Skip To Main Content

All graduate seminars are held on Mondays and Wednesdays from 4:10 p.m. to 5:25 p.m. in Peterson 118. Registering students for the seminar: See the requirements to receive credit. Join us for this open seminar! 


Spring 2022

"Houston, We Have a Narrative," a book by Randy Olson

Monday, Jan. 24

Dr. Ricardo Gutierrez-Osuna
Professor, Texas A&M University

Abstract
This talk describes the important role that storytelling plays in science communication. The talk is based on Randy Olson’s book titled “Houston, we have a narrative: why science needs story.” The talk will describe the basic structure of research articles and how that structure relates to narrative, going back to classical Greece and mythology. The talk will also present three simple techniques, the Word, Sentence and Paragraph templates, that can be used to inject a narrative component into science writing. 

Biography
Ricardo Gutierrez-Osuna received a Bachelor of Science degree in electrical engineering from the Polytechnic University of Madrid (Spain) in 1992, and Master of Science and Ph.D. degrees in computer engineering from North Carolina State University in 1995 and 1998, respectively. He is currently a professor in the Department of Computer Science and Engineering at Texas A&M University. He has broad research interests in speech processing, machine learning and models of human perception.

Faculty Contact: Ricardo Gutierrez-Osuna


SuiteSparse: GraphBLAS: Graph Algorithms in the Language of Sparse Linear Algebra

Monday, Jan. 31

Dr. Tim Davis
Professor, Texas A&M University

Abstract
SuiteSparse: GraphBLAS is a full implementation of the GraphBLAS standard, which defines a set of sparse matrix operations on an extended algebra of semirings using an almost unlimited variety of operators and types. When applied to sparse adjacency matrices, these algebraic operations are equivalent to computations on graphs. GraphBLAS provides a powerful and expressive framework for creating graph algorithms based on the elegant mathematics of sparse matrix operations on a semiring. Key features and performance of the SuiteSparse implementation of GraphBLAS package are described.  The implementation appears in Linux distros, forms the basis of the RedisGraph module of Redis (a commercial graph database system) and appears as C=A*B in MATLAB. Graph algorithms written in GraphBLAS can rival the performance of highly tuned specialized kernels, while being far simpler for the end user to write.

Biography
Tim Davis is a professor in the computer science and engineering department at Texas A&M University. His primary scholarly contribution is the creation of widely used sparse matrix algorithms and software (including x=A\b in MATLAB). Davis is a fellow of the Society for Industrial and Applied Mathematics, Association for Computing Machinery and Institute of Electrical and Electronics Engineers.

Faculty Contact: Ricardo Gutierrez-Osuna


The Early History of Computing

Wednesday, Feb. 2

Dr. John Keyser
Professor, Texas A&M University

Abstract
This talk is a walk through the early history of computing, from the earliest days of mechanical computation devices through the early forms of computers and software that we use today.  We’ll talk about people who were key parts of many of these developments, and we’ll discuss the factors that drove them into these developments.

Biography
John Keyser is a professor in the Department of Computer Science and Engineering at Texas A&M University. He joined Texas A&M in 2000 after receiving his Ph.D. in computer science from the University of North Carolina and earlier earned bachelor’s degrees in applied math, engineering physics and computer science from Abilene Christian University. Keyser’s research has spanned a wide range of graphics, with the majority of his work in the areas of geometric modeling, especially in robust solid modeling applications, and physically based simulation. He has also worked on topics in rendering, data visualization, and a large interdisciplinary project on scanning and reconstructing small animal brains at sub-micrometer resolution.

Faculty Contact: Ricardo Gutierrez-Osuna


"Writing Science," a book by Joshua Schimel

Monday, Feb. 7

Dr. Ricardo Gutierrez-Osuna
Professor, Texas A&M University

Abstract
This talk describes the basic structure of a scientific paper. The talk is based on Professor Joshua Schimel’s book titled “Writing science, how to write papers that get cited and proposals that get funded.” We will describe four core story structures (OCAR, ABDCE, LDR, and LR) and when to use them in storytelling. We will then focus on how to map the OCAR (Opening, Challenge, Action, and Resolution) structure, with its hourglass shape, into the traditional sections of a research article: Introduction, Methods, Results and Discussion. We will also discuss the basic structure of paragraphs and sentences, and how to use them effectively to improve flow.

