Computer Science Colloquia & Seminars are held each semester and sponsored by the Computer Science department. Faculty invite speakers from all areas of computer science, and the talks are open to all members of the RPI community.
Computer Science Colloquia & Seminars are held each semester and sponsored by the Computer Science department. Faculty invite speakers from all areas of computer science, and the talks are open to all members of the RPI community.
A quantum computer is as hard for us to build as for us to understand the algorithms they are built to run. In order to deliver asymptotic advantage over classical algorithms, quantum algorithms exploit inherently quantum phenomena – the ability for data to exist in a superposition of multiple states, exhibit constructive and destructive interference, and leverage the spooky phenomenon of entanglement. However, without appropriate and delicate manipulation of the quantum state stored by the computer, an implementation of an algorithm will produce incorrect outputs or lose its quantum computational advantage.
As a result, developers will face challenges when programming a quantum computer to correctly realize quantum algorithms. In this talk, I present these programming challenges and what we can do to overcome them. In particular, I address how basic programming abstractions – such as data structures and control flow – fail to work correctly on a quantum computer, and the progress we’ve made in re-inventing them to meet the demands of quantum algorithms.
Bio: Charles Yuan is a Ph.D. student at MIT working with Michael Carbin whose research interests lie in programming languages for quantum computation. His work has appeared in the ACM SIGPLAN POPL and OOPSLA conferences and has been recognized with the SIGPLAN Distinguished Artifact Award and the CQE-LPS Doc Bedard Fellowship.
Machine learning has revolutionized the usage of data, and proven of tremendous applicability due to its ability to find relations in data. One area of application is nanoscience, specifically, the investigation of monolayer protected nanoclusters (MPCs). Experimental research on MPCs requires expensive materials and equipment, and traditional computational research requires costly computation resources. ML is used in this context to alleviate the computational costs by utilizing distance-based regression models as surrogates for expensive Density Functional Theory-based calculations. Our research has so far primarily focused on feature selection, since MPCs provide high-dimensional data.
Bio: Joakim Linja received his Ph.D. degree in Mathematical Information Science from the University of Jyväskylä (JYU) in April 2023. He received his Masters degree in physics from JYU in 2017, specializing in nanoscience and computational science. He is currently working as a post doctoral scholar at JYU. His research interests lie in high-performance computing, machine learning, GPU computation, nanoscience and physics.
: Deep Learning techniques are underlying many amazing accomplishments in artificial intelligence and machine learning. Their theory does not match empirical achievements, but the applicable results have largely been in favor of DL. In our recently published paper [1], we question this belief. In the context of autoencoding, i.e., nonlinear dimension encoding-decoding, we propose a new, additive model that strictly separates approximation of bias, linear behavior, and nonlinear behavior. With this approximation, we encountered no help or even need of deeper network structures to encapsulate nonlinear behavior. We also witnessed worse data reconstruction results when typical data-batch driven optimization techniques were applied to train the additive autoencoder. It would be really an interesting endeavor to address the underlying reasons of the observed behavior of our extensive set of empirical experiments.
[1] Kärkkäinen, T., & Hänninen, J. (2023). Additive autoencoder for dimension estimation. Neurocomputing, Volume 551, 126520.
Biography: Tommi Kärkkäinen (TK) received the Ph.D. degree in Mathematical Information Technology from the University of Jyväskylä (JYU), in 1995. Since 2002 he has been serving as a full professor of Mathematical Information Technology at the Faculty of Information Technology (FIT), JYU. TK has led 50 different R&D projects and has been supervising over 60 PhD students. He has published over 200 peer-reviewed articles and received the Innovation Prize of JYU in 2010. He has served in many administrative positions at FIT and JYU, currently leading a Research Division and a Research Group on Human and Machine based Intelligence in Learning. The main research interests include data mining, machine learning, learning analytics, and nanotechnology. He is a senior member of the IEEE.
