Optimization Modeling and Algorithms For Politically Fair Redistricting
Abstract: Political districting in the U.S. is a decennial process of drawing legislative districts for political representation. Each district is a contiguous set of geographical units (such as census tracts) that elects a representative. As an optimization problem, districting is computationally intractable for large practical instances. This talk presents a class of approximation algorithms, called multilevel algorithms, for large-scale districting. Here, political fairness criteria such as the efficiency gap, partisan asymmetry, competitiveness and compactness are optimized using a multi-objective optimization method. The practical implications of adopting the algorithm for real-world districting is presented with a case study in Wisconsin. The results demonstrate the positive role of approximation algorithms in districting, and paves the way for algorithmic transparency in the future of electoral representation.
Bio: Rahul Swamy is a Ph.D. Candidate in Industrial Engineering (Operations Research) at the University of Illinois Urbana-Champaign. His research focus is on combinatorial optimization and game theory which spans applications such as political redistricting, election modeling, evacuation planning, and team formation in online health forums. His work on political redistricting has been recognized by First Place in the INFORMS Service Science Best Paper Award 2019, First place in the INFORMS Poster Competition 2019, and a Finalist at the INFORMS Public Sector Operations Research (PSOR) Best Paper Award 2018.
Before You Know How to Read, How Do You Learn to Code? Studying Tangible Programming and Early Childhood Computational Thinking in the Coding in Kindergarten (CiK) Project
Abstract: More and more young children are learning to code using a variety of commercially available coding toys. The toys are fun and engaging and appear user friendly. However, we do not know how preliterate children learn from these toys. What ways of thinking or problem-solving do the toys support? This seminar examines this issue through two lenses: programming for learning and programming as learning. I discusshow computer science educators are conceptualizing computational thinking (CT) skills and show how this is being studied in early childhood in the Coding in Kindergarten project. Findings highlight the design and use of robot coding toys with a special emphasis on debugging as a key problem-solving skill.
Bio: Dr. Deborah Silvis is a postdoctoral fellow in Instructional Technology and Learning Sciences at Utah State University. She received her PhD in Learning Sciences from the University of Washington’s College of Education. Her work focuses on theories of learning behind design and bridges sociotechnical and sociocultural perspectives on learning. Her work has appeared in journals like theInternational Journal of Computer Supported Collaborative LearningandLearning, Media, and Technology.
Web Programming - You Are All Doing It Totally Wrong
Abstract: In this talk we will revisit Roy Fielding's famous dissertation describing the network architecture of the Web. In particular, we will focus on the uniform interface constraint of REpresentational State Transfer (REST) and Hypermedia as the Engine of Application State (HATEOAS). We will discuss the history of REST as it evolved from a description of the web architecture to a prescriptive set of rules for API design (first in XML, then in JSON), and why this was misguided. We will talk about current web application architectures, particularly focusing on Single Page Application (SPA) frameworks, and an alternative approach, htmx, which matches the original REST description more faithfully.
Bio: CarsonGross is an instructor in the CS Department at MSU and is the creator of intercooler.js and htmx
Golden Helix - Clinical DNA Analytics
Abstract: Dr. Andreas Scherer is CEO of Golden Helix. The company has been delivering industry-leading bioinformatics solutions for the advancement of life science research and translational medicine for over two decades. Its innovative technologies and analytic services empower clinicians and scientists at all levels to derive meaning from the rapidly increasing volumes of genomic data produced from next-generation sequencing and microarrays. With its solutions, hundreds of the world’s hospitals, testing labs, academic research organizations, and governments are able to harness the full potential of genomics to identify the cause of disease, develop genomic diagnostics, and advance the quest for personalized medicine. Golden Helix products and services have been cited in thousands of peer-reviewed publications. In this talk, Dr. Scherer will give an overview of Golden Helix and some of key problems the company is solving to bring advanced analytics solution to the market.
Bio: Dr. Scherer has managed Golden Helix since 2013. Prior to that he held executive roles
in public and private companies managing P&L's in excess of $100 Million. He holds
a PhD in computer science from the University of Hagen, Germany, and a Master of Computer
Science from the University of Dortmund, Germany. He is author and co-author of over
40 international publications, recipient of multiple NIH grants and has written books
on project management, the Internet, and artificial intelligence. His latest book, “Be
Fast Or Be Gone”, is a prizewinner in the 2012 Eric Hoffer Book Awards competition,
and has been named a finalist in the 2012 Next Generation Indie Book Awards.
