2021 Computer Science Colloquium Series


Talks are held via Zoom.


Machine Learning, Networking, and Computer Science Education

R. Benjamin Shapiro

Wednesday, November 17, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

Today’s world is full of technologies that leverage machine learning (ML) and networking. Young people see and use many of them every day, including voice assistants like Alexa and Siri, messaging applications, multiplayer games, video filters, and even autonomous vehicles. But the ways we teach computing to young people largely ignores these technologies, and as such does not enable their creative agency and critical capacity with respect to them. In this talk I will describe work my research group has done, together with educator and researcher partners, to create new tools for youth to apply and learn about ML and networking within creative projects, and to study learning with these systems. Our work illustrates that nominally advanced topics in computing (like ML) can, in fact, be part of introductory computing, and that integrating them into beginning computing courses and programs can enable young people to deeply leverage their prior knowledge about dance, athletics, and music.

Bio:

R. Benjamin Shapiro is an Assistant Professor in the Department of Computer Science at the University of Colorado Boulder. He is also faculty, by courtesy, in Learning Sciences and Human Development (School of Education) and Information Science (College of Media, Communication, and Information). He leads the Education team in Apple’s AI/ML organization, which both conducts research and development of new technologies and partnerships for ML education and supports learning about AI and ML by Apple employees worldwide. He holds a B.A. in Independent Studies from the University of California San Diego, and a PhD in Learning Sciences from Northwestern University. He was a postdoc in the Wisconsin Institutes for Discovery, at the University of Wisconsin Madison.


Reconstructing Random Geometric Graphs

Varsha Dani

Wednesday, November 10, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

A unit-disk graph is obtained by taking a finite collection of points in the plane as vertex set, and putting an edge between any two vertices whose Euclidean distance is at most one. The reconstruction problem for such a graph asks, given the adjacency matrix of the graph as input, to approximately recover the coordinates of each vertex, up to symmetries. How accurately can this be done? I will present some recent progress on this problem under the additional assumption that the collection of points is chosen at random from a compact convex region of the plane. I'll also briefly discuss how to extend the same ideas to higher dimensions and arbitrary manifolds of bounded curvature.

Bio:

Varsha Dani is an Assistant Professor at the Rochester Institute of Technology, where she pretends to be a Computer Scientist despite being a Mathematician at heart. After getting a Ph.D in Computer Science from the University of Chicago, she spent many years in Albuquerque, dabbling in many things, including parenting, travel and hiking, edu-tainment, and, of course, mathematics ...er theoretical computer science. In her spare time she likes to... wait, what spare time?


Creative Learning through Expressive Making

HyunJoo Oh

Wednesday, November 03, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

Tools shape the way we think, make, and learn. As a designer and design researcher, I build tools that integrate everyday craft materials with computing, and study how the tools can engage and support other designers in investigating new expressive and technical possibilities. I employ familiarity and accessibility of everyday tools and materials to empower a broad range of designers, from professional designers and hobbyist makers to K-12 educators and students, to expand their capabilities for creative learning through making. In this talk, I’ll present my recent projects, developing a kit of materials for inclusive computing education and techniques of DIY sensing technology for design prototyping.

Bio:

HyunJoo Oh is an Assistant Professor with a joint appointment in the School of Industrial Design and the School of Interactive Computing at Georgia Tech, where she directs the CoDe Craft group (www.codecraft.group/). Her team investigates how computing technologies can extend and transform everyday craft materials and how these integrations can broaden creative possibilities for designers. She received her PhD in Technology, Media, and Society from University of Colorado Boulder and Master’s degrees in Entertainment Technology from Carnegie Mellon University and Media Interaction Design from Ewha Womans University.


Distributed connectivity and the k-out random graph conjecture

Valerie King

Wednesday, October 27, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

We consider the following problem. Each node in a graph has a distinct ID and each knows only the IDs of its neighbors. Suppose it can send one message to a referee who must determine the graph’s connected components. The graph sketching technique described by Ahn, Guha and McGregor in 2012 gives a method which requires only O(log^3 n) bits to be sent by each node, to compute the solution with high probability, and this is tight, according to a recent result of Nelson and Yu. However this method requires public randomness.

We began by investigating the one-way communication cost of this problem when there is private randomness, and ended up proving a surprising lemma about sampling in graphs and connectivity. This is joint work with Jacob Holm, Mikkel Thorup, Or Zamir, and Uri Zwick which appeared in FOCS 2019.

Bio:

Valerie King is an American and Canadian computer scientist who works as a professor at the University of Victoria. Her research concerns the design and analysis of algorithms; her work has included results on maximum flow and dynamic graph algorithms, and played a role in the expected linear time MST algorithm of Karger et al. King graduated from Princeton University in 1977. She earned a law degree (Juris Doctor) from the University of California, Berkeley in 1983, and became a member of the State Bar of California, but returned to Berkeley and earned a Ph.D. in computer science in 1988 under the supervision of Richard Karp with a dissertation concerning the Aanderaa–Karp–Rosenberg conjecture.

She became a Fellow of the Association for Computing Machinery in 2014.


Checking In With All Stakeholders: Reflecting on Research Methods While Designing an Electronic Toolkit with Older Adult Crafters

Katie Siek

Wednesday, October 20, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

Computing researchers develop technology for older adults – from innovating how to make commodity technology more accessible to designing aging in place systems for various stakeholders. Older adults are often consulted by research teams, but not integrally involved in the design process. The research teams themselves are often small with internal epistemologies that impact how research is conducted and analyzed.

In this talk, I reflect on how we iteratively developed an electronic toolkit with older adult crafters while collaborating with outside research teams. During our in-person and remote design workshops, we had to check in with older adult advisors to better understand how we could organize mutually beneficial design workshops that would also help us understand how to scaffold activities to build up their electronics knowledge. Likewise, we checked in with other research teams who integrated their own research approaches into our design process. These collaborations made us aware of new empirical understandings about our methods and how we studied older adults. I conclude with considerations researchers should take when designing for older adults and how to check-in on their own research practices.

Bio:

Katie Siek is a professor and chair of Informatics at Indiana University Bloomington. Her primary research interests are in human computer interaction, health informatics, and ubiquitous computing. More specifically, she is interested in how sociotechnical interventions affect personal health and well being. Her research is supported by the National Institutes of Health, the Robert Wood Johnson Foundation, and the National Science Foundation including a five-year NSF CAREER award. She has been awarded an NCWIT Undergraduate Research Mentoring Award (2019), a CRA-W Borg Early Career Award (2012), and Scottish Informatics and Computer Science Alliance Distinguished Visiting Fellowships (2010 & 2015). Prior to returning to her alma mater, she was a professor for 7 years at the University of Colorado Boulder. She earned her PhD and MS at Indiana University Bloomington in computer science and her BS in computer science at Eckerd College. She was a National Physical Science Consortium Fellow at Indiana University and a Ford Apprentice Scholar at Eckerd College.


Theory and Performance of Backoff Algorithms

Maxwell Young

Wednesday, October 13, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

Binary exponential backoff (BEB) is a decades-old algorithm for coordinating access to a common system resource, such as a shared communication channel. In modern networks, BEB plays a crucial role in WiFi and other wireless standards. Despite this track record, well-known theoretical results indicate that under bursty traffic, BEB has poor performance, and superior algorithms exist.

We investigate a challenging case for BEB: a single burst of packets that simultaneously contend for access on a wireless channel. Using Network Simulator 3, we incorporate into IEEE 802.11g several newer algorithms that have theoretically-superior performance guarantees. Surprisingly, we discover that these newer algorithms underperform BEB.

Investigating further, we identify as the culprit a common abstraction regarding the performance impact of collisions; that is, when two or more devices send at the same time, resulting in failed communication. Our experimental results are complemented by analytical arguments that the number of collisions is an important metric to optimize. We propose a new theoretical model that accounts for the cost of collisions, and we derive new asymptotic bounds on the performance of BEB and some newer backoff algorithms.