As a very hot topic today, near-data computing has a beautifully simple rationale: Moving computational tasks closer to where data reside could improve the overall system performance/efficiency. However, its large-scale commercial success has remained elusive so far, despite countless awesome research papers and 100s millions of dollars spent on its R&D. This disappointing status quo warrants doubts and skepticisms: Will it turn out to be a hype just like many others we have seen over the years? Are there any fatal flaws in this simple idea? Facing these questions, proponents of near-data computing must be brutally honest to themselves and humbly search for the (inconvenient) truth, other than conveniently blaming the industryÕs reluctance/laziness on embracing disruptive technologies. This talk will discuss the pitfalls of prior and on-going R&D efforts, and present the correct (or at least the most convenient) way to commence the commercialization journey of near-data computing. This talk will also show that there is still a huge space for research innovations in this area, despite intensive research over the past 20 years.
Bio: Tong Zhang is currently a Professor in the Electrical, Computer and Systems Engineering Department at Rensselaer Polytechnic Institute (RPI), NY. In 2002, he received the Ph.D. degree in electrical engineering from the University of Minnesota and joined the faculty of RPI. He has graduated 20 PhD students, and authored/co-authored over 160 papers, with citation h-index of 43. Among his research accomplishments, he made pioneering contributions to enabling the pervasive use of low-density parity-check (LDPC) code in commercial HDDs/SSDs and establishing the research area of flash memory signal processing. He co-founded ScaleFlux (San Jose, CA) to spearhead the commercialization of near-data computing, and currently serves as its Chief Scientist. He is an IEEE Fellow.
Machine Learning (ML) techniques, especially Deep Neural Networks (DNNs), have been driving innovations in many application domains. These breakthroughs are powered by the computational improvements in processor technology driven by Moore's Law. However, the need for computational resources is insatiable when applying ML to large-scale real-world problems. Energy efficiency is another major concern of large-scale ML. The enormous energy consumption of ML models not only increases costs in data-centers and decreases battery life of mobile devices but also has a severe environmental impact. Entering the post-Moore’s Law era, how to keep up performance and energy-efficiency with the scaling of ML remains challenging.
This talk addresses the performance and energy-efficiency challenges of ML. The core hypothesis can be encapsulated in a few questions. Do we need all the computations and data movements involved in conventional ML processing? Does redundancy exist at the hardware level? How can we better approach large-scale ML problems with new computing paradigms? This talk presents how to explore the elasticity in ML processing and hardware architectures: from the algorithm perspective, redundancy-aware processing methods are proposed for DNN training and inference, as well as large-scale classification problems and long-range Transformers; from the architecture perspective, balanced, specialized, and flexible designs are presented to improve efficiency.
Bio: Liu Liu is an Assistant Professor in the department of Electrical, Computer, and Systems Engineering at RPI. He has a Ph.D. in Computer Science at the University of California, Santa Barbara. His research interests reside in the intersection between computer architecture and machine learning, towards high-performance, energy-efficient, and robust machine intelligence.
As machine learning (ML) technologies get widely applied to many domains, it has become essential to rapidly develop and deploy ML models. Towards this goal, MLOps has recently emerged as a set of tools and practices for operationalizing production-ready models in a reliable and efficient manner. However, several open problems exist, including how to automate the ML pipeline that includes data collection, model training, and deployment (inference) with support for distributed data and models stored at multiple sites. In this talk, I will cover some theoretical foundations and practical approaches towards enabling distributed MLOps, i.e., MLOps in large-scale distributed systems. I will start with explaining the requirements and challenges. Then, I will describe how our recent theoretical developments in the areas of coreset, federated learning, and model uncertainty estimation can support distributed MLOps. As a concrete example, I will dive into the details of a federated learning algorithm with flexible control knobs, which adapts the learning process to accommodate time-varying and unpredictable resource availabilities, as often seen in systems in operation, while conforming to a given budget for model training. I will finish the talk by giving an outlook on some future directions.