Biography
Ricardo Gutierrez-Osuna received a Bachelor of Science degree in electrical engineering from the Polytechnic University of Madrid (Spain) in 1992, and Master of Science and Ph.D. degrees in computer engineering from North Carolina State University in 1995 and 1998, respectively. He is currently a professor in the Department of Computer Science and Engineering at Texas A&M University. He has broad research interests in speech processing, machine learning and models of human perception.

Faculty Contact: Ricardo Gutierrez-Osuna


The Sense of Structure

Monday, Feb. 14

Dr. Ricardo Gutierrez-Osuna
Professor, Texas A&M University

Abstract
This talk will discuss how readers of English tend to make decisions concerning what a given document means. It is based on the existence of recognizable patterns in the interpretative process of most readers. Readers take the greatest percentage of their clues not from word choice but rather from the location of words within the structure of a sentence or a paragraph. The talk is based on the book “The sense of structure: writing from the reader’s perspective” by George D. Gopen. It will describe and provide examples on reader’s expectations at the sentence and paragraph levels. It will also challenge pieces of advice that writers often receive on how to improve their writing.

Biography
Ricardo Gutierrez-Osuna received a Bachelor of Science degree in electrical engineering from the Polytechnic University of Madrid (Spain) in 1992, and Master of Science and Ph.D. degrees in computer engineering from North Carolina State University in 1995 and 1998, respectively. He is currently a professor in the Department of Computer Science and Engineering at Texas A&M University. He has broad research interests in speech processing, machine learning and models of human perception.

Faculty Contact: Ricardo Gutierrez-Osuna


Computing Machinery and Intelligence, a paper by Alan Turing

Monday, Feb. 21

Dr. Dylan Shell
Associate Professor, Texas A&M University

Abstract
Turing's 1950 paper shows that any attempt to create minds, which one might reasonably take as a core goal of artificial intelligence, is fundamentally of philosophical interest. Not only did this contribution to the literature plant a seed from which much followed, including ways of sharpening our thinking about the nature of intelligence, but it actively anticipates questions and challenges faced as part of the quest for artificial intelligence more generally. The paper is highly readable and as relevant today as ever. I don't know many other papers that fit that bill; it is one of my all-time favorites.

Biography
Dylan Shell received his Bachelor of Science degree in computational and applied mathematics and computer science from the University of the Witwatersrand, South Africa, and his Master of Science degree and Ph.D. in computer science from the University of Southern California. He is an associate professor of computer science and engineering at Texas A&M University, where he teaches and conducts research. His research aims to synthesize and analyze complex, intelligent behavior in distributed systems that interact with the physical world.

Faculty Contact: Ricardo Gutierrez-Osuna


Increasing Social Presence with Interactive Virtual Agents in eXtended Reality

Wednesday, Feb. 23

Dr. Kangsoo Kim
Assistant Professor, Department of Electrical and Software Engineering, University of Calgary

Abstract
Our society has been experiencing a convergence of technologies that merges and tightly integrates the physical, digital, and biological spheres. All the physical and digital things are becoming more and more intelligent and connected to each other, and the boundary between them becomes blurry and seamless. In particular, eXtended Reality (XR), ranging from virtual to augmented and mixed reality (VR/AR/MR), combines virtual contents with the real environment and is experiencing an unprecedented golden era along with dramatic technological achievements and increasing public interest. 

Dr. Kangsoo Kim’s research focuses on improving user experience and enhancing human abilities in XR through socially engaging and intelligent virtual agent interactions from the perspective of interdisciplinary human-centered computing research. In this talk, he will present his efforts and achievements on increasing the sense of social presence with of a virtual agent. His method particularly leveraged the agent’s visual embodiment and the physically coherent and plausible behaviors to achieve the effects. Multiple user studies conducted for the past years support the positive influence of such an interactive embodied virtual agent on user experience and perception as an effective user interface. He will also discuss the implications of the research while illustrating different application scenarios.