Bio:

Maxwell Young is an Assistant Professor at Mississippi State University. Previously, he completed his MS under the supervision of Jared Saia, received his PhD at the University of Waterloo, Canada, and did postdocs at the National University of Singapore and the University of Michigan, Ann Arbor. His work focuses on algorithm design and analysis for large-scale networks. Outside of research, Max fights a losing battle to civilize his children, and he occasionally makes fountain pens.


Expanding Digital Boundaries in Physical Space

John-Mark Collins

Wednesday, October 6, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

Our world is becoming more and more infused with digital pieces - what would this look like if we artists and creatives pushed something together, versus allowing the large tech companies to guide the deployment? How can we start in our galleries and facilities in a way that will empower the new digital reality to be more human-focused?

Electric Playhouse is an all-ages dining, gaming, and recreation wonderland that requires no goggles or equipment to transport you to another reality. The immersive projection-based play arena has 18 interactive areas that change constantly. Electric playhouse is located in Albuquerque, NM. 

Bio:

John-Mark is the co-founder and creative director of Electric Playhouse. He is a creative problem solver whose passion lies in using technology to augment the real world in beautiful and engaging ways. John-Mark received his BS in Computer Engineering and MBA from the University of New Mexico. In addition to technology, John-Mark has studied in both Art and Architecture and has participated in several public electronic arts exhibitions. He has worked with a variety of clients, including Coca-Cola, Starbucks, HP, Intel, several Smithsonian Institutions as well as Fermi, SLAC, and Sandia National Laboratories. The drive behind Electric Playhouse stems from the wonder and awe shown on a daily basis by his two young daughters, Lola and Mila.


Designing the Hybrid Body

Cindy Hsin-Liu Kao

Wednesday, September 29, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

Sensor device miniaturization and breakthroughs in novel materials are allowing for the placement of technology increasingly close to our physical bodies. However, unlike all other media, the human body is not simply another surface for enhancement - it is the substance of life, one that encompasses the complexity of individual and social identity. The human body is inseparable from the cultural, the social, and the political, yet technologies for placement on the body have often been developed separately from these considerations, with an emphasis on engineering breakthroughs. The Hybrid Body Lab investigates opportunities for cultural interventions in the development of technologies that move beyond wearable clothing and accessories, and that are purposefully designed to be placed directly on the skin surface. By hybridizing miniaturized robotics, machines, and materials with cultural body decoration practices, the Hybrid Body Lab investigates how technology can be situated as a culturally meaningful material for crafting our identities. Through these hybrid body interfaces, we investigate opportunities for designing new modes of self-expression, and also ways to interact with others and our surrounding environments.

Bio:

Cindy Hsin-Liu Kao is an Assistant Professor in the College of Human Ecology and graduate field faculty in Information Science and Electrical and Computer Engineering at Cornell University, where she founded and directs the Hybrid Body Lab. Her research practice themed Hybrid Body Craft blends aesthetic and cultural perspectives into the design of on-body interfaces. She also creates novel processes for crafting technology close to the body. Her research has been presented at leading computer science conferences and journals (ACM CHI, UbiComp/ISWC, TEI, UIST, IEEE Pervasive Computing) while receiving media coverage by CNN, TIME, Forbes, Fast Company, WIRED, among others. Her work has been exhibited and shown internationally at the Pompidou Centre, the Boston Museum of Fine Art, Ars Electronica, New York Fashion Week. Among her awards include the NSF CAREER Award, and several Honorable Mention/Best Paper Awards in top computer science conferences (ACM CHI, UIST, DIS and ISWC). The design community has also recognized her lab's work with the Fast Company Innovation by Design Award Finalist, an Ars Electronica STARTS Prize Nomination, and the SXSW Interactive Innovation Award. Dr. Kao holds a Ph.D. from the MIT Media Lab, along with a Master's degree in Computer Science; and two Bachelor’s degrees in Computer Science and in Technology Management, all from National Taiwan University.


On Power of Choice for k-colorability

Diksha Gupta

Wednesday, September 22, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

In studying the evolution of a random graph, one starts from an empty graph and adds edges one by one, generating each one independently and uniformly at random. An object of interest associated with the resultant graph is the size of the edge set at which some property changes. For instance, if we are interested in the k-colorability of the random graph, there will eventually be an edge that renders it non-k-colorable. Earlier work on the “power of choice” to affect the outcome of random processes has investigated questions like load-balancing in balls and bins models, scheduling, routing, and many more. When applied to the random graph process, it results in the r-choice Achlioptas process, wherein random edges are generated r at a time, and an online strategy selects one for inclusion in a graph. In this talk, we investigate the problem of whether such a selection strategy can shift the k-colorability transition; that is, the number of edges at which the graph goes from being k-colorable to non-k-colorable.

Bio:

Dr. Diksha Gupta is a Research Fellow at the National University of Singapore under Prof. Seth Gilbert. She obtained a Ph.D. in Computer Science from the University of New Mexico, USA, under the advisement of Prof. Jared Saia in Fall 2020. She holds an M.S. in Computer Science from UNM and an M.Tech. in Computer Science and Engineering from IIT Roorkee, India. Her current research focuses on designing provably secure, scalable, and efficient protocols for distributed systems. In general, topics related to Randomized Algorithms, Distributed Graph Algorithms, Algorithmic Game Theory, and Biologically-inspired Algorithms spike her interest. In her free time, she likes solving mathematical puzzles, reading sci-fiction, and traveling. As a grad student at UNM, she has served as a member for various student and departmental organizations – UNM CS Advisory Board (Graduate Representative), Computer Science Graduate Student Association (President), and UNM Women in Computing (Vice President). She will be joining IBM Innovations Lab Singapore as a Research Scientist in November this year.


Theatre of the Car

Wendy Ju

Wednesday, September 15, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

The advent of autonomous vehicles is both exciting and alarming. The success or failure of such systems will very much depend on the driver-vehicle interaction: whether people have a good assessment of what the car perceives and is likely to do, and how they might respond to different situations. In my research lab, we are looking at how people will interact with the cars and robots of tomorrow. By using simulation technologies and design techniques, we can prototype and test interfaces with real people to understand how best to design our future interactions with automation.

Bio:

Wendy Ju is an Associate Professor at the Jacobs Technion-Cornell Institute at Cornell Tech and the Technion and in the Information Science field at Cornell University. Her work in the areas of human-robot interaction and automated vehicle interfaces highlights the ways that interactive devices can be designed to be safer, more predictable, and more socially appropriate. Professor Ju has innovated numerous methods for early-stage prototyping of automated systems to understand how people will respond to systems before the systems are built. She has a PhD in Mechanical Engineering from Stanford, and a Master’s in Media Arts and Sciences from MIT.


A Class of Trees Having Near-Best Balance: a Competitor to Divide-and-Conquer

Laura Monroe

Wednesday, September 8, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

Full binary trees naturally represent commutative non-associative products. There are many important examples of these products: finite-precision floating-point addition and NAND gates, among others. Balance in such a tree is highly desirable for efficiency in calculation. The best balance is attained with a divide-and-conquer approach. However, this may not be the optimal solution, since the success of many calculations is dependent on the grouping and ordering of the calculation, for reasons ranging from the avoidance of rounding error, to calculating with varying precision, to the placement of calculation within a heterogeneous system.

We introduce a new class of computational trees having near-best balance in terms of the Colless index from mathematical phylogenetics. These trees are easily constructed from the binary decomposition of the number of terms in the problem. They also permit much more flexibility than the optimally balanced divide-and-conquer trees. This gives needed freedom in the grouping and ordering of calculation and allows intelligent efficiency trade-offs.