Bio: Shiqiang Wang is a Staff Research Scientist at IBM T. J. Watson Research Center, NY, USA. He received his Ph.D. from Imperial College London, United Kingdom, in 2015. His current research focuses on the intersection of distributed computing, machine learning, networking, and optimization, with a broad range of applications including data analytics, edge-based artificial intelligence (Edge AI), Internet of Things (IoT), and future wireless systems. He received the IEEE Communications Society (ComSoc) Leonard G. Abraham Prize in 2021, IEEE ComSoc Best Young Professional Award in Industry in 2021, IBM Outstanding Technical Achievement Awards (OTAA) in 2019, 2021, and 2022, and multiple Invention Achievement Awards from IBM since 2016. For more details, please visit his homepage at: https://shiqiang.wang
Vasundhara Acharya
Advisor: Prof. Bulent Yener
Title: Tuberculosis Prediction from Lung Tissue Images of Diversity Outbred Mice using Cell Graph Neural Network
Lorson Blair
Advisor: Prof. Stacy Patterson
Title: A Continuum Approach for Collaborative Task Processing in UAV MEC Networks
Jesse Ellin
Advisor: Prof. Alex Gittens
Title: Knowledge Graph Anomaly Detection via Probabilistic GANs
Shawn George
Advisor: Prof. Konstantin Kuzmin
Title: Synergy: Abstract2Gene
William Hawkins
Advisor: Prof. George Slota
Title: Accelerating Graph Neural Network Training using Dynamic Mode Decomposition
Neha Deshpande
Advisor: Prof. Chuck Stewart
Title: Tusk Detection for Elephant Re-Identification
Ian Conrad
Advisor: Prof. Sibel Adali
Title: Contextualized Moral Foundations Analysis
Ruixiong Hu
Advisor: Mark Shephard
Title: Mesh Adaptation in Multilayer Laser Powder Bed Fusion
Ashley Choi
Advisor: Prof. Sibel Adali
Title: News Story Collection API and Visualization
Ohad Nir
Advisor: Prof. Chuck Stewart
Title: Detection of Capuchin Monkeys
Connor Wooding
Advisor: Prof. George Slota
Title: GPU Parallelization for Biconnectivity Algorithms
Roman Nett
Advisor: Prof. Bulent Yener
Title: Graph Neural Network Using Local Cell Graph Features for Cancer Classification
Andy Bernhardt
Advisor: Prof. Tomek Strzalkowski
Title: Imageability as an Indicator of Authorship
Jacy Sharlow
Advisor: Prof. Barb Cutler
Title: Automating the Artistic Pipeline Regarding Skin Wrinkling in the Geometric Space
Steven Laverty
Advisor: Prof. Mohammed Zaki
Title: Protein Folding with Deep RL
Zachary Fernandes
Advisor: Prof. Mei Si
Title: Investigating the Impact of Self-Attention on Reinforcement Learning
Seth Laurenceau
Advisor: Prof. Ana Milanova
Title: Verification of Python Docstrings
Ryan Kaplan
Advisor: Prof. Alex Gittens
Title: Transfer Learning on Images for Graph Problems
Mohammed Shahid Modi
Advisor: Prof. Bolek Szymanski
Title: Poster on Dynamics of Ideological Bias Shifts of Users on Social Media Platforms
Dhruva Hiremagalur Narayan
Advisor: Prof. Mohammed Zaki
Title: Understanding forms using graph neural networks
Harshaa Hiremagalur Narayan
Advisor: Prof. Mohammed Zaki
Title: Using BERT-GCN on embeddings created using dictionary word
Artificial intelligence (AI) has become woven into therapeutic discovery to accelerate drug discovery and development processes since the emergence of deep learning. For drug discovery, the goal is to identify drug molecules with desirable pharmaceutical properties. I will discuss our deep generative models that relax the discrete molecule space into a differentiable one and reformulate the combinatorial optimization problem into a differentiable optimization problem, which can be solved efficiently. On the other hand, drug development focuses on conducting clinical trials to evaluate the safety and effectiveness of the drug on human bodies. To predict clinical trial outcomes, I design deep representation learning methods to capture the interaction between multi-modal clinical trial features (e.g., drug molecules, patient information, disease information), which achieves 0.847 F1 score in predicting phase III approval. Finally, I will present my future works in geometric deep learning for drug discovery and predictive model for drug development.