Publications and Grants: https://orcid.org/0000-0002-55
An Optimal, Practically Fast Algorithm for Selection on Cartesian Products
Abstract: Given two lists X and Y, generating the top values from their Cartesian product, X_i + Y_j, is an important problem with applications in chemistry and statistics. A novel algorithm is presented, which exactly computes the top k values from X+Y in optimal time. The new algorithm relies on a "layer-ordered heap," which generalizes sorted ordering to lesser degrees of order. The new method relies only on linear-time selection (e.g., median-of-medians) and is fast in practice. Given a median-of-medians implementation, the proposed method can be implemented in roughly one page of python code.
Bio: Oliver Serang received his B.S. in Computer Engineering from North Carolina State University and his Ph.D. in Genome Sciences from the University of Washington. He has been a visiting scientist in Sao Paulo, a research fellow at Harvard Medical School and Boston Children's Hospital, a Software Research Scientist at Thermo Fisher Scientific, a lecturer in Informatik at the Universität Bremen, and a professor of Informatik at Freie Universität Berlin. He is now on the faculty of Computer Science at the University of Montana. He lives on a mountain with his Finnish masonry stove.
Adventures in Quantum Computing
Abstract: Quantum computing is a novel form of computing that has the potential to perform certain computations much more efficiently than classical algorithms. This talk aims to give a glimpse of this emerging field by exploring a few examples. After a brief introduction to quantum computing, a couple vignettes will be presented on using quantum computers to solve problems related to partial differential equations. One of these vignettes shows fairly promising results on a Noisy Intermediate-Scale Quantum (NISQ) computer without any theoretical guarantees of strong performance. The other can potentially leverage theoretically efficient algorithms, but shows poor performance on NISQ hardware. The remainder of the talk will focus on an approach to matrix factorization that utilizes quantum annealing. This approach enables the factorization of arbitrarily large matrices despite the small number of qubits available. The number of qubits only limits the rank of the factorization.
Bio: Daniel O'Malley is a scientist in the Computational Earth Science group at Los Alamos National Laboratory (LANL). He studied at Purdue University, receiving a B.S. degree in computer science and mathematics, an M.S. in mathematics and a Ph.D. in applied mathematics. His research interests include computational science (with an emphasis on subsurface flow and transport), quantum computing, and machine learning. He has won numerous awards including a LANL Early Career Research award, a LANL Director's Postdoctoral Fellowship, the InterPore-Fraunhofer Award for Young Researchers from the International Society for Porous Media, a Charles C. Chappelle Fellowship from Purdue University, and the Meyer E. Jerison Memorial Award in Analysis from the Department of Mathematics at Purdue University.
An Introduction to Aurora and A Few Thoughts on Sensors for Autonomous Vehicle
Abstract: Aurora is developing the software, hardware and data services to safely deploy autonomous vehicles. With the acquisition of Blackmore in 2019, Aurora added a location in Bozeman. The first half of this presentation will introduce Aurora and explain the importance of the lidar technology developed in Bozeman. From there, we'll discuss the unexpected roots of various autonomous vehicle sensor technologies and also highlight the broad set of career opportunities for CS students to consider in the robotics industry.
Bio: Stephen Crouch is a technologist focused on applied sensing problems in radar and lidar. Mr. Crouch holds a BS/MS in Physics from Stevens Institute of Technology where he focused on integrated photonics and an MS in Physics from Montana State University where he completed a thesis on synthetic aperture lidar. He began his career at Delphi Automotive focusing on automotive radar technology and later returned to Bozeman to work at Bridger Photonics. As CTO and co-founder of Blackmore, he led the company through rapid growth and ultimately acquisition. Mr. Crouch now works with novel sensor technologies at Aurora.
An Improved Belief Rule-Based Expert System with an Enhanced Learning Mechanism
Abstract: Belief rule-based expert systems (BRBESs) are widely used in various domains which provide an integrated framework to handle qualitative and quantitative data by addressing several kinds of uncertainty. The correctness of the data significantly affects the accuracy of the BRBESs. Learning plays an important role in BRBESs to upgrade their knowledge base and parameters values, necessary to improve the accuracy of prediction. In addition, comparatively larger datasets hinder the accuracy of BRBESs. Therefore, my research focuses on four different aspects of BRBESs, namely, the accuracy of data, multi-level complex problem, learning of BRBES, and accuracy of prediction for comparatively large dataset.