Biography
Kim is an assistant professor in the Department of Electrical and Software Engineering at the University of Calgary. After completing his Ph.D. in computer science at the University of Central Florida (2018), he was a postdoctoral researcher at the same institute with an appointment in the Institute for Simulation and Training and the College of Nursing (2019 –2020), and at the University of Delaware (2021). His research interests broadly cover pervasive context-aware XR systems and intelligent social interactions in XR, and he has co-led multiple research projects sponsored by the National Science Foundation and Office of Naval Research. His research has been published in top-tier conferences and journals, including Institute of Electrical and Electronics Engineers (IEEE) Transactions on Visualization and Computer Graphics, IEEE International Symposium on Mixed and Augmented Reality, IEEE Conference on Virtual Reality and 3D User Interfaces, and Association for Computing Machinery (ACM) International Conference on Intelligent Virtual Agents, achieving multiple Best Paper Awards at ACM Symposium on Virtual Reality Software and Technology and ACM Spatial User Interaction. 

Faculty Contact: Jeeeun Kim


Studies on 3D Reconstruction

Wednesday, March 2

Dr. Wenping Wang 
Professor, Head of the Department of Visualization, Texas A&M University

Abstract
3D reconstruction is about producing faithful digital models of real-world physical objects or scenes from 2D image input. The task of 3D reconstruction is encountered in many applications, such as VR content creation, urban visualization, CAD/CAM, robotics and autonomous driving. In the first part of this seminar, I will present an efficient pipeline for scanning and reconstructing sherds (i.e., excavated ceramic fragments at archaeological sites), which is a long-standing problem in digitization for archaeology. Existing image acquisition systems typically take several minutes to scan a single sherd, so they are impractical for field studies which usually produce hundreds of sherds every day. Our system is capable of scanning and reconstruct over a thousand pieces per day. It is not only efficient but also portal, affordable and accurate. The system has been tested in archaeological fields and demonstrated expected efficacy and robustness.

In the second part I will present a series of works on 3D reconstruction of thin structures and objects with weak textures, which are among the most challenging types of objects to reconstruct. In particular, I will introduce a neural rendering method, dubbed NeuS, for 3D reconstruction. NeuS integrates the signed distance function (SDF) representation with NeRF [Mildenhall et al., 2020], a recent neural rendering method for novel view synthesis. By encoding a 3D surface as the zero-level set of a network-encoded SDF, NeuS achieves robust and high-quality reconstruction of complex objects, especially objects with complex structures, self-occlusion and weak texture.

Biography
Dr. Wenping Wang conducts research in computer graphics, computer vision, visualization, robotics, medical image processing and geometric modeling. He has published over 190 journal papers in these fields. He is the associate editor of several journals, including Computer Aided Geometric Design, Computers & Graphics, and the Institute of Electrical and Electronics Engineers’ (IEEE) Transactions on Visualization and Computer Graphics, and has chaired over 20 international conferences, including Pacific Graphics (2012), Association for Computing Machinery (ACM) Symposium on Physical and Solid Modelling (2013), SIGGRAPH Asia (2013), and Geometry Summit (2019). He received the John Gregory Memorial Award for contributions in geometric modeling. He is an IEEE Fellow and ACM Fellow.

Faculty Contact: John Keyser


Modeling and Analysis of Plant Roots from 3D Imaging

Monday, March 21

Dr. Tao Ju
Professor, Department of Computer Science and Engineering, Washington University in St. Louis 

Abstract
Roots, the "hidden" half of plants, play a vital role in the development and function of plants. Recent advances in imaging (e.g., CT and MRI) have allowed biologists to "see" the structure and growth of roots in 3D. Due to the complexity of roots, computational methods that extract useful biological information (e.g., shape and branching hierarchy) from 3D imaging are still under-developed. In this talk, I present our recent work on algorithms and software tools for 3D root modeling and analysis. Our work presents novel contributions to both the fields of plant phenotyping and computer graphics, particularly in shape and topology analysis.