Bio: Tianfan Fu is a Ph.D. candidate in the School of Computational Science and Engineering at the Georgia Institute of Technology, advised by Prof. Jimeng Sun. His research interest lies in machine learning for drug discovery and development. Particularly, he is interested in generative models on both small-molecule & macro-molecule drug design and deep representation learning on drug development. The results of his research have been published in leading AI conferences, including AAAI, AISTATS, ICLR, IJCAI, KDD, NeurIPS, UAI, and top domain journals such as Nature, Cell Patterns, Nature Chemical Biology, and Bioinformatics. His work on clinical trial outcome prediction has been selected as the cover paper on Cell Patterns. In addition, Tianfan is an active community builder. He co-organized the first three AI4Science workshop on leading AI conferences (https://ai4sciencecommunity.github.io/); he co-founded Therapeutic Data Commons (TDC) initiative (https://tdcommons.ai/), an ecosystem with AI-solvable tasks, AI-ready datasets, and benchmarks in therapeutic science. Additional information is available at https://futianfan.github.io/.
The rapid progress of deep learning in recent years has led to significant advances in various fields such as computer vision, natural language processing, and speech recognition. The success of deep learning models heavily relies on the availability of large-scale and high-quality datasets. To address this challenge, active learning is a representative strategy that interactively queries human annotators for efficient data annotation. I will discuss how to design both theoretically and empirically effective active learning strategy for deep neural networks. On the other hand, powerful deep learning models have the potential to create high-quality data for human needs nowadays. I will demonstrate this by exploring recent advancements in generative methods. By taking the neural style transfer problem as an example, I will discuss how to achieve a desirable balance between content, style, and visual quality when creating visual contents. I will also share potential future directions of data-aspect AI, as well as the applications to biomedical domains.
Bio: Dr. Siyu Huang is a postdoctoral fellow at the John A. Paulson School of Engineering and Applied Sciences, Harvard University. He received his B.E. and Ph.D. degrees from Zhejiang University in 2014 and 2019, respectively. Prior to joining Harvard, he was a visiting scholar at Carnegie Mellon University in 2018, a research scientist at Baidu Research from 2019 to 2021, and a research fellow at Nanyang Technological University in 2021. His research interests include computer vision, deep learning, and generative AI, with 30 publications on top-tier conferences and journals.
Program specifications provide clear and precise descriptions of behaviors of a software system, which serves as a blueprint for its design and implementation. They help ensure that the system is built correctly and the functions work as intended, making it easier to troubleshoot, modify, and verify the system if needed. NIST suggests that the lack of high-quality specifications is the most common cause of software project failure. Nowadays, successful projects have an equal or even higher number of specifications than code (counted by lines).
In this talk, I will present my research on synthesizing both informal and formal specifications for software systems. I will explain how we use a combination of program and natural language semantics to automatically generate informal specifications, even for native methods without implementation in Java which previous methods could not handle. By leveraging the generated specifications, we successfully detect many code bugs and code-comment inconsistencies. Additionally, I will describe how we derive formal specifications from natural language comments using a search-based technique. The generated formal specifications have been applied to facilitate program analysis for existing tools. They have been shown to greatly improve the capabilities of these tools, by detecting many new information leaking paths and reducing false alarms in testing. Overall, the talk will highlight the importance of program specifications in software engineering and demonstrate the potential of our techniques to improve the development and maintenance of software systems.