Bio: Dr. Raihan Ul Islam is a postdoctoral researcher at Luleå University of Technology, Sweden. He did both his M.Sc. and Ph. D. degree in Computer Science from Luleå University of Technology, Sweden. Previously he worked as a software engineer at NEC Laboratories Europe, in the Context-aware Services (CAS) and Smart Environments Technologies Group. His research interests also include Mobile Edge Computing, 5G, Machine Learning, M2M Communication, Smart Homes and Cities, Mobile Systems, and Pervasive and Ubiquitous Computing.
Data Science in Industry
Abstract: Dr. Neal Richter will give a tour of data science in anapplied industry setting. Drawing from his experience and those ofworking data scientists he will present what a new undergrad or gradwill typically work on, methodologies and various concerns on scalingstatistical and machine learning algorithms.
Bio: Dr. Richter (MSU CS alumni BS 1998, PhD 2010) has worked inindustry since 1996 and since 1999 focused on data science, ML and AIapplications. He has served as staff software scientist at RightNowTech, a series of startups, as CTO of Rubicon Project and RakutenMarketing. He is now Chief Scientist at SpotX, a platform focused onvideo advertising for streaming TV platforms.
Computing Haplotype Blocks to Investigate Natural Selection
Abstract: Recent work provides the first method to measure recent natural selection across an entire genome from population data that scales to a large number of individuals. A key part of the computation is finding maximal blocks of genetic variation that are conserved across individuals. In this talk, I discuss our work to extend the problem of computing these blocks in two ways. First, we consider the case when the genomes of individuals may contain unknown regions due to sequencing errors. Then, we consider the problem setting where more complex variation across a genome is allowed (a pangenome), versus the original setting where only very simple variation can be represented.
Bio: Lucy Williams began her PhD at MSU in Computer Science in the fall of 2017. Her advisor is Brendan Mumey, and she is interested in data structures and algorithms for computational biology, particularly for the assembly and analysis of DNA sequence data. She received her B.S. in Applied Math from the University of Washington in 2014.
Pheasants Forever, MSU, and John Deere’s Joint Effort to Harness Precision Agriculture for the Benefit of Montana Wildlife
Abstract: Pheasants Forever’s vision for precision agriculture in Montana is to demonstrate, through the use of technology and collaboration, that farmers can increase their profitability while simultaneously providing benefits for soil health, water, and wildlife habitat. MSU is collaborating with them and with John Deere to map crop yields at no cost to the farmer, where the farmers agree to have us do so. We examined 16,000 acres of cropland, primarily wheat and barley, for two grain farmers in the pilot year of the project. The examination of yields from individual fields is quite simple and uses mapping and analysis software in the John Deere cloud that is available to any farmer who is a John Deere customer. Their yield and other data are automatically uploaded to the cloud databases directly from their farm equipment. However, they often do not either have the time or expertise to complete major yield mapping projects. Where our analyses show that yields are consistently low, the door is opened to having discussions with the farmer about alternative management options for that ground. We talk about turning “red acres” to “green”. I.e., by taking portions of a field out of crop production that are losing money, the producer can actually increase the overall profitability of their entire field. Then, wildlife habitat biologists can suggest more profitable options for those acres, such as joint management of pasture for cattle and nesting upland gamebirds, in place of growing grain that requires greater cost of input without an accompanying increase in revenue. We recognize that yield is a simplistic surrogate for profitability and return on investment analysis, but the operators with whom we have visited understand that.
Bio: Rick Sojda has been a part-time Research Professor in GSOC since 2013 after he retired from a career doing scientific research and wildlife management for the U.S. Fish and Wildlife Service and the U.S. Geological Survey for 34 years. He has degrees from Cornell University, Iowa State University, and Colorado State University. His doctoral work entailed using AI to build a multiagent system to model migration of trumpeter swans using interacting expert systems and a queuing model of swan migration. He was elected a fellow of the International Environmental Modelling and Software Society in 2011.
Abstract: This seminar will provide new and continuing graduate students with (1) useful information, (2) an opportunity to meet other students, staff and faculty, and (3) an opportunity to ask questions.
Date/Time: Wednesday, May 13 2:00 p.m - 3:00 p.m.
