uber and Lyft, have dramatically changed the way these kinds of services are and can be priced.
This talk will illustrate why the pricing of these services is interesting using some simple examples. It will begin with a review of equilibrium pricing in perfectly competitive markets (for those who have never studied microeconomics, or for those who have forgotten what they once knew). It will then review some of the literature on modeling equilibria in spatially separated markets. With that as background, it will use some simple models to explain how models of the pricing of taxi services are similar to and different from those models. It concludes with a discussion of how one can develop existence and uniqueness conditions for general problems, and how general problems might be solved numerically.
* Apr 15 Prof. Jason Forysth -- Designing Wearable Systems for Movement Correction in Physical Therapy Applications
Abstract: Musculoskeletal injuries are the leading cause of disability worldwide. Patients suffering from these injuries are often given exercises to perform that strengthen muscle groups or joints that have been injured. Typically, these exercises are observed in person at the physician's office, and the patient is then instructed to continue them at home; however, challenges exist with patient adherence to these exercise regimes, with some studies showing that 37% of patients failed to continue their exercises at home. Failure to complete these activities leads to longer recovery times, additional visits to the physician, and generally lower health outcomes.
In this talk, we will examine research conducted by the Wearable Computing Group to address these challenges by developing a wearable computing system to monitor patient at-home exercises. The system will actively monitor patients to provide real-time feedback to improve their functional gains and adherence to exercises. Prototype demonstrations will be presented along with discussions of specific research questions including capturing patient posture through various sensor systems, designing methods for feedback through visual, haptic, and audible modalities, and machine learning techniques to estimate patient improvement over time.
Faculty Research Talks
Faculty Research Talks
Undergraduate Thesis Presentations
Do You Understand IEEE Floating Point?
Floating-point arithmetic is used for nearly all real-valued computation in modern software. CS 261 teaches the basics of the IEEE standard format as well as some of its caveats (e.g., rounding error and its non-uniform distribution). However, there are many other floating-point issues that cannot be covered in that course due to time constraints, and many of them may surprise even an experienced software developer. This talk will discuss some of these issues and explore the published results of a recent survey from Northwestern University that attempted to quantify the level of understanding of floating-point issues among software developers across academia, national labs, and industry. The talk will also highlight the importance of these findings to ongoing research projects here at JMU.
POST: A Machine Learning Based Paper Organization and Scheduling Tool
Organizing and assigning papers into sessions within a large conference is a formidable challenge. Some conference organizers, who are typically volunteers, have utilized event planning software to ensure simple constraints, such as two people cannot be scheduled to talk at the same time. In this work, we proposed utilizing natural language processing to find the topics within a corpus of conference submissions and then cluster them together into sessions. As a preliminary evaluation of this technique, we compare session assignments from previous conferences to ones generated with our proposed techniques.
Advisors: Profs. Kevin Molloy and Michael Stewart
Machine Learning Application to Sand Dune Model Prediction
The ability to simulate the movement of sand dunes over a certain period of time is beneficial as they have a significant impact on a wide variety of people ranging from dry to coastal areas. Current physics-based simulation models are highly inefficient and computationally expensive. Here, we train a Generative Adversarial Network (GAN) to make video predictions of sand dune data in an attempt to decrease the computational effort required to run these simulations.
Advisor: Dr. Barry Rountree (Lawrence Livermore National Laboratory)
In high performance computing, many applications can easily generate terabytes to petabytes of floating-point data. Moving this data between nodes, over the internet, or within the memory, hierarchy presents a significant bottleneck to these programs. We propose a specialized accelerator to preform floating-point array compression using the ZFP compression algorithm. This talk will explore the potential benefits and difficulties of developing this accelerator.
Advisor: Dr. Scott Lloyd (Lawrence Livermore National Laboratory)
What is CS research? (Profs. Lam and Weikle)
Followed by
Faculty Flash Talks (Everyone)
Join us for an overview of computer science research followed by 5-minute flash talks giving you an overview of ongoing research within the JMU CS department. Whether you are looking to get into research or just interested in learning more, you are welcome!
The Effects of Finite Precision on the Simulation of the Double Pendulum
We use mathematics to study physical problems because abstracting the information allows us to better analyze what could happen given any range and combination of parameters. The problem is that for complicated systems mathematical analysis becomes extremely cumbersome. The only effective and reasonable way to study the behavior of such systems is to simulate the event on a computer. However, the fact that the set of floating-point numbers is finite and the fact that they are unevenly distributed over the real number line raises a number of concerns when trying to simulate systems with chaotic behavior. In this research we seek to gain a better understanding of the effects finite precision has on the solution to a chaotic dynamical system, specifically the double pendulum.