Bio:
Juan Zhai is an Assistant Teaching Professor in the Department of Computer Science at Rutgers University. Previously, she was a Postdoctoral Research Associate, working with Prof. Xiangyu Zhang in the Department of Computer Science at Purdue University. She also worked as a tenure-track Assistant Professor at Nanjing University, where she obtained her Ph.D. degree. Her research interests lie in software engineering, natural language processing, and security, focusing on specification synthesis and enforcement. She is the recipient of the Distinguished Paper Award of USENIX Security 2017 and the Outstanding Doctoral Student Award in NASAC 2016.
While decades of AI research on high-level reasoning have yielded many techniques for many tasks, we are still quite far from having artificial agents that can just “sit down" and perform tasks like intelligence tests without highly specialized algorithms or training regimes. We also know relatively little about how and why different people approach reasoning tasks in different (often equally
successful) ways, including in neurodivergent conditions such as autism. In this talk, I will discuss: 1) my lab's work on AI approaches for reasoning with visual imagery to solve intelligence tests, and what these findings suggest about visual cognition in autism; 2) how imagery-based agents might learn their domain knowledge and problem-solving strategies via search and experience, instead of these components being manually designed, including recent leaderboard results on the very difficult Abstraction & Reasoning Corpus (ARC) ARCathon challenge; and 3) how this research can help us understand cognitive strategy differences in people, with applications related to neurodiversity and employment. I will also discuss 4) our Film Detective game that aims to visually support adolescents on the autism spectrum in improving their theory-of-mind and social reasoning skills.
Bio: Maithilee Kunda is an assistant professor of computer science at Vanderbilt University. Her work in AI, in the area of cognitive systems, looks at how visual thinking contributes to learning and intelligent behavior, with a focus on applications related to autism and neurodiversity. She directs Vanderbilt’s Laboratory for Artificial Intelligence and Visual Analogical Systems and is a founding investigator in Vanderbilt’s Frist Center for Autism & Innovation.
She has led grants from the US National Science Foundation and the US Institute of Education Sciences and has also collaborated on large NSF Convergence Accelerator and AI Institute projects. She has published in Proceedings of the National Academy of Sciences (PNAS) and in the Journal of Autism and Developmental Disorders (JADD), the premier journal for autism research, as well as in AI and cognitive science conferences such as ACS, CogSci, AAAI, ICDL-EPIROB, and DIAGRAMS, including a best paper award at the ACS conference in 2020. Also in 2020, her research on innovative methods for cognitive assessment was featured on the national news program CBS 60 Minutes, as part of a segment on neurodiversity and employment. She holds a B.S. in mathematics with computer science from MIT and Ph.D. in computer science from Georgia Tech.
Many learning tasks in Artificial Intelligence require dealing with graph data, ranging from biology and chemistry to finance and education. As powerful learning tools for graph inputs, graph neural networks (GNNs) have demonstrated remarkable performance in various applications. Despite their success, unlocking the full potential of GNNs requires tackling the limitations of robustness and scalability. In this talk, I will present a fresh perspective on enhancing GNNs by optimizing the graph data, rather than designing new models. Specifically, first, I will present a model-agnostic framework which improves prediction performance by enhancing the quality of an imperfect input graph. Then I will show how to significantly reduce the size of a graph dataset while preserving sufficient information for GNN training.
My ultimate research vision is to develop an AI model that can emulate human reasoning and thinking, which requires building a differentiable Neural-Symbolic AI. This approach involves enabling neural models to interact with external symbolic modules, such as knowledge graphs, logical engines, math calculators, and physical/chemical simulators. This will facilitate end-to-end training of such a Neural-Symbolic AI system without annotated intermediate programs.
During this talk, I will introduce my two research endeavors focused on building differentiable neural symbolic AI using knowledge graphs. Firstly, I will discuss how Symbolic Reasoning can help Neural Language Models. I designed OREO-LM, which incorporates knowledge graph relational reasoning into a Large Language Model, significantly improving multi-hop question answering using a single model. Secondly, I will discuss how Neural Embedding can help Symbolic Logic Reasoning. I solve complex first-order logic queries in neural embedding space, using fuzzy logic operators to create a learning-free model that fulfills all logic axioms. Finally, I will discuss my future research plans on applying differentiable neural-symbolic AI to improve program synthesis, architecture design, and scientific discovery.