Speaker: Ambika Murali
Abstract: This session will cover the complexities of launching a new product or scaling a product by discussing the role of a Product Manager (PM) within technology. We'll discuss what is product management, how ideas and concepts become reality, and how a PM works cross-functionally to launch and scale products to serve users' pain points. Takeaways for attendees include
- An understanding of product management & its collaboration with engineering
- How to pitch ideas and prototype
- How to optimize the backend processes for speed
Biosketch:Born and raised in India, Ambika's passion for global education and culture has led her to live and work in 5 countries. Academically trained in data science, Ambika currently works as a Technical Group Product Manager in the Bay Area for US Bank. She is responsible for launching their new digital experience for checking accounts (mobile and web), with a keen focus to achieve the fastest time-to-fund a digital checking account in the US. Previously, Ambika worked at SoFi and launched multi-billion dollar new product lines and helped deploy machine learning models which attained SoFi the fastest time-to-fund a personal loan in the US. Prior to tech, Ambika worked in the financial industry for Goldman Sachs, Deutsche Bank, Central Bank of Chile, and the International Monetary Fund. Ambika is committed to the causes of education and theadvancement of women. Ambika is an Advisory Board Member of the School of Computing at Montana State University and Head of Programming for a nonprofit called Advancing Women in Product (AWIP).
Date/Time: Monday, April 22 4:10 p.m - 5:00 p.m.
Facilitator: John Paxton
Abstract: At the end of every academic year, we celebrate the accomplishments of members of the Gianforte School of Computing. Join us for this year's celebration where we will reflect on our accomplishments and present awards.
Continuum and Stochastic Models for Transcription on a Crowded Gene
Date/Time: Monday, March 9 4:10 p.m - 5:00 p.m.
Location: Barnard Hall 108
Speaker: Lisa Davis
Abstract: In fast-transcribing prokaryotic genes, such as an rrn gene, many RNA polymerases (RNAPs) transcribe the DNA simultaneously. Active elongation of RNAPs involves periods of fast forward motion that are often interrupted by pauses. In some literature, this has been observed to cause RNAP traffic jams. However, other studies indicate that elongation is faster in the presence of multiple RNAPs than elongation by a single polymerase. Over the course of this research project, we have considered several mathematical models to capture the essential behaviors known to this phenomena. I will give a brief overview of the essential biological quantities of interest, and the remainder of the talk will focus on an overview of two mathematical models that have been proposed. The first is a continuum model taking the form of a nonlinear conservation law PDE where transcriptional pausing is incorporated into the flux term with a piecewise continuous density-velocity relationship. The velocity relation is parametrized according to the user-specified (or randomly generated) spatial locations and time duration of the pauses. The second model is a stochastic one that is based on the classical TASEP model but with added complexity to account for the interactions among neighboring RNAPs that can influence local elongation velocities. I'll mention the algorithms that were used for model simulation for a series of parameter studies. If there's time, I'll discuss future directions to combine the lessons learned from previous models into the development of a specific second order PDE formulation which allows for an adaptive density-velocity relationship.
Strategies for Resisting the Racial Injustice of Predictive Policing
Date/Time: Monday, March 2 4:10 p.m - 5:00 p.m.
Location: Barnard Hall 108
Speaker: Bonnie Sheehey
Abstract: The recent use of predictive technologies in fields of criminal justice like policing has become a growing site of concern for activists, theorists, and technicians. Prevalent among these concerns is the way in which these technologies reinforce, and potentially even exacerbate, conditions of racial injustice that have long plagued the criminal justice system in the U.S. One common response to this problem is to call for more transparency in the deployment of predictive policing technologies. An ethics of transparency, however, is insufficient to address the racial harms of predictive policing. What is needed, I argue, is a set of ethical strategies that aim at resisting the racial injustice of predictive policing. By thinking of ethics in terms of resistant practices, we can begin to consider a notion of responsibility that holds us and the technologies we bind ourselves to accountable for the harms created by this bond.
Analyzing time series data as graphs
Date/Time: Monday, February 24 4:10 p.m - 5:00 p.m.
Location: Barnard Hall 108
Speaker: Bree Cummins
Abstract: The timing of the extrema of an experimental time series (or its derivative) contains information about the underlying dynamical system. Time series often contain significant measurement errors that interfere with the ability to extract information from the data. I will discuss a method for characterizing a time series for any assumed level of measurement error ε by a sequence of intervals, each of which is guaranteed to contain an extremum for any function that ε-approximates the time series. I will show that there is a well-defined partial order on these intervals across a collection of time series comprising a dataset. These partial orders can be thought of as graphical representations of the time series data. My collaborators and I have used these graphs to assess similarity between experimental replicates and between gene expression in different malaria strains. To do this, we defined a metric on graphs based on the maximum common edge-induced subgraph problem for directed graphs (DMCES). I will discuss that the definition fulfills the criteria for a metric and introduce a newly defined structure called the labeled line graph to show that DMCES can be reduced to the maximum clique-finding problem.