Bio:

Laura Monroe (HPC-DES) is a research scientist at Los Alamos National Laboratory. She received her Ph.D. in Mathematics and Computer Science from the University of Illinois at Chicago, where she studied the theory of error-correcting codes. She worked at NASA Glenn following graduation and joined LANL in 2000.

She was formerly the project leader for the laboratory’s Production Visualization project, where her last project was leading the rebuild of LANL’s state-of-the-art visualization corridor, including large-scale visualization systems, large virtual reality theaters and networking and systems for desktop visualization. She now works at LANL’s Ultrascale Systems Research Center in the field of novel computing, in particular probabilistic computing for high-performance applications. She has published in the fields of probabilistic computing, resilience, error-correcting codes, combinatorics and visualization, and her interests are in those fields as well as in the mathematical bridge between the computer as physical object and as ideal system.

She has received several Defense Program Awards of Excellence and several LANL Distinguished Performance awards, both as team leader and team member, and received an R&D 100 award in 2006 as part of the PixelVizion team. She was named one of the 2019 NM Technology Council Women in Technology awardees.


Dynamical system models for politics and voting

Vicky Chuqiao Yang, Santa Fe Institute

Wednesday, September 1, 2021, 2:00 PM

Join via Zoom: https://unm.zoom.us/j/95844274177
Passcode: 856723

Abstract:

The recent US political landscape brings many puzzling questions. For example, the two major parties have become increasingly polarized since the 1960s, while most voters maintained moderate policy positions. What can lead to the disconnect between the parties and the voters? Also, a sizable proportion, often the majority, of the voting population is uninformed about facts relevant to their voting decisions, such as policies proposed by the candidates. Can such a voting body deliver good collective decisions? In this talk, I will summarize research projects that address these complex issues. These projects leverage dynamical-systems models, recent findings in psychology, and data analysis. This approach takes into account the impact of multiple, complex, and often non-linear factors, and aim to give a coherent understanding of complex social phenomena.

Bio:

Vicky Chuqiao Yang is a fellow at the Santa Fe Institute. Her research uses mathematical tools to understand complex phenomena of human society. She wants to understand both human's collective smarts and their collective stupidity. Her recent applications of interest are urban areas and collective decision-making. Her approach involves two aspects: building mathematical models informed by psychological and social principles of human behavior and using real-world datasets to inform and confront these models. Vicky received a Ph.D. in Applied Mathematics from Northwestern University.


Adding Fast GPU Derived Datatype Handing to Existing MPIs

Carl Pearson, PhD, Sandia National Labs

Wednesday, May 5, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

MPI derived datatypes are an abstraction that simplifies handling of non-contiguous data in MPI applications.These datatypes are recursively constructed at runtime from primitive Named Types defined in the MPI standard.

More recently, the development and deployment of CUDA-aware MPI implementations has encouraged the transition of distributed high-performance MPI codes to use GPUs. Such implementations allow MPI functions to directly operate on GPU buffers, easing integration of GPU compute into MPI codes.

This talk presents a novel datatype handling strategy for nested strided datatypes on GPUs, and its evaluation on a leadership-class supercomputer that does not have built-in support for such datatypes. It focuses on the datatype strategy itself, implementation decisions based off measured system performance, and a technique for experimental modifications to closed software systems.

Bio:

Carl Pearson is a postdoctoral appointee at Sandia National Labs in the Center for Computing Research. There, he works in the Scalable Algorithms group on multi-GPU communication and leveraging specialized GPU computation hardware for generic computing tasks. He received his Ph.D in Electrical and Computer Engineering from the University of Illinois.


Internet of Things-enabled Passive Contact Tracing in Smart Cities

Zeinab Akhavan, UNM CS

Wednesday, April 28, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

Contact tracing has been proven an essential practice during pandemic outbreaks and is a critical non-pharmaceutical intervention to reduce mortality rates. While traditional contact tracing approaches are gradually being replaced by peer-to-peer smartphone-based systems, the new applications tend to ignore the Internet-of-Things (IoT) ecosystem that is steadily growing in smart city environments. This work presents a contact tracing framework that logs smart space users’ co-existence using IoT devices as reference anchors. The design is non-intrusive as it relies on passive wireless interactions between each user’s carried equipment (e.g., smartphone, wearable, proximity card) with an IoT device by utilizing received signal strength indicators (RSSI). The proposed framework can log the identities for the interacting pair, their estimated distance, and the overlapping time duration. Also, we propose a machine learning-based infection risk classification method to characterize each interaction that relies on RSSI-based attributes and contact details. Finally, the proposed contact tracing framework’s performance is evaluated through a real-world case study of actual wireless interactions between users and IoT devices through Bluetooth Low Energy advertising. The results demonstrate the system’s capability to accurately capture contact between mobile users and assess their infection risk provided adequate model training over time.

Bio:

Zeinab is a PhD student in Computer Science at UNM. She has served as UNM Women in Computing Vice President from 2018-2020 and helped CS undergraduate and graduate students build their network within the department. Zeinab’s research interests lie on applying machine learning/deep learning techniques into Internet of Things applications, such as smart buildings/cities.


Upgrading the Product Development Process to Foster ML Fairness and Ethical AI

Donald Martin, Google

Wednesday, April 21, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

Although technology and computer algorithms have significantly advanced health care in recent times, we’ve seen examples where faulty assumptions or biased training data have only served to increase health disparities. Recent research on algorithmic fairness has highlighted that the problem formulation phase of the development of systems that use machine learning can be a key source of bias and have significant downstream impacts on fairness outcomes. However, very little attention has been paid to methods for improving the fairness of this critical phase of machine learning system development. Current practice neither accounts for the dynamic complexity of high-stakes domains nor incorporates the perspectives of vulnerable stakeholders. This talk will explore the application of community-based system dynamics during the product development process in order to foster equitable and inclusive technology based on machine learning and artificial intelligence.

Bio:

Donald Martin, Jr. is currently Sr. Staff Technical Program Manager and Social Impact Technology Strategist at Google. He focuses on driving innovation in the spaces where Google's products and services intersect with society as well as understanding the intersections between Trust and Safety, Machine Learning (ML) Fairness and Ethical Artificial Intelligence (AI). He holds a Bachelor of Science degree in Electrical Engineering from the University of Colorado at Denver and founded its National Society of Black Engineers (NSBE) chapter. Donald has over 30 years of technology leadership experience in the telecommunications and information technology industries. He has held CIO, CTO, COO, VP of IT and product manager positions at global software development companies and telecommunications service providers. Donald holds a US utility patent for "problem modeling in resource optimization." His most recent publication is the Harvard Business Review article "AI Engineers Need to Think Beyond Engineering.”


Free Space Optics based Backhaul/Fronthaul Design for 5G and Beyond

Xiang Sun, PhD, UNM Department of Electrical and Computer Engineering

Wednesday, April 14, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

As 5G technologies are being rolled out, research effort is concentrated on the evolutionary solutions for post-5G and 6G eras. In order to meet the requirements of increased capacity, reduced latency, and on-demand service for mobile networks, various technologies, such as free space optics (FSO) and drone mounted base stations, have been proposed to be the enabling solutions for the next generation mobile networks. In this talk, a tunable FSO system will be introduced to provide low-cost and high-speed backhaul/fronthaul links between geographically distributed base stations and the gateway, where the base stations are communicating with the gateway via the FSO links in the time-division multiplexing manner. In addition, we will illustrate the FSO based drone assisted mobile network framework, where drone mounted base stations are deployed over some places of interest (such as hotspots and disaster struck areas) to assist the base stations to communicate with the mobile users in the places of interest via FSO links. Some challenges and potential solutions to achieve the FSO based drone assisted mobile network will also be presented.

