Computer Science Seminars

Solving Global Grand Challenges with High Performance Data Analytics

Emerging real-world graph problems include: detecting and preventing disease in human populations; revealing community structure in large social networks; protecting our elections from cyber-threats; and improving the resilience of the electric power grid. Unlike traditional applications in computational science and engineering, solving these problems at scale often raises new challenges because of the sparsity and lack of locality in the data, the need for additional research on scalable algorithms and development of frameworks for solving these problems on high performance computers, and the need for improved models that also capture the noise and bias inherent in the torrential data streams. In this talk, Bader will discuss the opportunities and challenges in massive data-intensive computing for applications in computational science and engineering.

David A. Bader is a Distinguished Professor and founder of the Department of Data Science and inaugural Director of the Institute for Data Science at New Jersey Institute of Technology. Prior to this, he served as founding Professor and Chair of the School of Computational Science and Engineering, College of Computing, at Georgia Institute of Technology. Dr. Bader is a Fellow of the IEEE, ACM, AAAS, and SIAM; a recipient of the IEEE Sidney Fernbach Award; and the 2022 Innovation Hall of Fame inductee of the University of Maryland’s A. James Clark School of Engineering. He advises the White House, most recently on the National Strategic Computing Initiative (NSCI) and Future Advanced Computing Ecosystem (FACE). Bader is a leading expert in solving global grand challenges in science, engineering, computing, and data science. His interests are at the intersection of high-performance computing and real-world applications, including cybersecurity, massive-scale analytics, and computational genomics, and he has co-authored over 300 scholarly papers and has best paper awards from ISC, IEEE HPEC, and IEEE/ACM SC. Dr. Bader is Editor-in-Chief of the ACM Transactions on Parallel Computing and previously served as Editor-in-Chief of the IEEE Transactions on Parallel and Distributed Systems. He serves on the leadership team of Northeast Big Data Innovation Hub as the inaugural chair of the Seed Fund Steering Committee.  In 2012, Bader was the inaugural recipient of University of Maryland’s Electrical and Computer Engineering Distinguished Alumni Award. In 2014, Bader received the Outstanding Senior Faculty Research Award from Georgia Tech.  Bader has also served as Director of the Sony-Toshiba-IBM Center of Competence for the Cell Broadband Engine Processor and Director of an NVIDIA GPU Center of Excellence. In 1998, Bader built the first Linux supercomputer that led to a high-performance computing (HPC) revolution, and Hyperion Research estimates that the total economic value of Linux supercomputing pioneered by Bader has been over $100 trillion over the past 25 years. Bader is a cofounder of the Graph500 List for benchmarking “Big Data” computing platforms. He is recognized as a “RockStar” of High Performance Computing by InsideHPC and as HPCwire’s People to Watch in 2012 and 2014.

Forgotten Promise, Current Peril, & Future Potential of the Internet Trust Architecture

The Public Key Infrastructure (PKI) determines the code our computers install, the web sites we recognize as trustworthy, and what apps our phones will accept.   The reliability of the PKI ecosystem depends on the trustworthiness of the Certificate Authorities (CAs), the code, the cryptography, and the selection of keys.  It also depends on the governance structure and human factors. Who decides what roots of trust are shipped as part of browsers and phones, and in the future automobiles, toys, appliances, and airplane components?  How do certificates fail?  Beginning with a machine learning approach to identify failures, then moving to qualitative analyses. I argue for a more nuanced understanding of trust in the Internet ecosystem. The talk includes an overview of emerging standards, current state, and past practice in PKI.

Jean Camp is a Professor of Informatics and Computer Science. Her research focuses on the intersection of human and technical  trust, with the goal of building for end to end empowerment. She was a member of the 2022 class of Fellows of the ACM. She was selected as a Fellow of the Institute of Electronic and Electrical Engineers in 2018. She was elected a Fellow of the American Association for the  Advancement of Science in 2017.  She was inducted into Sigma Xi – the national research honor society - in 2017. She is currently employed as a Professor at the Luddy School with appointments in Informatics and Computing Science at Indiana University. She joined Indiana after eight years at Harvard’s Kennedy School where her courses were  also listed in Harvard Law, Harvard Business, and the Engineering Systems Division of MIT. She spent the year after earning my doctorate from Carnegie Mellon as a Senior Member of the Technical Staff at Sandia National Laboratories. She began her career as an engineer at Catawba Nuclear Station after a double major in electrical engineering and mathematics, followed by a MSEE in optoelectronics at University of North Carolina at Charlotte.