Hyperspectral Image Classification with Low-Cost 3D-2D Convolutional Neural Networks
Date/Time: Monday, February 10 4:10 p.m - 5:00 p.m.
Location: Barnard Hall 108
Speaker: Giorgio Morales
Abstract: Hyperspectral images provide us a useful tool for extracting complex information when visual spectral bands are not enough to solve certain tasks. However, usually, processing them is computationally expensive due to the great amount of both spatial and spectral data they incorporate. In this seminar, I will present a low-cost convolutional neural network designed for hyperspectral image classification. Its architecture consists of two parts: a series of densely connected 3-D convolutions used as a feature extractor, and a series of 2-D separable convolutions used as a spatial encoder. We show that this design involves fewer trainable parameters compared to other approaches, yet without detriment to its performance. What is more, we achieve comparable state-of-the-art results testing our architecture on four public remote sensing datasets: Indian Pines, Pavia University, Salinas, and EuroSAT; and a dataset of kochia leaves (Bassia scoparia) with three different levels of herbicide-resistance.
Developing chemistries to produce nonlinear signal gain
Date/Time: Monday, February 3 4:10 p.m - 5:00 p.m.
Location: Barnard Hall 108
Speaker: Stephanie McCalla
Abstract: Kinetic measurements to quantify RNA and DNA are easily shifted by reaction conditions such as enzyme efficiency, salt concentrations, or inhibitors that are ubiquitous in biological samples. As a result, it is difficult to determine if a signal is uninhibited and originates from spurious amplification, similar nucleic acid sequences, or from the target of interest. Digital electronics revolutionized computing by creating a robust method to process data: one that relied on digital (on/off) signals that would not be affected by drifting signal amplitude and noise. An analogous method is gaining traction in the field of nucleic acid measurement, where individual molecules are amplified and detected in a multitude of small microfluidic wells. In this seminar, I will discuss our work on designing new reaction chemistries with highly non-linear signal output that are ideal for use in digital systems. These high-gain outputs can be applied to digital detection of specific disease-associated molecules, DNA computing, and DNA-based logic gates.
Designing and Implementing Culturally Responsive Computing Curricula and Pedagogy in the Elementary STEM Teaching Integrating Textiles and Computing Holistically (ESTITCH) Project
Date/Time: Thursday, January 30 11:00 a.m - 11:50 a.m.
Location: Roberts 113
Speaker: Kristin A. Searle
Abstract:As computer science education becomes more and more prevalent in K-12 classrooms, access and equity remain serious issues. We must design and implement computing curricula that make sense to a wide range of learners from a variety of backgrounds. Through professional development, we must also give teachers the tools to engage their students and the computing content in meaningful ways. One technology that has been shown to be especially promising for broadening participation in computing is electronic textiles (e-textiles). E-textiles bring together familiar aspects of fabric crafts with electronic components that are sewable and programmable, allowing for electronics to be embedded in items like clothing, bags, and decorative pillows. In the Elementary STEM Teaching Integrating Computing and Textiles Holistically (ESTITCH) project, we leverage the affordances of e-textiles for integrated science, social studies, and computing in grades 3-6. Project ESTITCH is a curriculum and professional development project serving elementary teachers and students in two rural school districts in Utah. Drawing on the qualitative and design-based aspects of the work, I discuss the design of curriculum and professional development, paying particular attention to the aspects that are culturally responsive, and share stories from our first year of implementation. Findings highlight both the affordances of culturally responsive computing and also the areas where we have more work to do.
Representing Shapes Using Topological Descriptors
Date/Time: Monday, January 13 4:10 p.m - 5:00 p.m.
Location: Barnard Hall 108
Speaker: Samuel Micka
Abstract: Techniques developed in the field of topological data analysis have grown increasingly popular in recent years for gathering insight into complex data. One example is the use of topological descriptors, such as the persistence diagram (PD), for representing and discriminating between different shapes. In this talk, we offer insight into the algorithms behind generating a set of descriptors which encode all of the geometric and topological information of a shape. Furthermore, we discuss the benefits of using PDs for representing shapes and describe efficient methods for comparing PDs to one another for the purposes of shape differentiation.
Seminars from 2019.