Bio:

Xiang Sun is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of New Mexico. He received the B.E. and M.E. degrees from the Hebei University of Engineering in 2008 and 2011, respectively, and the Ph.D. degree in electrical engineering from the New Jersey Institute of Technology (NJIT) in 2018. He has (co-)authored 44 technical publications, held one U.S. patent, and filed six U.S./PCT non-provisional patent applications. His research interests include mobile edge computing, wireless networks, distributed machine learning, and Internet of Things. He has received several honors and awards, including the 2016 IEEE International Conference on Communications Best Paper Award, the 2017 IEEE Communications Letters Exemplary Reviewers Award, the 2018 NJIT Hashimoto Price, the 2018 Inter Digital Innovation Award on IoT Semantic Mashup, the 2019 NJIT Outstanding Doctoral Dissertation Award, and the 2019 IEICE Communications Society Best Tutorial Paper Award. He currently serves as Associate Editor of the Digital Communications and Networks and the IEEE Open Journal of the Computer Society.


Designing and Optimizing MPI for Next-generation Applications and Systems

Patrick Bridges, PhD, UNM Computer Science

Wednesday, April 7, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

The Message Passing Interface (MPI) has been the standard for building distributed scientific applications for more than three decades, but both computer architectures and scientific applications have changed dramatically in this time, resulting in systems that are significantly more complex and difficult to program and optimize than the systems for which MPI was originally designed. Because of this, MPI has had to and must continue to change as well, taking on more responsibility for managing and optimizing the complex communication required to effectively leverage modern supercomputers. In this talk, I will describe the challenges faced by next-generation parallel communication systems and describe several research directions we are pursuing to address these challenges. This includes research on the communication performance of HPC applications on modern systems, how to effectively model the performance of communication primitives so they can be successfully optimized, and higher-level communication patterns and primitives commonly used by HPC applications and how new MPI abstractions such as neighbor collectives and persistent, partitioned communication can or could be used to improve their performance.

Bio:

Patrick Bridges is a Professor of Computer Science and the Director of the UNM Center for Advanced Research Computing. He received his B.S. in Computer Science from Mississippi State University in 1994 and his Ph.D. in Computer Science from the University of Arizona in 2002 and joined the faculty of UNM immediately thereafter. His research interests cover a wide range of topics related to operating systems and networking for large-scale high-performance computing systems, including virtualization, performance measurement and modeling, fault tolerance, and general system design issues.


Time-series data mining and machine learning techniques for seismic signal detection and classification

Mohammad Ashraf Siddiquee, PhD Candidate, UNM Computer Science

Wednesday, March 24, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

Current seismic data processing pipeline is surprisingly human-dependent. With the rapid increase of seismic-sensor data availability, all manual data processing approaches fail to detect, classify, and analyze seismic activity within a reasonable amount of time. An automated, fast, and reliable seismic data processing pipeline is desired for meaningful analysis of massive seismic datasets. In this talk, we will show how advanced time-series data-mining and machine learning techniques can be leveraged to resolve this issue. We will particularly focus on seismic activity detection, classification, and inspection using our techniques that would help us better understand the surrounding earth structure, earthquake evaluation, and seismic monitoring.

Bio:

Ashraf is a PhD candidate in the UNM Computer Science Department. He works under the supervision of Dr. Abdullah Mueen, and his research revolves around designing and developing novel data mining and machine learning approaches that can be applied to large time-series (i.e. seismic sensor) datasets. For his research work, he collaborates with Los Alamos National Laboratory and Air Force Research Laboratory. He has interned at NEC laboratories America Inc where he implemented a domain agnostic anomaly detection system in univariate time-series. In his free time, Ashraf goes outdoors for biking or fishing.


Financial Science is Computer Science

Donour Sizemore, PhD, Two Sigma Investments

Wednesday, March 10, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

In this talk, the speaker will survey the types of problems that finance companies solve, from the mundane to the novel. With this motivation, we introduce examples of the technical problems that investment managers see on a daily basis. Highlights will include automated decision making (modeling/forecasting), scalable data collection, and technical debt. No finance background is necessary.

Bio:

Donour is Vice President at Two Sigma Investments, where he has worked as a software engineer since 2014. He has worked on data science problems in a variety of industries and is a UNM Computer Science alumnus Previously, he was Trackside Systems Director at Michael Waltrip Racing (2011-2013), Visiting Researcher at Sun (2009), and Researching Computing Director at the Chicago Economic Research Center (2003-2005). Donour holds a Ph.D. in Computer Science from the University of New Mexico (2011) and a BS in Mathematics and Computer Science from the University of Chicago (2003).


From Robot Swarms to COVID-19: Interactions in Space Determine Temporal Dynamics

Melanie Moses, PhD, The University of New Mexico Department of Computer Science

Wednesday, February 24, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

In complex systems high level patterns emerge from small scale interactions. The nature of those interactions depends on where agents are located in physical space. In this talk I highlight the importance of understanding small scale interactions to predict behavior of robot swarms and disease dynamics. The talk reviews our bio-inspired algorithms for robot foraging, the VolCAN swarm of volcano monitoring robots, and how the spatial dynamics of viral infection in the lung determines viral load which ultimately influences epidemic spread of COVID-19.

Bio:

I am a Professor of Computer Science with a secondary appointment in Biology at the University of New Mexico. I'm also an external faculty member of the Santa Fe Institute. I earned my undergraduate degree in Symbolic Systems at Stanford University and my PhD in Biology at UNM. I've recently run the NASA Swarmathon and NM CSforAll educational programs, and I am a Co-PI of the UNM Advance program to support women and underrepresented faculty in STEM. My current research projects include the VolCAN project to develop a swarm of autonomous adaptive robots to monitor volcanoes and predict eruptions, and SIMCov, a spatial model of COVID-19 lung infection and immune response. I am also a co-PI on two AI research institute planning grants, one on the foundations of intelligence at the Santa Fe Institute, and the Proteus Institute at the University of Vermont.


Hand and Machine

Leah Buechley, PhD, UNM CS Department

Wednesday, February 10, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

The aim of my research is to introduce the creative and intellectual potential of computers and electronics to new audiences. I believe that by making technology more accessible and building artifacts that look and feel different from anything that has been built in the past, I can change and broaden the culture of technology. I can get a diverse range of people excited by the ways that computers and electronics can be used to build beautiful, expressive, and useful objects. I can also illuminate the deep relationships between the tools that we use and the communities that we create. To achieve these goals, I integrate computation and electronics with materials from art and design, like paper, textiles, ceramics, and wood. I then use these integrations to develop new tools and approaches that others can employ. I research the adoption of the tools I develop to understand how different people and communities use and learn from different materials. This talk will present an overview of my work, focusing in particular on two recent projects, designing and building Interactive Murals and exploring computational design and ceramics.

Bio:

Leah Buechley is an associate professor in the computer science department at the University of New Mexico, where she directs the Hand and Machine research group. Her work explores integrations of electronics, computing, art, craft, and design. She is a pioneer in paper and fabric-based electronics and her inventions include the LilyPad Arduino, a construction kit for sew-able electronics. Previously, she was a professor at the MIT Media Lab, where she founded and directed the High-Low Tech group. Her work has been featured in publications including The New York Times, Boston Globe, and Wired and exhibited in venues including Ars Electronica, the Exploratorium, and the Victoria and Albert Museum. In 2017, her work was recognized with the Edith Ackerman award for Interaction Design and Children. Leah received a PhD in computer science from the University of Colorado at Boulder and a BA in physics from Skidmore College.