Threshold Cryptography: From Private Federated Learning to Protecting Your Cryptocurrency

We will present the notion of Threshold Cryptography which aims to secure cryptographic keying materials.  The keys in a cryptographic system are the most critical part, and losing them can cause considerable damage. We will explain how to enhance the means for storing them and how to apply these techniques to specific applications. 

The talk is for a general audience and will be self contained.

Tal Rabin is a Rachleff Family Professor of Computer Science at the University of Pennsylvania. Prior to joining UPenn she has been the head of research and Algorand Foundation and prior to that she had been at IBM Research for 23 years as a Distinguished Research Staff Member and the manager of the Cryptographic Research Group.  She received her PhD from the Hebrew University in 1995. 

Tal’s research focuses on cryptography and, more specifically, on secure multiparty computation, threshold cryptography, and proactive security and recently adapting these technologies to the blockchain environment. Her works have been instrumental in forming these areas.  She has served as the Program and General Chair of the leading cryptography conferences and as an editor of the Journal of Cryptology.   She has initiated and organizes the Women in Theory Workshop, a biennial event for graduate students in Theory of Computer Science. Tal currently serves as the chair of the ACM SIGACT Executive Board.

Tal is an ACM Fellow, an IACR (International Association of Cryptologic Research) Fellow and member of the American Academy of Arts and Sciences. Tal’s work won the 30-year test of time award at STOC. She is the 2019 recipient of the RSA Award for Excellence in the Field of Mathematics. She was named by Forbes in 2018 as one of the Top 50 Women in Tech in the world. In 2014 Tal won the Anita Borg Women of Vision Award winner for Innovation and was ranked by Business Insider as the #4 on the 22 Most Powerful Women Engineers.  

Pursuing Transparency and Accountability in Data and Decision Processes

Algorithmic systems and data processes are being deployed to aid a wide range of high-impact decisions: from school applications or job interviews to gathering, storing, and analyzing personal data, or even performing critical tasks in the electoral process. These systems can have momentous consequences for the people they affect but their internal behaviors are often inadequately communicated to stakeholders, leaving them frustrated and distrusting of the outcomes of the decisions. Transparency and accountability are critical prerequisites for building trust in the results of decisions and guaranteeing fair and equitable outcomes.

In this talk, I will present my work on making these opaque processes more transparent and accountable to the public in several real-world applications.  In particular, I will discuss how ranking aggregation functions traditionally used in decision systems inadequately reflect the intention of the decision-makers. Providing transparent metrics to clarify the ranking process, by assessing the contribution of each parameter used in the ranking function in the creation of the final ranked outcome, more accurately captures the true impact of each parameter in the ranking decision. Furthermore,  ranking functions that are used in resource allocation systems often produce disparate results because of bias in the underlying data.  I will show how a sample-based mechanism based on the use of compensatory bonus points can transparently address disparity in ranking applications.

Organizations and agencies do not have strong incentives to explain and clarify their decision processes; however, stakeholders are not powerless and can strategically combine their efforts to push for more transparency. I will discuss the results and lessons learned from such an effort: a parent-led crowdsourcing campaign to increase transparency in the New York City school admission process. This work highlights the need for oversight and AI governance to improve the trust of stakeholders who have no choice but to interact with automated decision systems.

Amélie Marian is an Associate Professor in the Computer Science Department at Rutgers University, where she was Director of the Undergraduate program from 2014 to 2020. Her research interests are in Explainable Rankings, Accountability of Decision-making Systems, Personal Digital Traces, and Data Integration. Her recent public scholarship work on explaining the NYC School Admission lottery process to families, in collaboration with elected parent representatives, was instrumental in increasing transparency and accountability in the NYC high school application system. Amélie received her Ph.D. in Computer Science from Columbia University in 2005. She is the recipient of a Microsoft Live Labs Award, three Google Research Awards, and an NSF CAREER award.

Bringing Time and Social Space into Natural Language Processing

Human languages have extremely large vocabularies, and by assembling words into sequences, humans can express complex and novel ideas to each other. The likelihood of selecting any given word at any point in time varies greatly as a function of the context. Understanding and formalizing these contextual influences is essential for building robust and adaptable NLP systems.  This talk will focus on two sources of variability: variation over time, and variation across speakers. I will first consider the how individual words behave as a function of who is speaking and what the topic of discussion is.  I will explain how vector space representations of words (also known as word embeddings) make it possible to investigate the abstract concepts that underpin the use of groups of semantically related words. Finally, using social networks defined from social media posts, I will illustrate how graph neural networks can be used to investigate the structure and dynamics of opinions in the social space.