Advisor: Prof. Mike Lam
A Study of the Effect of Memory System Configuration on the Power Consumption of an FPGA Processor
With electrical energy being a finite resource, feasible methods of reducing system power consumption continue to be of great importance within the field of computing, especially as computers proliferate. A victim cache is a small fully associative cache that “captures” lines evicted from L1 cache memory, thereby reducing lower memory accesses and compensating for conflict misses. Little experimentation has been done to evaluate its effect on system power behavior and consumption. This project investigates the performance and power consumption of three different processor memory designs for a sample program using a field programmable gate array (FPGA) and the Vivado Integrated Development Environment.
Advisor: Prof. Dee Weikle
Data Science and Fake News Detection
When classifying text with Machine Learning algorithms, textual features have to be extracted and used to train the classifier. We extract several different features: word counts, ngram counts, term frequency-inverse document frequency, sentiment analysis, lemmatization, and named entity recognition. Different features give classifiers different accuracies. We test the features on multiple classifiers to determine the best features for Fake News detection. We explore classifying news articles as either Fake News or as not Fake News using three datasets, which in total contain over 40,000 articles.
Advisor: Prof. Ramon Mata-Toledo
Visualizing Chaos in the Bunimovich Stadium
The Bunimovich stadium is a chaotic dynamical system in which a single particle, known as a billiard, moves indefinitely within a barrier without loss of momentum. Mathematicians and physicists have been actively researching its properties since it was discovered to be chaotic in the 1970’s. Current tools for visualizing the system's behavior are inadequate for visualizing a key aspect of its chaos: sensitive dependence. We developed software called the Bunimovich Stadia Evolution Viewer (BSEV) to ameliorate this issue. It provides visualizations that could help researchers gain insights into sensitive dependence.
Advisor: Prof. Ramon Mata-Toledo
WHAT THE WASM? Learn you a WebAssembly for great good!
Javascript’s monopoly on interactivity in the web is coming to an end. WebAssembly is a new open standard for delivering compiled client-side programs to web browsers. Currently JavaScript is the only native option for client-side scripting in the browser. WebAssembly aims to change this while delivering increased performance. At this point the tools are green-to-nonexistent but the standard is published and implemented in all major browsers. In this talk we will survey the state of WebAssembly, highlight the new scripting pipeline that WebAssembly provides, and in the course of this talk attendees will code an interactive webpage in a language that is not JavaScript.
AlphaGo Zero and the Downfall of Humanity
In 1997 IBM's Deep Blue defeated World Chess Champion Gary Kasparov, marking the end of human dominance in the game of Chess. For nearly twenty years a similar victory proved elusive in the game of Go. Even with dramatically more powerful computers, the best Go programs only reached the level of strong amateur play.
This changed In October 2015 when DeepMind's AlphaGo became the first computer program to defeat a professional Human Go player. The AlphaGo team accomplished this using deep neural networks combined with Monte-Carlo Tree Search. Then, in 2017 DeepMind introduced AlphaGo Zero, an even stronger program trained entirely through self-play. This talk will provide an overview of the AlphaGo Zero algorithm, as well as some discussion of the implications of this work.
POGIL in Computer Science: Faculty Motivation and Challenges
Introductory computer science courses face multiple challenges, including a broad range of content, diverse teaching methods, and the need to help students develop skills such as problem solving, critical thinking, and teamwork. One evidence-based approach to address these challenges is Process Oriented Guided Inquiry Learning (POGIL), in which student teams work on classroom activities specifically designed to help them construct understanding and develop key skills. While POGIL has been widely used and studied in Chemistry and other STEM fields, much less is known about how it is used in Computer Science.
In this study, we examined how faculty adopt POGIL for the first time in introductory CS classrooms. Using qualitative in-depth interviews, we investigated why faculty chose to adopt POGIL and their concerns about using it. Our results suggest that faculty motivations to use POGIL centered around improving student outcomes, including their learning and engagement. However, faculty also had concerns about using POGIL, which ranged from how POGIL impacts the curriculum to logistical and institutional barriers. We discuss the implications of these findings with respect to faculty development and active learning pedagogies.