Locality-Aware Data Movement on Modern Supercomputers

Amanda Bienz, PhD, UNM CS Department

Wednesday, February 3, 2021, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

The performance of parallel sparse linear solvers, such as algebraic multigrid, is limited by inter-process communication constraints. The cost of communication is dominated by inter-node messages, while on-node messages are relatively inexpensive. Therefore, communication can be optimized by limiting inter-node communication, exchanging it for additional intra-node messages. During this talk, I will discuss performance expectations of irregular communication as well as locality-aware methods for optimizing the cost. Furthermore, I will present extensions of this work to modern heterogeneous architectures.

Bio:

Amanda Bienz is an assistant professor in the department of computer science at the University of New Mexico. She received her PhD from the University of Illinois at Urbana-Champaign in 2018. Her research is focused on improving the performance and scalability of numerical methods on modern parallel architectures, with a focus on sparse matrix operations, locality-awareness, data aggregation, and heterogeneous architectures.


Characterizing Biomolecular Binding with a Docking Game

Bruna Jacobson, PhD, UNM CS Department

Wednesday, December 2, 2020, 2:00-2:50 PM

Join via Zoom: https://unm.zoom.us/j/98715707842
Meeting ID: 987 1570 7842
Passcode: 9620277

Abstract:

Biomolecular interactions are vital to the large majority of processes that sustain life. When these interactions go astray, they can lead to the onset of diseases and allergies. Research on molecular binding and recognition is central to the development of new drugs and vaccines. We are witnessing this right now, as the research community is trying to identify potential molecular targets for vaccines and treatments against COVID-19. Simulations of biomolecular binding can greatly contribute by filling in knowledge gaps from wet-lab experiments. Computer simulations at the atomic scale can help to elucidate these interactions in detail. However, atomistic simulation is currently limited to short time and length scales and may not capture the phenomena we want to observe. In this colloquium, I will address the design and preliminary results from DockAnywhere, a molecular puzzle game to characterize molecular binding. In this molecular docking game, players can help find bound positions of a small molecule and a protein by manipulating it in the high-dimensional space of molecular interactions. The crowdsourced molecular motion data are used to identify potential binding sites and reconstruct possible pathways of binding with techniques from robotics and biophysics.

Bio:

Dr. Bruna Jacobson is an Assistant Professor at UNM Computer Science. She has a Ph.D. in Physics from the University of Southern California, 2012, and more recently held a Research Assistant Professor position at the Tapia Lab at UNM Computer Science. Her research interests are varied and include Computational Biology, Human-Automation Collaboration, Machine Learning for Biomaterials, Biophysics, Bioinformatics, and Computational Soft Matter Physics. She has often showcased her research in outreach activities around Albuquerque.


Realtime.Earth: Collective Intelligence from Distributed Imagery for Wildland Fire

Kasra “Kaz” Manavi, PhD, Director of Research and Communications, Simtable

Wednesday, November 11, 2020, 2:00-2:50 PM

Abstract:

We are currently fighting “blind” on wildland fire incidents. Fire location and behavior intelligence is crucial during the initial phase of an incident, but reports of wildfire can be delayed for hours. To make matters worse, changes in fuel loads and forest composition along with increasing fire season lengths are resulting in larger and more intense fires. With recent events like the Tubbs, Atlas and Camp Fires, more and more catastrophic wildland fire events are causing significant structure damage and considerable numbers of lives are being lost. Real-time data streams relevant to wildland fire are diversifying e.g. increased activity on social media and publicly accessible imagery. With the increase in these streams, more and more sources of relevant imagery are becoming available during an incident. We suggest the fusion of these data outlets coupled with streaming camera feeds directly from mobile phone browsers can provide real-time situation awareness during the critical first hours of an incident. In this talk we discuss observations obtained using Realtime.Earth, a web-based platform for real-time collective intelligence enabled by imagery capture and collection, data distribution and model visualization, all in the browser. We discuss how imagery captured on mobile devices from citizens, crews and social media can be fused together into live 3D models for real-time fire behavior monitoring.

Bio:

Kasra “Kaz” Manavi is the Director of Research and Communications at Simtable. He received a M.S. in Computer Science from Texas A&M University with an emphasis on robotic motion planning and received a PhD in Computer Science from the University of New Mexico with a focus on computational structural biology. After graduation, he started working at Simtable LLC in Santa Fe, NM, where he has been working on developing a web-based platform to enable real-time collective intelligence by providing users the ability to seamlessly incorporate agent-based modeling, ambient computing, photogrammetry, geospatial information systems and distributed computation into solutions that helps users better understand complex environmental and social phenomena in their community, primarily in the wildland fire space.


Augmenting Human-Infrastructure Interfaces

Fernando Moreu, The University of New Mexico

Wednesday, November 4, 2020, 2:00-2:50 PM

Abstract:

This seminar builds up from existing human-computer and human-machine interfaces, and challenges existing human-infrastructure interfaces with new paradigms and decision-making scenarios. In this seminar, human-infrastructure interfaces are developed within the area of structural health monitoring, exploring Augmented Reality (AR) as the new interface between the deterioration of infrastructure and the humans making real-time decisions. The contents of this talk will emphasize advancing human decisions and cognition of the built environment enabled by a human access to databases, sensors, and assessment tools. To date, new technologies collecting data of the built environment are cheaper, more accurate, diverse, and more accessible than ever before. However, the use and implementation of these new technologies to structural engineers to assess, inspect, or inform actions have been very limited. The examples will present attempts and results of empowering human-machine interfaces in the context of the built environment (human-infrastructure interfaces) and increasing human involvement and participation (human-in-the-loop) through AR. This seminar will present specific practical implementations about how the collection of data, their analysis, and their interpretation can inform human decisions in the area of structural engineering, emergencies and rescue, smart cities and communities, and other overlapping research themes grounded in computer science and engineering.

Bio:

Dr. Fernando Moreu, PE is an assistant professor in structural engineering at the Department of Civil, Construction and Environmental Engineering (CCEE) at the University of New Mexico (UNM) at Albuquerque, NM. He holds courtesy appointments in the Departments of Electrical and Computer Engineering and Mechanical Engineering, both at UNM. He is the founder and director of the Smart Management of Infrastructure Laboratory (SMILab) at UNM (http://smilab.unm.edu/). SMILab is headquartered at the Center for Advanced Research and Computing (CARC) at UNM and aims to develop the use of next-generation smart sensing technologies and strategies towards safer, cost-effective, resilient and sustainable structure . Dr. Moreu’s industry experience includes ESCA Consultants, Inc. for over ten years, with experience in the design, construction and replacement of over thirty bridges in the US. Research interests include structural dynamics and vibrations, structural health monitoring, wireless smart sensor networks, field monitoring of critical infrastructure, augmented reality, unmanned aerial systems, human-machine interfaces, nonlinear dynamics, cyber-physical systems, and aerospace structures design, monitoring, and reusability. Dr. Moreu received his MS and PhD degrees in structural engineering from the University of Illinois at Urbana-Champaign (2005 and 2015, respectively.)


Screening at Scale: Best Practices for Research and Development in Eye Disease Detection

Jeremy Benson, The University of New Mexico

Wednesday, October 14, 2020, 2:00-2:50 PM

Abstract:

In this talk, we will cover some common sight-threatening complications that occur alongside Diabetes. Using tools from image processing and machine learning, we will highlight some successful approaches to making general solutions that work across datasets -- using both low-cost and high-end camera technologies -- as well as mitigating variations that appear throughout different demographics. Finally, we will discuss some tactics to deploy services on cloud infrastructure (AWS) so that they are widely available, cost-effective, and yield quick results.

Bio:

Jeremy Benson is a PhD candidate in Computer Science at the University of New Mexico. He is a member of Dr. Estrada's Data Science Group, where his research focuses on semi-supervised approaches to data discovery and labeling. For the past 5 years, he has been with VisionQuest Biomedical (ABQ, NM), where he works on software solutions for medical diagnostics. Outside of work and school, Benson enjoys playing board games like Chess and Go, or video games like Among Us, which is pretty suspicious.