Biography
Tao Ju is a professor in the Department of Computer Science and Engineering at Washington University in St. Louis. He obtained his Bachelor of Science and Bachelor of Arts degrees from Tsinghua University in 2000, and his Master of Science and Ph.D. degrees in computer science from Rice University in 2005. He conducts research in computer graphics and bio-medical applications, and is particularly interested in geometric modeling and shape analysis. He is currently the associated editor-in-chief for IEEE Transactions on Visualization and Computer Graphics, and he has served as associate editors for Computer Graphics Forum, Computer-Aided Design, Graphical Models and Computational Visual Media. He has served on the program committees of key conferences in computer graphics and has chaired the program committees of Pacific Graphics (2007), SGP (2018) and GMP (2019). His research is funded by the National Science Foundation (NSF) and National Institutes of Health, including an NSF CAREER award in 2009.

Faculty Contact: Scott Schaefer


Self-Learning Governance of Multi-Agent Systems

Wednesday, March 23

Michael Oesterle 
Doctoral Researcher, University of Mannheim, Germany

Abstract
Multi-agent systems with full agent autonomy often exhibit undesirable behavior with respect to overall system performance or social welfare. To solve this challenge while preserving agent freedom, we propose a restriction-based governance which learns optimal restrictions as a reinforcement learning policy.

Biography
Michael Oesterle is a Ph.D. researcher at the University of Mannheim, Germany, and is currently spending four months at Texas A&M University as a visiting researcher. He holds a bachelor’s and a master’s degree in mathematics from the Technical University of Munich, Germany, and has several years of industry experience in strategy consulting and applied game theory.

Faculty Contact: Guni Sharon


Vision and Language: Compositionality and Fairness

Monday, March 28

Dr. Vicente Ordóñez-Román
Associate Professor, Department of Computer Science, Rice University

Abstract
Compositionality is the ability for an intelligent system to recognize a concept based on its parts or constituents. This ability is essential for humans to use language effectively as there exists a very large combination of plausible objects, attributes and actions in the world. Computational models for visual recognition should ideally be able to represent these parts as well as their combinations in a way that mirrors the compositional structures in language. We argue that compositionality should enable models that are more robust and can be trained with fewer samples. Compositionality could also help mitigate the impact of spurious correlations that amplify societal biases. In this talk, I will describe how current methods fall short in terms of compositionality and outline possible directions going forward. I will also present some of the work in our group in this direction and in trying to leverage language either as a supervisory signal or as a prompt to enable better models that can represent complex visual scenes for synthesis, retrieval and visual recognition.

Biography
Vicente Ordóñez-Román is an associate professor in the Department of Computer Science at Rice University and an Amazon Visiting Academic at Amazon Alexa AI. His research focus is on building visual recognition models that can perform tasks that leverage both images and text. Ordóñez-Román is a recipient of a best paper award at the conference on Empirical Methods in Natural Language Processing (2017) and the Best Paper Award — Marr Prize at the International Conference on Computer Vision (2013). He has also been the recipient of a National Science Foundation Early Career Development Award, an IBM Faculty Award, a Google Faculty Research Award and a Facebook Research Award. From 2016-2021, Ordóñez-Román was an assistant professor in the Department of Computer Science at the University of Virginia. He obtained a Ph.D. in computer science at the University of North Carolina at Chapel Hill and has also been a visiting researcher at the Allen Institute for Artificial Intelligence and a visiting professor at Adobe Research.

Faculty Contact: Ruihong Huang


Combining Networks and Learning to Improve Social and Infrastructure Systems

Wednesday, March 30

Dr. Arlei Silva
Assisstant Professor, Department of Computer Science, Rice University

Abstract
Graphs are a powerful framework for modeling complex systems, such as those arising from society and infrastructure. That is because they can capture how local interactions determine the global properties of the system. Due to their highly combinatorial structure, the study of real graph properties, which is known as network science, often relies on decades of research on efficient algorithms and their theoretical guarantees. However, recently, learning-based approaches have achieved promising results in several tasks, especially in the supervised setting (e.g., node classification, link prediction). One of the key advantages of these recent approaches is their ability to learn representations for heterogeneous data (e.g., structure, attributes). On the other hand, they still lack the same theoretical guarantees and the interpretability of classical approaches. In this talk, we will show how to combine classical and learning-based algorithms for the solution of problems motivated by social and infrastructure systems. We will start by describing how representation learning on graphs can capture node degrees, achieving state-of-the-art results on two link prediction tasks. Next, we will show how deep learning and graph optimization algorithms can be combined for better flow prediction in infrastructure networks. We will also present recent results on graph classification based on the dynamics and sparsification. Finally, we will discuss how representation learning can support novel algorithms for the challenging problem of optimizing networked systems.