Professor Janet Pierrehumbert has an interdisciplinary background from Harvard and MIT in linguistics, mathematics, and electrical engineering and computer science. Her PhD dissertation developed a model of English intonation that was applied to generate pitch contours in synthetic speech. She began her career as a Member of Technical Staff at AT&T Bell Laboratories in Linguistics and Artificial Intelligence Research. From there, Pierrehumbert moved to Northwestern University, where she headed a research group that used experimental and computational methods to understand lexical systems in English and many other languages.  Pierrehumbert joined the University of Oxford faculty in 2015 as Professor of Language Modelling in the Oxford e-Research Centre,  Her current focusses on robust and interpretable natural language processing methods, in particular ones that can handling variation across different topics, topics, and social contexts. She has held visiting appointments at Stanford, the Royal Institute of Technology, the École Normale Superieure, and the University of Canterbury.

Pierrehumbert is a Member of the National Academy of Sciences, a Fellow of the American Academy of Arts and Sciences, a Fellow of the Cognitive Science Society and a Fellow of the Linguistic Society of America. She won the Medal for Scientific Achievement of the International Speech Communication Association (ISCA) in 2020.

Human-centric Natural Language Processing for Social Good and Responsible Computing

Large language models (LLMs) constitute a paradigm shift in Natural Language Processing and its applications across all domains. To move towards human-centric NLP designed for social good and responsible computing, I argue we need knowledge-aware NLP systems and human-AI collaboration frameworks.  NLP systems that interact with humans need to be knowledge aware (e.g., linguistic, commonsense, sociocultural norms) and context aware (e.g., social, perceptual) so that they communicate better and in a safer and more responsible fashion with humans. Moreover, NLP systems should be able to collaborate with humans to create high-quality datasets for training and/or evaluating NLP models, to help humans solve tasks, and ultimately to align better with human values. In this talk, I will give a brief overview of my lab’s research around NLP for social good and responsible computing (e.g., misinformation detection, NLP for education and public health, building NLP technologies with language and culture diversity in mind). I will highlight key innovations on theory-guided and knowledge-aware models that allow us to address two important challenges: lack of training data, and the need to model commonsense knowledge. I will also present some of our recent work on human-AI collaboration frameworks for building high-quality datasets for various tasks such as generating visual metaphors or modeling cross-cultural norms similarities and differences. 

Smaranda Muresan is a Research Scientist at the Data Science Institute at Columbia University and an Amazon Scholar. Before joining Columbia, she was a faculty member in the School of Communication and Information at Rutgers University where she co-founded the Laboratory for the Study of Applied Language Technologies and Society. At Rutgers, she was the recipient of the Distinguished Achievements in Research Award. Her research focuses on human-centric Natural Language Processing for social good and responsible computing. She develops theory-guided and knowledge-aware computational models for understanding and generating language in context (e.g., visual, social, multilingual, multicultural) with applications to computational social science, education, and public health. Research topics that she worked on over the years include: argument mining and generation, fact-checking and misinformation detection, figurative language understanding and generation (e.g., sarcasm, metaphor, idioms), and multilingual language processing for low-resource and endangered languages. Recently, her research interests include explainable models and human-AI collaboration frameworks for high-quality datasets creation. She received best papers awards at SIGDIAL 2017 and ACL 2018 (short paper). She served as a board member for the North American Chapter of the Association for Computational Linguistics (NAACL) 2020-2021, as a co-founder and co-chair of the New York Academy of Sciences’ Annual Symposium on NLP/Dialog/Speech (2019-2020) and as a Program Co-Chair for SIGDIAL 2020 and ACL 2022.  

Misinformation, Harassment, and Violence through a Cybersecurity and Privacy Lens

Image

Technology companies play a central role in mediating online discourse and monitoring people's actions. Unfortunately, these products are spreading misinformation, harassment, and enabling violence. Currently, technology companies have struggled to mitigate these problems. In this talk, I will discuss how we use robust cybersecurity, data science, and independent data collection techniques to better understand these issues. I will show how this approach can illuminate the systemic incentives and design choices that likely contribute to unsafe technology products that are vulnerable to attacks. In addition, I will show how we can leverage those insights to design safer technology systems and improve resources for those targeted by these attacks. In cases where companies' interests are not aligned with their users, effecting changes that result in safer technology products often requires independent data collection and engaging with civil society, journalists, regulators, and policymakers.