Software Tools for Mixed-Precision Program Analysis
Floating-point representation and arithmetic remain the predominant mechanism for real-valued computation despite many well-known issues such as roundoff error. As high-performance computing continues to scale, many computational scientists are seeking to reduce precision when possible to accelerate performance and alleviate memory requirements. In this talk, I will describe several projects I have worked on that seek to facilitate such reduction of precision or in some other way provide insights about the accuracy or behavior of floating-point code, including a recent experimental prototype of an end-to-end source-level mixed-precision tuning system. All of the tools discussed are available as open source software.
Applications of Wearable Computing for Physical Therapy and Rehabilitation
Patients undergoing physical therapy or rehabilitation are often given exercises to strengthen muscle groups or joints that have been injured. Typically these exercises are observed in-person at the physician's office and the patient is instructed to continue them at home. Significant challenges exist in physical rehabilitation with patient adherence to their exercise regimen (often they forget to do the activities) and patient access to the physician (because fewer doctors are available). In this talk we will discuss wearable computing approaches to address this challenge through a system that will utilize inertial measurement units to provide real-time feedback to patients at home as they perform their exercises. Furthermore the system will capture those patient movements to transmit clinically relevant information to their physician for review. The wearable computing system addresses both adherence and access by increasing a patient’s self-efficacy when performing activities and enabling the physician to monitor their patient without having to travel to the doctor’s office.
Throughout the talk we will discuss applications of wearable computing for healthcare, related research questions in human-computer interaction and machine learning derived from this research, and an overview of initial results from the Wearable Computing Research Group in the Department of Engineering at James Madison University.
Discussion of the RISC-V Instruction Set Architecture
“The hardware-software interface, embodied in the instruction set architecture (ISA), is arguably the most important interface in a computer system. Yet, in contrast to nearly all other interfaces in a modern computer system, all commercially popular ISAs are proprietary. A free and open ISA standard has the potential to increase innovation in microprocessor design, reduce computer system cost, and, as Moore’s law wanes, ease the transition to more specialized computational devices.” -- Andrew Waterman, University of California, Berkeley
Are you interested in open source? Are you interested in computer architecture? Dr. Weikle will lead a discussion of the RISC-V instruction set architecture that will be featured in her computer architecture class this spring. The RISC-V claims to be extensible and fix many of the problems of the first RISC (Reduced Instruction Set Computer). Come see if you agree. If you would like a paper to look at about the architecture ahead of time, check outside Dr. Weikle's office (205) or send her an email at weikleda@jmu.edu.
Automatic Generation of Mixed-Precision Programs
Floating-point arithmetic is foundational to scientific computing in HPC, and choices about floating-point precision can have a significant effect on the accuracy and runtime of HPC codes. Unfortunately, current precision optimization tools require significant user interaction and few work on the scale of HPC codes due to significant analysis overhead. We propose an automatic search and replacement system that finds the optimal mixed precision configuration given a required level of accuracy. To achieve this, we integrated three existing analysis tools into a system that requires minimal input from the user. If a speedup is found, our system can provide a ready-to-compile mixed-precision version of the original program.
Faculty Advisor: Prof. Lam
Should we auto-generate compilers?
Compilers draw from many areas of CS to solve a single problem: translating from one computer language to another. Some aspects of a compiler (e.g., parsing) are considered “solved” from a theoretical perspective, and others (e.g., optimization) are still open. A recent preprint paper from heavy-hitters in the world of compilers research asserts that the optimization portions of tomorrow's compilers will be autogenerated just like the parsing portions are now. Come discuss this possibility, and along the way you'll learn a bit about SMT solvers and superoptimizers.
What is CS research? (Profs. Lam and Weikle)
Followed by
Faculty Flash Talks (Roughly 5 minutes each)
Join us for an overview of computer science research followed by 5-minute flash talks giving you an overview of ongoing research within the JMU CS department. Whether you are looking to get into research or just interested in learning more, you are welcome!
Less-Java, More Learning: Language Design for Introductory Programming
Less-Java is a new procedural programming language with static, strong, and inferred typing, native unit testing, and support for basic object-oriented constructs.
These features make programming in Less-Java more intuitive than traditional introductory languages, which will allow professors to dedicate more class time to overarching computer science concepts and less to syntax and language-specific quirks.