Bio: Ziniu Hu is a fifth-year PhD student in computer science at UCLA. His research focuses on integrating symbolic knowledge reasoning with neural models. Under the guidance of Professors Yizhou Sun and Kai-Wei Chang, he has developed several models that have successfully solved complex question-answering and graph mining problems. His research has received support from Baidu Ph.D. Fellowship and Amazon Ph.D. Fellowship. He also contributed to the research community as the research-track workflow co-chair for KDD'23 and was awarded the top reviewer at NeurIPS'22. His research has been deployed on various industrial applications, including Tiktok unbiased Recommendation, Google YouTube Shorts recommendation, Microsoft Graph anomaly detection, and Facebook hate speech detection service. His research has received several awards, including the best paper award at WWW'19, the best student paper award at DLG-KDD'20 workshop, and the best paper award at SoCal-NLP'22.
Traditionally, multimodal information consumption has been entity-centric with a focus on concrete concepts (such as objects, object types, physical relations, e.g., a person in a car), but lacks ability to understand abstract semantics (such as events and semantic roles of objects, e.g., driver, passenger, mechanic). However, such event-centric semantics are the core knowledge communicated, regardless whether in the form of text, images, videos, or other data modalities.
At the core of my research in Multimodal Information Extraction (IE) is to bring such deep semantic understanding ability to the multimodal world. My work opens up a new research direction Event-Centric Multimodal Knowledge Acquisition to transform traditional entity-centric single-modal knowledge into event-centric multi-modal knowledge. Such a transformation poses two significant challenges: (1) understanding multimodal semantic structures that are abstract (such as events and semantic roles of objects): I will present my solution of zero-shot cross-modal transfer (CLIP-Event), which is the first to model event semantic structures for vision-language pretraining, and supports zero-shot multimodal event extraction for the first time; (2) understanding long-horizon temporal dynamics: I will introduce Event Graph Model, which empowers machines to capture complex timelines, intertwined relations and multiple alternative outcomes. I will also show its positive results on long-standing open problems, such as timeline generation, meeting summarization, and question answering. Such Event-Centric Multimodal Knowledge Acquisition starts the next generation of information access, which allows us to effectively access historical scenarios and reason about the future. I will lay out how I plan to grow a deep semantic understanding of language world and vision world, moving from concrete to abstract, from static to dynamic, and ultimately from perception to cognition.
Bio: Manling Li is a Ph.D. candidate at the Computer Science Department of University of Illinois Urbana-Champaign. Her work on multimodal knowledge extraction won the ACL'20 Best Demo Paper Award, and the work on scientific information extraction from COVID literature won NAACL'21 Best Demo Paper Award. She was a recipient of Microsoft Research PhD Fellowship in 2021. She was selected as a DARPA Riser in 2022, and a EE CS Rising Star in 2022. She was awarded C.L. Dave and Jane W.S. Liu Award, and has been selected as a Mavis Future Faculty Fellow. She led 19 students to develop the UIUC information extraction system and ranked 1st in DARPA AIDA evaluation in 2019 and 2020. She has more than 30 publications on multimodal knowledge extraction and reasoning, and gave tutorials about event-centric multimodal knowledge at ACL'21, AAAI'21, NAACL'22, AAAI'23, etc. Additional information is available at https://limanling.github
Today connected devices, as well as various smart sectors generate a significant amount of data. Tailoring machine learning algorithms to exploit this massive amount of data can lead to many new applications and enable ambient intelligence. The question is how to use this decentralized data to enhance the system intelligence beneficial for everyone while protecting the sensitive information. It is not desirable to offload such massive amounts of data available at the edge devices to a cloud server for centralized processing due to storage, latency, bandwidth, and power constraints, as well as privacy concerns of users. Furthermore, due to the growing storage and computational capabilities of the edge devices, it is increasingly attractive to store and process the data locally by shifting network computations to the edge. This enables decentralized intelligence where local computations on the data converts decentralized data to a global intelligence; hence, enhancing data privacy while learning from the collection of data available across the network. In this talk, I highlight some of the challenges and advances in enabling decentralized intelligence by integrating computations, collaboration, and communications, the three essential components of enabling collective intelligence.