Damon McCoy (he/she) is an Associate Professor of Computer Science and Engineering at New York University's Tandon School of Engineering and the co-director of Cyber Security for Democracy. Her research focuses on empirically understanding the security and privacy of technology systems and their intersection with society. In particular, he investigates problems through the lens of cybersecurity and privacy, such as hate, harassment, misinformation, and violence, that are more traditionally explored by social scientists. She is normally a down to earth person and only talks about herself in the third person when requested.

Multispectral Analysis and Deep Learning for Life Science and Biomedical Research

Several plant and animal species are more comprehensively understood by multispectral analysis. For example, ultraviolet fluorescence reveals original color patterns on colorless fossils for species classification, while Infrared imaging permits study of subsurface materials hidden under pigments. However, faded color, and material layers that exhibit subsurface scattering and spatially varying surface reflectance make it difficult to reconstruct the shape and appearance of biological materials. This talk presents a texture transfer framework that reconstructs invisible (or faded) appearance properties in organic materials with complex color patterns. I will motivate the project with a study that computes surface orientation (normals) at different material layers as a function of emission wavelength for effective scientific analysis in life science. Key contributions include a novel ultraviolet illumination system that records changing material property distributions, and a color reconstruction algorithm that uses spherical harmonics and principles from chemistry and biology to learn relationships between color appearance and material composition and concentration. Finally, I will explain a novel algorithm that extends the effective receptive field of a convolutional neural network for multi-scale detection of cancerous tumors in high resolution slide scans. The results permit efficient real-time analysis of medical images in pathology and biomedical research fields.

Corey Toler-Franklin is an Assistant Professor of Computer Science at the University of Florida where she directs the Graphics, Imaging & Light Measurement Laboratory. Dr. Toler-Franklin obtained a Ph.D. in Computer Science from Princeton University. She earned an M.S. degree from the Cornell University Program of Computer Graphics, and a B. Arch. degree from Cornell. Before joining UF faculty, Dr. Toler Franklin was a UC President's Postdoctoral Fellow at UC Davis, and a researcher at the UC Berkeley CITRIS Banatao Institute. She also held positions at Autodesk, Adobe, and Google.

Dr. Toler-Franklin’s research in computer graphics and vision includes Machine Learning, Data Acquisition, Appearance Modeling, Imaging Spectroscopy and Non-Photorealistic Rendering, with real-world applications in Life Science, Bio-Medical Research and Archaeology. Her algorithms use mathematical principles in optics to capture and analyze the shape and appearance of complex materials. Her recent work develops AI algorithms for biomedical research. Collaborating with the UF College of Medicine Oncology and Pathology Departments, and the UF Neuroscience Department, Dr. Toler-Franklin developed deep learning algorithms for diagnosing metastatic cancers and studying behaviors associated with neurological disorders (Alzheimer's and autism).

Actual Causality in Reactive Systems

Image

Counterfactual reasoning is an approach to infer the cause of an observed effect by comparing a given scenario in which the suspected cause and the effect are present, to the hypothetical scenarios where the suspected cause is not present. The seminal works of Halpern and Pearl have provided a definition of counterfactual causality for finite settings. In this talk, we propose an approach to check causality for reactive systems, i.e., systems that interact with their environment over a possibly infinite duration. First, we focus on finding causes for violations of hyperproperties. Hyperproperties, unlike trace properties, can relate multiple traces and thus express complex security properties. Here, the suspected cause is represented by a finite set of events occurring on the set of traces. Then, we lift Halpern and Pearl's definition to the case where the causes themselves (as well as effects) are omega-regular properties, not just sets of events. Given the causality algorithms, our tool HyperVis generates interactive visualizations of the given model, specification, and cause of the counterexample.

Niklas Metzger is a PhD student at CISPA Helmholtz Center for Information Security in Germany. He is advised by Bernd Finkbeiner and a member of the Reactive Systems Group. Before joining CISPA, he received his BSc and MSc at Saarland University in Germany. Niklas’ research focuses on compositional reactive synthesis guided by the principles of knowledge, actual causality in reactive systems, and building machine learning models as heuristics in complex formal method tasks.

Music Information Retrieval and Hit Songs