Fingerprint-protected storage security: the good, the bad, and the ugly
How to use a fingerprint alone to protect a USB drive, a hard disk, or a remote web folder? Unlike passwords or cryptographic keys, whose comparisons are exact, even two consecutive readings of the same finger are not exactly the same; they are just close, making it hard to use a fingerprint to “encrypt” a USB or hard drive.
In this talk, we will discuss the security problem, some insecure implementations (the bad), some elegant theoretical results that may be suitable for the problem (the good), and the still-open road to adapt the theoretical results to solve the problem (the ugly).
A Plethora of Parallelism
From the late 1980s to the early 2000s, computer architects have been using advanced architectural and organizational ideas to contribute significantly to the speed up of computing systems. In the early 2000s, power consumption became an issue that forced architects to develop multi-core solutions. While this has provided increased throughput it has also necessitated that programmers have a higher awareness of how the hardware manages and supports parallelism of different forms. This talk will discuss types of parallelism and introduce the latest attempt to continue performance improvements in computer architecture, domain specific accelerators (DSAs).
Code x Design
Code is a powerful secret weapon for creating 3D-printable designs, and mathematics gives you a universe of beautiful abstract forms to build from. In this talk we’ll discuss how software like Grasshopper, OpenSCAD, Structure Synth, and TopMod can be used to turn mathematical knots, curves, polyhedral wireframes, and procedurally generated forms into physical 3D printed objects. If you’re new to 3D design then this talk will show you ways that you can get started; if you’re an expert then you’ll learn some new tools to add to your design library. We’ll also have lots of cool 3D prints to show and pass around.
Meltdown and Spectre: Complexity and the Death of Security
Meltdown and Spectre are two newly announced computer vulnerabilities. These attacks exploit flaws in how computers are designed and cannot be fixed at this time. That is, these attacks constitute a completely new type of attack by manipulating complex design structures to create unanticipated behavior. This talk will examine how Meltdown and Spectre work, and show how they highlight a long- known principle of computer security: complexity makes devastating attacks possible.
What is CS Research?
We will begin with a brief overview of computer science research given by Profs. Lam and Weikle followed by a series of 5-minute flash talks giving you a window into ongoing research projects at JMU by Profs. Bowers, Lam, Mayfield, Stewart, Taalman, Weikle, and Yang. Whether you are looking to get into research or just interested in learning more, you are more than welcome!
Algorithms ruin everything
I’ll be presenting 2 papers that touch on people’s perceptions of algorithms and their impact on society:
Michael A. DeVito, Darren Gergle, and Jeremy Birnholtz. 2017. “Algorithms ruin everything”: #RIPTwitter, Folk Theories, and Resistance to Algorithmic Change in Social Media. In Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems (CHI '17). ACM, New York, NY, USA, 3163-3174. DOI: https://doi.org/10.1145/3025453.3025659
Tufekci, Zeynep. “Algorithmic Harms beyond Facebook and Google: Emergent Challenges of Computational Agency,” Colorado Technology Law Journal vol. 13, no. 2 (2015): p. 203-218.
Artificial Echolocation Using Deep Neural Networks
Our study explores the effectiveness of bat-inspired echolocation as a practical sensory modality for extracting high-resolution depth information. Traditional ultrasonic depth sensors provide a single scalar depth estimate based on the time delay between an emitted pulse and the detection of an echo. Bats use a similar mechanism to extract depth information. However, bats and other echolocating animals are able to create a sensory percept that is much richer than a single distance measurement. We demonstrate that a deep neural network can be trained to accurately reconstruct two-dimensional depth fields by analyzing the echoes from a single 10 millisecond frequency-modulated chirp.
Preventing Blackout Catastrophe When Learning Sequentially with Elastic Weight Consolidation
Artificial neural networks have become the dominant machine learning approach across a wide range of domains. Significant strides have been made on single-task learning. The problem of sequentially learning multiple tasks has proved to be more challenging. Naive approaches suffer from “catastrophic forgetting”: networks lose accuracy on previously learned tasks when trained on new tasks. The recently introduced Elastic Weight Consolidation (EWC) algorithm is a novel and promising approach that calculates the importance of individual weights to previously learned tasks. A penalty term is introduced that preserves weights in proportion to their importance. However, preserving weights via EWC eventually results in total network failure due to the limited capacity of a fixed-size network, a phenomenon referred to as “Blackout Catastrophe”. Our proposed algorithm addresses this problem by employing EWC until the network capacity has been reached and then increasing the size of the network to accommodate additional tasks. We further investigate the potential of Fisher Information, which is used by EWC to evaluate the importance of each weight for each task, to probabilistically predict when the network needs to expand to prevent blackout catastrophe - an insight which could theoretically allow the network to learn sequentially in perpetuity without the need for human supervision.