Bio: Mohammad received the B.Sc. degree in Electrical Engineering from the Iran University of Science and Technology in 2011 and the M.Sc. degree in Electrical and Computer Engineering from the University of Tehran in 2014, both with the highest rank in classes. He also obtained the Ph.D. degree in Electrical and Electronic Engineering at Imperial College London in 2019. He then spent two years as a Postdoctoral Research Associate in the Department of Electrical and Computer Engineering at Princeton University. He is currently a Postdoctoral Associate at MIT where he joined in early 2022. He received the Best Ph.D. Thesis Award from the Department of Electrical and Electronic Engineering at Imperial College London, as well as the IEEE Information Theory Chapter of UK and Ireland in the year 2019. He is also the recipient of the IEEE Communications Society Young Author Best Paper Award (2022) for the paper titled "Federated learning over wireless fading channels". His research interests include machine learning, information theory, distributed computing, privacy and security, and data science.
The fuel of machine learning models and algorithms is the data usually collected from users, enabling refined search results, personalized product recommendations, informative ratings, and timely traffic data. However, increasing reliance on user data raises serious challenges. A common concern with many of these data-intensive applications centers on privacy — as a user’s data is harnessed, more and more information about her behavior and preferences is uncovered and potentially utilized by platforms and advertisers. These privacy costs necessitate adjusting the design of data markets to include privacy-preserving mechanisms.
This talk establishes a framework for collecting data of privacy-sensitive strategic users for estimating a parameter of interest (by pooling users' data) in exchange for privacy guarantees and possible compensation for each user. We formulate this question as a Bayesian-optimal mechanism design problem, in which an individual can share her data in exchange for compensation but at the same time has a private heterogeneous privacy cost which we quantify using differential privacy. We consider two popular data market architectures: central and local. In both settings, we use Le Cam's method to establish minimax lower bounds for the estimation error and derive (near) optimal estimators for given heterogeneous privacy loss levels for users. Next, we pose the mechanism design problem as the optimal selection of an estimator and payments that elicit truthful reporting of users' privacy sensitivities. We further develop efficient algorithmic mechanisms to solve this problem in both privacy settings.
Finally, we consider the case that users are interested in learning different personalized parameters. In particular, we highlight the connections between this problem and the meta-learning framework, allowing us to train a model that can be adapted to each user's objective function.
Bio: Alireza Fallah is a Ph.D. candidate at the department of Electrical Engineering and Computer Science (EECS) and the Laboratory for Information and Decision Systems (LIDS) at Massachusetts Institute of Technology (MIT). His research interests are machine learning theory, data market and privacy, game theory, optimization, and statistics. He has received a number of awards and fellowships, including the Ernst A. Guillemin Best MIT EECS M.Sc. Thesis Award, Apple Scholars in AI/ML Ph.D. fellowship, MathWorks Engineering Fellowship, and Siebel Scholarship. He has also worked as a research intern at the Apple ML privacy team. Before joining MIT, he earned a dual B.Sc. degree in Electrical Engineering and Mathematics from Sharif University of Technology, Tehran, Iran.
Addressing the performance gap between software and hardware is one of the major challenges in computer science and engineering. Software stacks and optimization approaches have long been designed targeting regular programs — programs that operate over regular data structures such as arrays and matrices using loops, partly due to the abundance of regular programs in computer software. But irregular programs — programs that traverse over irregular or pointer-based data structures such as sparse matrices, trees, and graphs using a mix of recursion and loops — also appear in many essential applications such as simulation, data mining, graphics, etc. Loop transformation frameworks are good examples of performance-enhancing scheduling transformations for regular programs. Generally, these frameworks reason about transformations in a composable manner (i.e., reason about a sequence of transformations).