Biography
Arlei Silva is an assistant professor of computer science at Rice University. His research focuses on developing algorithms and models for mining and learning from complex datasets, broadly defined as data science, especially for data represented as graphs/networks. Silva is particularly interested in problems motivated by computational social science, infrastructure and healthcare. To address these problems, he applies tools from machine learning, network science, graph theory, linear algebra, optimization and statistics.

Silva received his Ph.D. in computer science from the University of California, Santa Barbara, advised by Ambuj Singh, where he was also a postdoctoral scholar. Before that, he received a Bachelor of Science and Master of Science degrees in computer science from Universidade Federal de Minas Gerais, in Brazil, advised by Wagner Meira Jr. Silva has also been a visiting scholar at the Rensselaer Polytechnic Institute, hosted by Mohammed J. Zaki.

Faculty Contact: Dilma Da Silva


Using Linearizable Objects in Randomized Concurrent Programs

Wednesday, April 6

Dr. Jennifer Welch
Chevron Profesor II, Regents Professor, Texas A&M University

Abstract
Atomic shared objects, whose operations take place instantaneously, are a powerful technique for designing complex concurrent programs. Since they are not always available, they are typically substituted with software implementations. A prominent condition relating this implementations to their atomic specifications is linearizability, which preserves safety properties of programs using them. However, linearizability does not preserve hyper-properties, which include probabilistic guarantees of randomized programs. A more restrictive property, strong linearizability, does preserve hyper-properties, but it is impossible to achieve in many situations. In particular, we show that there are no strongly linearizable implementations of multi-writer registers or snapshot objects in message-passing systems. On the other hand, we show that a wide class of linearizable implementations, including well-known ones for registers and snapshots, can be modified to approximate the probabilistic guarantees of randomized programs when using atomic objects.

Biography
Jennifer L. Welch received her S.M. and Ph.D. from the Massachusetts Institute of Technology and her B.A. from The University of Texas at Austin. She is currently the holder of the Chevron II Professorship and Regents Professorship in the Department of Computer Science and Engineering at Texas A&M University and is an Association for Computing Machinery Distinguished Member. Her research interests are in the theory of distributed computing, especially dynamic networks and distributed data structures.

Faculty Contact: Ricardo Gutierrez-Osuna


Golden Speakers: Transforming Non-Native Speech via Machine Learning

Monday, April 11

Dr. Ricardo Gutierrez-Osuna
Professor, Texas A&M University

Abstract
Despite years or decades of immersion in a new culture, adult second-language (L2) learners typically speak with a so-called ‘‘foreign accent.” A few studies have suggested that it would be beneficial for such learners to be able to listen to their own voices producing native-accented speech, a so-called golden speaker. As a step towards this goal, my group develops speech processing methods that can modify the perceived accent of utterances produced by L2 speakers of English. At a conceptual level, this requires decomposing the speech signal of an L2 learner and a native teacher into two components: one that carries the speakers’ voice quality and a second one that contains their linguistic gestures. By combining the voice quality of the L2 learner with the linguistic gestures of the teacher, we can then generate a “morphed” voice that is perceived to be like that of the L2 speaker but has the pronunciation patterns (or accent) of the native speaker. 

In this talk, I will introduce the general problem of accent conversion, its connection to voice conversion and its role in pronunciation training. I will then review two general approaches to accent conversion that operate in the articulatory and acoustic domains, respectively. I will also describe several accent conversion systems that we have developed, including techniques based on statistical machine learning, sparse representation and deep-learning models. I will conclude the talk with an overview of ongoing work on developing implicit and explicit feedback mechanisms for computer-assisted pronunciation training via accent conversion and mispronunciation detection, respectively. 