ANALYSIS OF PARALLEL IMPLEMENTATIONS OF CENTRALITY ALGORITHMS
Speaker: Patricia D. Soriano
This talk explores parallel implementations of three network analysis algorithms for detecting node centrality: betweenness centrality (BC), eigenvalue centrality (EC), and degree and line importance (DIL). All solutions were written in the C programming language using OpenMP library for parallelization. We evaluated these implementations for accuracy and parallel scaling performance using five example networks. We found that the algorithms accurately reflect different notions of centrality. While DIL performs better in general because it is asymptotically faster than the other two algorithms, BC demonstrates better parallel strong scaling.
TRAVELING SALESMAN: A HEURISTIC SCALING ANALYSIS
Speaker: Garrett Folks
In this talk, we analyze two heuristics that approximate the Traveling Salesman Problem: K-Opt search and ant colony optimization. Our goal was to explore how these heuristics perform when run in parallel on multiple CPU cores as well as using GPU computing. We found that the K-Opt search heuristic showed impressive performance scaling results, especially when executed on a GPU. We also parallelized portions of the ant colony optimization and found good scaling. We conjecture that the ant colony optimization could be greatly improved with the use of GPU computing.
Scheduling, Reproducibility, and Resilience
Dr. Lam and JMU student Garrett Folks spent Summer 2017 doing research at Lawrence Livermore National Laboratory (LLNL) in California. LLNL is a Department of Energy lab that supports a variety of computational research in the broad area of high-performance and scientific computing. This survey talk will give an overview of three ongoing research collaborations between LLNL and academic institutions with a general theme of improving the performance and reliability of numeric code.
Development of voice user interfaces and their impact to the userexperience of mobile applications
Voice user interfaces help users in hands-free situations and whenever graphical interfaces cannot be used. Developing voice user interfaces is a complex process. Besides the technical challenges of understanding and synthesizing human speech, the design of the dialogs is the key part to create a good interface. An easy dialog design includes the right wording and sentence structure. In addition, voice user interfaces can be combined with graphical interfaces to ensure a better usability. This presentation introduces the process of voice user interface development and the design of understandable dialogs. It displays this process through a practical example. This example explains the development of an Android application to experience interactive fiction. The application implements a voice user interface, created during the research for the thesis which this presentation is based on.
A situation-centric, knowledge-driven requirements elicitation approach
Human factors have been increasingly recognized as one of the major driving forces of requirement changes. We believe that the requirements elicitation (RE) process should largely embrace human-centered perspectives, and my work focuses on changing human intentions and desires over time. To support software evolution due to requirement changes, Situ framework has been proposed to model and detect human intentions by inferring their desires through monitoring environmental contexts and human behavioral contexts prior to or after system deployment. Earlier work on Situ reported that the technique is able to infer users’ desires with a certain degree of accuracy using the Conditional Random Fields method. However, new intention identification and new requirements elicitation still primarily depends on manual analysis.
In this talk, I will discuss our attempt to find a computable way to identify users’ new intentions with limited help from human oracle. I will first discuss the feasibility of implementing the concept of Data-Information-Knowledge-Wisdom (DIKW) to bridge the gap between requirements and data pertaining to user behaviors and environmental contexts, and will then introduce our proposed situation-centric, knowledge-driven requirements elicitation approach using the Multi-strategy, Task-adaptive Learning (MTL) method and the Strategic Rationale (SR) model. Our case study shows that the proposed approach is able to identify users’ new intentions, and is especially effective to capture alternatives of low-level tasks. I will also demonstrate how these newly identified intentions can be fused to the existing domain knowledge network using the SR model, and harvest high-level wisdom, in terms of new requirements and design insights.
Adopting CS Principles in a Breadth-First Survey Course
With the recent launch of AP CS Principles in 2016-17, many efforts are currently underway to share curriculum resources and prepare new teachers. The community has primarily focused on high school implementations, which have different situational factors than university courses (e.g., amount of class time). In this seminar, we present the design of a survey course that aligns with CS Principles and also continues the long tradition of breadth-first introductions to computer science at the college level. We describe the instructional strategies, assessments, and curriculum details, providing a model for how to modify existing CS0 courses. We also outline twelve lab activities that support the computational thinking practices and learning objectives of the AP curriculum framework. The course has run successfully for the past four years at two universities and three high schools via dual enrollment. Initial results suggest that the curriculum has a positive impact on student confidence levels and attitudes toward computer science.