How to Perform Binary Classification without a Binary Classifier?

Abhinav Aggarwal, PhD, Amazon

Wednesday, September 30, 2020, 2:00-2:50 PM

Abstract:

Binary classification is a fundamental task in machine learning. The success of any such classifier is often measured through real-valued statistical aggregates computed using the classifier's predictions on a set of test data points against their true labels. In this talk, I will present algorithms that use common performance metrics like AUC, Log-Loss, Precision, Recall, or F1-Scores to infer the true labels of arbitrarily many test data points without training any model and requiring only the knowledge of the size of the test dataset. This helps provide insight into the extent of information leakage from exposing these statistical aggregates and how it can be exploited.

Bio:

Abhinav is an Applied Scientist at Amazon Alexa, working on differentially private mechanism design for data de-identification and making machine learning models robust to privacy leakage attacks. Prior to this, he has worked with VISA Research, Google, Microsoft, Cornell University and University of Saskatchewan for internships. He obtained his Ph.D in Computer Science in 2019 from the University of New Mexico (UNM), under the supervision of Prof. Jared Saia. During this time, he worked on fault tolerant distributed computing and its applications to bio-inspired algorithms for swarm foraging, bagging the best paper award at SIROCCO 2020 and a similar nomination at ICDCN 2020.


Beating Sybil with Resource Burning

Diksha Gupta, The University of New Mexico

Wednesday, September 23, 2020, 2:00-2:50 PM

Abstract:

In a permissionless system, due to the absence of a certifying authority, participants can join and depart at will. Taking advantage of this, an attacker can inject a large number of adversarial pseudo participants, thereby launching a Sybil attack on the system. Existing defense techniques use computational puzzles to limit the number of adversarial participants proportional to the fraction of computational power with the attacker. But these impose a computational cost on the system even in the absence of an attack.

In this talk, we will first discuss an algorithm- ESTIMATE, that provides bounds on the number of new honest participants in a permissionless system. To this end, we present an empirical study of the performance of our algorithm on a number of real-world permissionless systems. Then, we will discuss the application of this algorithm to design an efficient Sybil defense algorithm - ERGO. This algorithm gives the following guarantees: 1) always maintains a majority of honest participants in the system, and 2) the cost to the honest participants grows sub-linearly in the cost to the attacker.

Bio:

Diksha is a Ph.D. Candidate in the Department of Computer Science at The University of New Mexico (UNM), under the advisement of Prof. Jared Saia. Her current research is focused on designing provably secure, scalable, and efficient protocols for distributed systems. She obtained an M.S. in Computer Science at UNM. Prior to this, she completed an M.Tech in Computer Science and Engineering from the Indian Institute of Technology (IIT), Roorkee, India. In her free time, she likes solving puzzles, reading sci-fiction, and rock climbing. She has served on the boards of various student and departmental organizations – UNM CS Advisory Board (Graduate Representative), Computer Science Graduate Student Association (CSGSA) (President), and UNM Women in Computing (Vice President). She will be joining as a Research Fellow at the National University of Singapore (NUS) in December 2020.


Imputation and characterization of uncoded self-harm in major mental illness using machine learning

Praveen Kumar, The University of New Mexico

Wednesday, September 16, 2020, 2:00-2:50 PM

Abstract:

Suicide is 1 of the 10 leading causes of death in the United States with self-harming behavior being a major risk factor for suicide. Inadequate coding of suicidality and self-harm in medical records has been consistently reported.  Underreporting of self-harm impedes the ability to estimate event prevalence and reduces the statistical power to perform time-to-event comparative effectiveness pharmacotherapy studies. The objective of this study was to apply machine learning algorithms at the visit level to impute self-harm events that were uncoded in claims data of individuals with major mental illness (MMI) (schizophrenia, schizoaffective disorder, major depressive disorder, and bipolar disorder),  to identify factors associated with coding discrepancies, and to characterize coded vs imputed self-harm incidence in various demographic groups.

Bio:

Praveen is a computer science Ph.D. student at The University of New Mexico, working with Prof. Christophe G. Lambert at the Department of Internal Medicine of UNM. Medical claims data are often not clean; come with noisy labels and missing phenotypes/outcomes. Missing information poses several challenges while applying machine learning algorithms on such data. Working on claims data, Praveen primarily focuses on developing techniques to impute missing phenotypes/outcomes. His work on imputing missing phenotypes for self-harm won the best poster award at the OHDSI (Observational Health Data Sciences and Informatics) Symposium. He contributed to the OHDSI open-source software to convert the CMS (Centers for Medicare & Medicaid Services) Data to OMOP CDM (Observational Medical Outcomes Partnership Common Data Model) compatible files. He has a Computer Engineering Bachelor's degree from the National Institute of Technology, Surat(India), and a Computer Science Master's degree from the University of New Mexico. After finishing his Bachelor's degree, he worked for IT companies - DXC Technology, Fiserv, and Travelport. At these companies, he worked for the development and enhancement of software products related to banking, insurance, and travel domains.


A new interpolation algorithm for the theory of Equality with Uninterpreted Functions

Jose Abel Castellanos Joo, The University of New Mexico

Wednesday, September 9, 2020, 2:00-2:50 PM

Abstract:

An interpolant for a pair (A, B) of inconsistent formulas is a formula C such that: A implies C; B is inconsistent with C; and C  only contains common symbols between A and B. Modern techniques for interpolant generation rely on special deductive calculus and unsatisfiability proofs.  In this talk, we will discuss a new algorithm to compute the interpolation formula for the theory of Equality with Uninterpreted Functions (EUF) that does not require unsatisfiability proofs.  We will discuss an observation made during the implementation of the algorithm, introducing a new Horn-unsatisfiability algorithm that uses a congruence closure with explanations as the mechanism for equality propagation.

Bio:

Jose is a Ph.D. student in the computer science department at the University of New Mexico working with Prof. Deepak Kapur. His research interests span from formal methods to computer algebra. His goal is to combine decision procedures from sums of squares algorithms to provide new answers to verification problems. Jose enjoys reading about self-reference, learning programming languages/tools, and playing the guitar whenever he finds free time.


A Multi-Robot Loss-Tolerant Algorithm for Surveying Volcanic Plumes

John Ericksen, The University of New Mexico

Wednesday, September 2, 2020, 2:00-2:50 PM

Abstract:

Measurement of volcanic CO2 flux by a drone swarm poses special challenges. Drones must be able to follow gas concentration gradients while tolerating frequent drone loss. We present the LoCUS algorithm as a solution to this problem and prove its robustness. LoCUS relies on swarm coordination and self-healing to solve the task. As a point of contrast, we also implement the MoBS} algorithm, derived from previously published work, which allows drones to solve the task independently. We compare the effectiveness of these algorithms using drone simulations, and find that LoCUS provides a reliable and efficient solution to the volcano survey problem. Further, the novel data-structures and algorithms underpinning LoCUS have application in other areas of fault-tolerant algorithm research.

Bio:

John Ericksen is a software developer with Honeywell Federal Manufacturing and Technologies and a computer science Ph.D. student at the University of New Mexico with the Moses Biological Computation Lab. Working with the earth and planetary sciences department, John's research focus is on autonomous airborne robot swarms used to sample volcanic CO2 plumes. The goal of this is to link volcanic CO2 output with volcanic behavior to better understand the precursors to life-threatening eruptions. John has also published on a variety of other research topics including software architecture, evolutionary complex systems, and intelligent swarm robotics. John holds a computer science Bachelor's degree from Western Washington University and a computer science Master's degree from the University of New Mexico. At Honeywell, John works with a team of software developers that develop software solutions in a variety of contexts to the Federal Government. In his free time, John likes to spend his time underwater. He enjoys scuba diving throughout the United States and the Caribbean, works as a scuba diving instructor at a shop in Albuquerque, and loves nature photography, especially the underwater variety.