Biography
Ricardo Gutierrez-Osuna received a Bachelor of Science degree in electrical engineering from the Polytechnic University of Madrid (Spain) in 1992, and Master of Science and Ph.D. degrees in computer engineering from North Carolina State University in 1995 and 1998, respectively. He is currently a professor in the Department of Computer Science and Engineering at Texas A&M University. He has broad research interests in speech processing, machine learning and models of human perception.

Faculty Contact: Ricardo Gutierrez-Osuna


Communication Algorithms via Deep Learning

Monday, April 18

Dr. Hyeji Kim 
Assistant Professor, Department of Electrical and Computer Engineering, The University of Texas at Austin

Abstract
The design of codes for communicating reliably over a statistically well-defined channel is an important endeavor involving deep mathematical research and wide-ranging practical applications. In this talk, we demonstrate that the discovery of codes can be automated via deep learning. We first show that creatively designed and trained neural networks can decode well-known codes with close-to-optimal performance on the additive white Gaussian noise (AWGN) channels and are much more robust and adaptive to deviations from the AWGN setting compared to existing decoders. Next, we present a family of codes obtained via deep learning, which notably outperforms state-of-the-art codes for challenging communication scenarios, such as multi-terminal channels and feedback channels. Finally, we demonstrate the use of channel coding problems as a benchmark for deep learning (e.g., meta-learning).

This talk is intended for a general audience; no prior knowledge of coding theory or communications will be assumed in the talk.

Biography
Hyeji Kim is an assistant professor in the Department of Electrical and Computer Engineering and a fellow of the Advanced Micro Devices Chair in Computer Engineering at The University of Texas at Austin. She received her Ph.D. in electrical engineering from Stanford University in 2016. Kim was a postdoctoral researcher at the University of Illinois at Urbana-Champaign from 2016 to 2018, and she worked as a researcher at the Samsung AI Research Cambridge from 2018 to 2020.

Faculty Contact: Yoonsuck Choe


Dynamic Systems Modeling of Humans to Improve Proactive Healthcare

Monday, April 25

Dr. Misha Pavel
Professor of the Practice, Khoury College of Computer Sciences and Bouvé College of Health Sciences, Northeastern University

Abstract
The vision of transforming healthcare from reactive sick care to proactive healthcare requires new approaches to assessing individuals’ physical, physiological and mental states and their dynamics. Emerging advances in sensing, computation and communication technology have the potential to enable intensive longitudinal monitoring, assessment and prediction to close the loop by optimizing early detection and tailored intervention. To reach our goal, we use computational modeling, prediction and optimization applicable to individuals in specific contexts and scenarios. This presentation will discuss examples of robust computational modeling and predicting individuals’ behaviors by combining machine learning, hybrid dynamic systems and statistical signal processing with psychological knowledge. First, we describe inferences of cognitive functionality from computer interactions and games that can be used for the early detection of changes in cognitive function. Second, we describe approaches to the inference of stress levels from physiological measurements. Finally, I will describe a principled approach using intensive longitudinal health behavior monitoring to help individuals to increase their physical activity and engagement. We note that an additional benefit of this approach is that it provides transparent explanations of the inferences and recommendations.

Biography
Misha Pavel, Ph.D., holds a joint faculty appointment in Northeastern University’s Khoury College of Computer Sciences and Bouvé College of Health Sciences, and he is a visiting faculty at the University of California, Davis. His background comprises of electrical engineering, computer science and experimental psychology. His research includes multi-scale dynamic computational modeling of behaviors and psychological states, with applications ranging from elder care to augmentation of human performance. Pavel uses these model-based approaches to develop algorithms transforming unobtrusive monitoring from smart homes and mobile devices to practical and actionable knowledge for diagnosis and intervention in care for older adults. Under the auspices of the Northeastern-based Consortium on Technology for Proactive Care, Pavel and his colleagues target technological innovations to support the development of economically feasible, proactive, distributed and individual-centered healthcare. In addition, Pavel is investigating approaches to inferring and augmenting human cognition using computer games, EEG, gait characteristics and transcranial electrical stimulation. Prior to his current positions, he was a program director at the National Science Foundation, faculty at New York University, Oregon Health and Science University, and Stanford University. Before his academic career, he was a Member of Technical Staff at Bell Laboratories.

Faculty Contact: Bobak Mortazavi