Workload Characterization: Some Motivation and Some Math
Workload characterization, including establishing benchmarks, has become a critical part of computer architecture research. At its core, though, it is manipulating very large amounts of data. This talk will discuss some of the motivation behind doing workload characterization and at least one of the mathematical tools, Principal Component Analysis, that computer architects use to do characterization.
Deep Neural Networks vs. Space Invaders: Teaching a Computer to Play Atari
Neural networks have seen a resurgence over the last decade as a result of algorithmic and hardware advances that have enabled the use of deep network architectures. These deep neural networks are now the dominant approach for a wide range of challenging computational problems. Some networks have shown human-level (or better) performance on tasks including visual object recognition, game playing, and translation. In this talk I will introduce artificial neural networks, describe recent progress in deep neural networks, and discuss the application of these networks to problems that involve reasoning and decision making.
Two Talks for the Price of One: Evaluating CS1 Changes and Cognitivism as a CS Ed Foundation
The first part of this talk presents the results of empirical research examining the impact of recent changes made to the introductory CS courses at JMU. We have examined 10 years of data to examine whether the JMU CS introductory sequence is achieving its goal of providing a basis for all students to succeed in the major, regardless of prior experience.
The second part of this talk discusses why more research on CS education is needed. After presenting some counterintuitive and surprising empirical results, we will introduce cognitivism as a theory of learning. This theory uses principles of human cognitive architecture to explain how learning occurs. Cognitive load theory (CLT) builds on these principles to create a basis for efficient and enduring acquisition of biologically secondary knowledge. We will close by discussing the possibility that CLT provides a good framework for effective CS teaching.
Circle packings and Polyhedra
In this talk I will survey some recent developments in the field of circle packing, a theory of circle patterns that are packed together so that neighboring circles are tangent. Circle packings are theoretically interesting as discretizations of analytic functions and provide a method of computing maps between spaces, called quasi-conformal maps, that (approximately) maintain angles. A variety of interesting applications make use of circle packings, including brain anatomy mapping, computational experiments for investigating certain random-walk behaviors in quantum mechanics, and graph drawing. I will introduce the field of circle packing, discuss several of our recent theoretical results, and discuss a new heuristic for computing circle packings that we are currently applying to 3D-printing related applications.
LogicBench: Web Tools for Learning Logic
There are many tools for helping students learn logic. Most tend to be rather clunky and restricted to classical propositional and predicate logic. The LogicBench project is an effort to make more elegant pedagogical tools for a wide range of logics, including modal and temporal logics.
Finding Alternatives to the Best Path
This talk considers several different ways of thinking about alternatives to the best (e.g., shortest, minimum cost) path. It then defines the notion of the Best k-Similar Path and considers a linear optimization formulation of the problem of finding such paths. Next it considers a Lagrangian Relaxation heuristic for solving this problem (and other related problems). Finally, it concludes with some empirical results. Along the way it provides some background (for those who need it) on multivariate calculus and linear optimization.
Office Space and Salami - Automated Floating-Point Program Analysis
Most computers use floating-point arithmetic to perform non-integer computations. However, floating-point representations provide limited precision and it can be difficult to quantify the resulting loss of accuracy. Because of this, computer programmers tend to use the highest available precision and “hope for the best.” My research addresses this situation by providing automated techniques for analyzing and providing insights about the floating-point behavior of computer programs. Most of these techniques operate at the assembly and machine code level, adding instrumentation to detect problems or simulate alternative representations. In this talk, I will describe past efforts as well as my current research, including concrete projects that I would like to work on with undergraduate students.
Cauchy Rigidity of convex c-Polyhedra
A c-polyhedron is a generalization of circle packings on the sphere to circle patterns with specified inversive distances between adjacent circles where the underlying 1-skeleton need not be a triangulation. In this talk we prove that any two convex c-polyhedra with inversive congruent faces are inversive congruent. The proof follows the pattern of Cauchy’s proof o f his celebrated rigidity theorem for convex Euclidean polyhedra. The trick in applying Cauchy’s argument in this setting is in constructing hyperbolic polygons around each vertex in a c-polyhedron on which a variant of Cauchy’s arm lemma can be applied.