Advancing Machine Learning and Machine Vision Using Topological Graph-Based Representations, Methods, and Algorithms

Liping Yang, UNM Department of Geography and Environmental Studies

Wednesday, March 4, 2020
Centennial Engineering Center 1041
2:00-3:00 PM

Abstract:

Google can tell what it is for photo search well, but not for drawing search. Drawing-based search remains challenging because drawing images contain much less information compared with natural images; no color and texture, only shape and topology. In this talk, we will present remaining challenges in computer vision, and we will show how we overcome the obstacles by developing new image representation, methods and algorithms based on topological graph and computational geometry. Our image representation, methods and algorithms can make machine learning and machine vision learn and see better. We will show the effectiveness of our topological graph-based image representation and methods using two applications: image classification and image denoising.

Bio:

Dr. Liping Yang is an assistant professor of geographic information science (GIScience) and geospatial artificial intelligence (GeoAI) in the Department of Geography and Environmental Studies at The University of New Mexico (UNM). Dr. Yang received her Ph.D. in Spatial Information Science and Engineering from the University of Maine in 2015; after that she was a Postdoctoral Researcher at Penn State University, where she worked on machine learning and deep learning to analyze big geospatial data, including high resolution aerial images. After Penn State and prior to UNM , Dr. Yang was a postdoctoral research associate in the Information Sciences group of the Computer, Computational, and Statistical Sciences Division at Los Alamos National Laboratory (LANL), focusing on computer vision and machine learning algorithm development for technical diagram image analysis. Dr. Yang has worked many years at the intersection of Computer Science, Mathematics, and GIScience. Her multidisciplinary background on GIScience, graph theory, computational geometry, and machine learning provides her a solid foundation to develop creative and novel solutions to advance computer vision tasks such as image representation, retrieval, and analysis to advance machine vision. Dr. Yang has multiple top-tier journal papers (e.g., IJGIS, Soft Computing) and conference papers (e.g., ACM SIGSPATIAL GIS, CVPR, ICCV, KDD) in GIScience, GeoAI, and computer vision areas.


Compiler Directed Lightweight Resilience Mechanisms for HPC Applications

Chao Chen, CS Assistant Faculty Candidate

Wednesday, February 26, 2020
Centennial Engineering Center 1041
2:00-3:00 PM

Abstract:

Transient faults are becoming a significant concern for emerging extreme-scale high performance computing (HPC) systems. This nascent problem is exacerbated by technology trends toward smaller transistor size, higher circuit density and the use of near-threshold voltage techniques to save power. They could corrupt the execution of long-running scientific applications by leading to either SDCs (incorrect values in outputs) or soft failures (abnormal termination, e.g., process crashes). While SDCs harm the confidence in computations and could lead to inaccurate and untrustworthy scientific insights, soft failures degrade system efficiency and performance since they require the impacted jobs to be restarted from their checkpoints and re-executing the lost computations before continuing the normal operation. As a consequence, transient faults detection as well as recovery must be dealt with in the HPC system design for its usability (trust in the output results) and efficiency (speedup and energy efficiency). In particular, solutions must be designed that have very low regular execution overheads, as well as an ability to detect (and potentially recover from) a large set of faults with negligible downtime.

In this talk, I will present two compiler driven resilience techniques, called LADR and CARE, which are designed respectively for SDC detection and soft failure (SF) recovery. By exploring applications’ knowledge via compiler techniques, they both achieve high fault coverage (~80%), but incur negligible or even zero runtime overheads. I will first describe LADR which detects the SDCs in scientific applications by watching for data anomaly of their state variables (those of scientific interest), and employs compile-time data-flow analysis to minimize the number of monitored variables, thereby reducing runtime and memory overheads. The compiler analysis uses the algebraic properties of the underlying data-flow to select the variables where the fault appears in a magnified manner. The technique is able to maintain a high level of fault coverage with low false positive rates. I will then introduce CARE, a compiler-assisted online recovery technique against soft failures. The advantages of CARE are that it can quickly (with milliseconds) repair the (crashed) process on-the-fly allowing applications to continue their executions instead of being simply terminated and restarted, and incur zero runtime overhead during the normal execution of applications. For recovery, it utilizes the live variables of the program resident in registers and reconstructs the failed computation. Finally, I will conclude my talk by describing future directions towards applying compiler technologies for efficient implementation of the desired system properties.

Bio:

Chao Chen is a Ph.D. candidate in the School of Computer Science at Georgia Tech, advised by Santosh Pande and Greg Eisenhauer. His research interests are broadly in the areas of compilers and systems, with a thesis research on lightweight resilience techniques for HPC applications by exploring applications’ properties. His work appears in top-tier HPC venues, and was nominated for Best Student Paper at SC '19.


High-Performance Graph Analytics with GraphBLAS and LAGraph

Scott Kolodziej, Texas A&M University

Wednesday, February 19, 2020
Centennial Engineering Center 1041
2:00-3:00 PM

Abstract:

Graph-structured data continues to grow in size and complexity, from social networks to graph databases to genome graphs. Even deep neural networks, which have traditionally been treated as dense networks, are now being formulated with sparse connectivity layers to resemble graphs. At the core of these challenges is the field of graph analytics: the application of novel graph algorithms to answer questions about the relationships between data. Efficient solutions to these problems often require a varied approach that utilizes state-of-the-art algorithm design, mathematics, high-performance computing, and software engineering. In this talk, we will step through some recent advances in graph partitioning, as well as how the graph analytics landscape is being transformed by the GraphBLAS standard and associated algorithmic developments. We will also explore a variety of domains and applications where graph analytics is being used and demonstrate how new high-performance graph algorithms and libraries are being built to enable novel research in these areas.

Bio:

Scott Kolodziej is an Assistant Research Scientist at Texas A&M University in the Department of Computer Science & Engineering. He received his Ph.D. in Computer Science from Texas A&M in 2019 for work on hybrid combinatoric and optimization-based graph partitioning methods while working with Dr. Tim Davis. His research interests include high-performance graph analytics, computational optimization, and software engineering of scientific software. In addition to being named an HPEC 2019 Graph Challenge Champion and Affiliate of the Texas A&M Institute of Data Science for his work in graph algorithms and analytics, Scott was also an ACM Student Research Competition Grand Finalist for his work in software engineering and documentation. His background additionally includes degrees in chemical engineering and industrial experience as an Optimization Engineer at Shell.


Graph-Based Exploration of Energy Landscapes in Biomolecular Interactions

Bruna Jacobson, CS Assistant Faculty Candidate

Wednesday, February 12, 2020
Centennial Engineering Center 1041
2:00-3:00 PM

Abstract:

To understand the biological processes that sustain life requires elucidating how biomolecules interact with each other and their environment. However, these interactions are highly complex and involve molecular motion in a high-dimensional space. Biomolecules such as proteins, nucleic acids, and lipids consist of a large number of atoms and hence exhibit many degrees of freedom. Moreover, their behavior is dependent on external factors such as the solvent, temperature, and pH. This complexity explains the immense effort over the last forty years to create computational models that accurately represent molecular interactions. While there has been significant progress in computational methods and the development of high-performance computing systems to run them, enabling atomistic simulations of relatively large systems, there is a significant limit on the ability to simulate processes that occur over timescales longer than about one microsecond. In this talk, I will present how graph-based models of molecular interactions can reduce the dimensionality of motion of a large protein complex by mapping a weighted directed graph onto a rugged and dynamic high-dimensional energy landscape. Protein motion is then derived from pathway determination on the graph. Chemical and conformational changes can be simulated for biomolecular interactions via chemical reaction networks. I will show how transition rate parameters in these networks can be optimized via supervised learning. I will highlight the potential of such networks to be combined with the graph-based model of the energy landscape to simulate biochemical pathway determination. Lastly, I discuss how we are using this modeling approach as a game, DockAnywhere, that will allow users to experience biomolecular interactions on their mobile devices.

Bio:

Dr. Bruna Jacobson is currently a Research Assistant Professor at the Computer Science Department at UNM. She has a Ph.D in Physics from the University of Southern California, 2012, and more recently held a postdoctoral position at the Tapia Lab at UNM Computer Science. Her research interests are varied and include Computational Biology, Human-Automation Collaboration, Machine Learning for Biomaterials, Biophysics, Bioinformatics, and Computational Soft Matter Physics. She is the Principal Investigator of an NSF Core Program Award. She has often showcased her research in outreach activities around Albuquerque. Dr. Jacobson was the recipient of a Chateaubriand Fellowship from the French Embassy in Washington D.C. as a graduate student in 2010.


Programming energy landscapes for absolute molecular positioning & robust molecular computation

Chris Thachuk, Senior Postdoctoral Researcher at Caltech

Wednesday, February 5, 2020
Centennial Engineering Center 1041
2:00-3:00 PM

Abstract:

The promise of molecular programming lies in its ability to (i) self-assemble structures with nanometer precision and (ii) process information autonomously in a biochemical context in order to sense and actuate matter. How do you 'program' an energy landscape so that DNA-based devices follow designed reaction pathways, yet incur significant kinetic and thermodynamic energy penalties for spurious pathways? I'll focus on two different projects butting up against this same theme in different ways. (Part I) A very successful example of self-assembly driven by molecular forces is DNA origami. This process can result in the assembly of ~10^10 copies of a designed 2D or 3D shape, with feature resolution of 6 nanometers. By designing the energy landscape of the interaction between a DNA origami shape and a flat surface, we demonstrate that single molecules can be placed with orientation that is absolute (all degrees of freedom are specified) and arbitrary (every molecule's orientation is independently specified). (Part II) The most sophisticated molecular computing systems have been built upon the DNA strand displacement primitive, where a soup of rationally designed nucleotide sequences interact, react, and recombine over time in order to carry out sophisticated computation. Existing systems are often slow, error-prone, require bespoke design and weeks of labor to realize experimentally. I will detail our efforts to fix these issues by introducing a molecular breadboard, capable of computing billions of functions including all 2^32 Boolean predicates with 5 distinct inputs. Its purpose is to "scale-up" what is possible with this technology and to "scale-out" its adoption to new contexts. In order to facilitate the rapid design of new circuits from a common molecular broth, we have developed a compiler that takes as input a logic description and provides as output the optimized set of breadboard components necessary to activate the desired logic behavior. By mixing these preexisting components as prescribed, it is possible to achieve fast, autonomous and robust molecular circuits, from conception to implementation, within a single afternoon. Due to the large separation of time scales between designed and spurious computation, we expect the breadboard architecture will open new research directions in molecular sensing, actuation and interfacing with self-assembly systems.

Bio:

Chris Thachuk is a Banting Fellow awardee and Senior Postdoctoral Researcher at Caltech, with Erik Winfree. Chris works in the areas of DNA computing and molecular programming - how one might compute or build new structures at the nano-scale with bio-molecules such as DNA. Prior to Caltech, Chris was a postdoc at Oxford Computer Science and a James Martin Fellow at the Institute for the Future of Computing, Oxford. He received his PhD in Computer Science from UBC in 2013, advised by Anne Condon. Much of Chris' computations now happen in a test tube.


Reducing Parallel Communication Costs on Emerging Architectures

Amanda Bienz, Postdoctoral Researcher at University of Illinois at Urbana-Champaign

Wednesday, January 29, 2020
Centennial Engineering Center 1041
2:00-3:00 PM

Abstract:

Advances in parallel architectures yield the potential to solve increasing large and difficult problems efficiently. However, parallel communication demands create performance bottlenecks across applications domains. Communication overhead remains a challenge due to the fact that demands are hard to predict, particularly when moving across difference machines. As a result, it is difficult to design algorithms that are generally efficient across a variety of architectures. In this talk, I will present an analysis of parallel communication, dis-playing a heavy correlation between cost and relative location of the sending and receiving processes, with inter-node communication costing significantly more than intra-node. Furthermore, I will present methods for reducing communication costs and improving the performance of parallel applications on emerging architectures, with a focus on re-routing MPI messages on each node to reduce the amount of costly inter-node communication.

Bio:

Amanda Bienz is a Postdoctoral Researcher at University of Illinois at Urbana-Champaign.


Exploring Supercomputer I/O Systems: Performance Learning, Optimization and Beyond

Bing Xie, HPC System Engineer at Oak Ridge National Laboratory

Monday, January 27, 2020
Farris Engineering Center 3100
2:00-3:00 PM

Abstract:

Supercomputer I/O systems are built around scientific codes. These codes issue periodic write bursts to the file systems for various purposes and with various I/O patterns. From the application’s viewpoint, if its I/O system does not absorb data fast enough, then memory to buffer the output is exhausted, forcing the computation to stall before it can output more data. Output stalls leave precious CPU resources underutilized, extending application runtimes and compromising system throughput. In this talk, I will discuss the study on the write performance of production supercomputers, ranging from quantitative I/O behavior analysis to predictive performance modeling with machine learning techniques. In particular, I will talk about the challenges of benchmarking, profiling and modeling on the write performance of supercomputer I/O systems under production load, and discuss the techniques and methods I proposed to analyze the target systems based on the system design, deployment and configuration. Moreover, I will also show my works on data management among heterogeneous filesystems and resource management for workflows on elastic virtual infrastructure, emphasizing on the challenges, opportunities and my approach.

Bio:

Bing Xie is an HPC System Engineer at Oak Ridge National Laboratory. Bing received her Ph.D. in 2017 from the Computer Science Department at Duke University, where she was advised by Jeff Chase. Her research develops performance analysis and prediction methods for supercomputer I/O systems. More broadly, her research interests span distributed systems, storage systems, high-performance computing, and cloud computing. Her papers are published at HPDC, SC, ACM TOS, etc. Among her works, the petascale filesystem study was nominated for Best Paper and also for Best Student Paper at SC’12.


Accelerating the Analysis of Massive-scale Graph-structured Data

George Slota, Assistant Professor, Computer Science Department, Rensselaer Polytechnic Institute

Wednesday, January 22, 2020
Centennial Engineering Center 1041
2:00-3:00 PM

Abstract:

This talk considers the study of large-scale social, informational, and biological network data, topologically represented as graphs. Such graphs are common, complex, and can be very large, which makes them important to study yet computationally difficult to work with. Developing scalable parallelization methods for graph analytical algorithms is an interesting research area with many significant challenges. I will present results from my ongoing collaborations with Sandia National Labs that involve the development of such techniques for effective multicore, manycore, and distributed parallelization on modern high performance computing architectures. This work includes the tera-scale graph partitioning software PuLP/XtraPuLP, low overhead layout and distribution methods that have enabled the complex study of the largest publicly available web crawl to date via my HPCGraph framework, and A-BTER, a new graph generator designed to facilitate benchmarking studies for the community detection problem at a new massive scale.

Bio:

Dr. George Slota is an Assistant Professor in the Computer Science Department at Rensselaer Polytechnic Institute. He previously worked at Sandia National Labs in the Scalable Algorithms Department from 2013-2016. He graduated with his Ph.D. in Computer Science and Engineering from Penn State in 2016 after working in the Scalable Computing Lab with his advisor, Kamesh Madduri. He was partially supported by a Blue Waters Fellowship during my graduate studies. His research interests are in the areas of graph and network mining, big data analytics, combinatorial algorithms, and their relation to parallel, scientific, and high performance computing.