In the past, scheduling transformations for irregular programs were ad-hoc, and they were considered on the horizon by loop transformation frameworks. Even the few existing ones were applied in isolation, and the composability of these transformations was not studied extensively. In this talk, I will discuss a composable framework for verifying the correctness of scheduling transformations for irregular programs. We will explore the abstractions used in different parts of our framework, and I will show ways to extend these abstractions to capture a wide variety of scheduling transformations for irregular programs. Finally, I will discuss future directions on incorporating dependence analyses and data layout abstractions into this framework.
Bio: Kirshanthan (“Krish”) Sundararajah is a PhD candidate in the Elmore Family School of Electrical and Computer Engineering, advised by Milind Kulkarni. He earned his Bachelor's degree from the University of Moratuwa, Sri Lanka, and his Master’s degree from Purdue University. His research interests lie in the areas of compilers, programming languages, and high-performance computing. He is particularly interested in solving the performance challenges of irregular applications. He has published in top conferences such as ASPLOS, OOPSLA, and PLDI and is a recipient of the Bilsland Dissertation Fellowship.
Neural network models have been pushing computers’ capacity limit on natural language understanding and generation while lacking interpretability. The black-box nature of deep neural networks hinders humans from understanding their predictions and trusting them in real-world applications. In this talk, I will introduce my effort in bridging the trustworthy gap between models and humans by developing interpretation techniques, which cover three main phases of a model life cycle—training, testing, and debugging. I will demonstrate the critical values of integrating interpretability into every state of model development: (1) making model prediction behavior transparent and interpretable during training; (2) explaining and understanding model decision-making on each test example; (3) diagnosing and debugging models (e.g., robustness) based on interpretations. I will discuss future directions on incorporating interpretation techniques with system development and human interaction for long-term trustworthy AI.
Bio: Hanjie Chen is a Ph.D. candidate in Computer Science at the University of Virginia. Her research interests lie in Trustworthy AI, Natural Language Processing (NLP), and Interpretable Machine Learning. She is a recipient of the Carlos and Esther Farrar Fellowship and the Best Poster Award at the ACM CAPWIC 2021. Her work has been published at top-tier NLP/AI conferences (e.g., ACL, AAAI, EMNLP, NAACL) and selected by the National Center for Women & Information Technology (NCWIT) Collegiate Award Finalist 2021. Besides, as the primary instructor, she co-designed and taught a cross-listed course, CS 4501/6501 Interpretable Machine Learning, at UVA. Her effort in teaching was recognized by the UVA CS Outstanding Graduate Teaching Award and University-wide Graduate Teaching Awards Nominee (top 5% of graduate instructors).
The vast amount of digital data we create and collect has revolutionized many scientific fields and industrial sectors. Yet, despite our success in harnessing this transformative power of data, computational and societal trends emerging from the current practices of data science necessitate upgrading our toolkit for data analysis. In this talk, we discuss how practical considerations such as privacy and memory limits affect statistical inference tasks. In particular, we focus on two examples: First, we consider hypothesis testing with privacy constraints. More specifically, how one can design an algorithm that tests whether two data features are independent or correlated with a nearly-optimal number of data points while preserving the privacy of the individuals participating in the data set. Second, we study the problem of entropy estimation of a distribution by streaming over i.i.d. samples from it. We determine how bounded memory affects the number of samples we need to solve this problem.
Bio: Maryam Aliakbarpour is a postdoctoral researcher at Boston University and Northeastern University, where she is hosted by Prof. Adam Smith and Prof. Jonathan Ullman. Before that, she was a postdoctoral research associate at the University of Massachusetts Amherst, hosted by Prof. Andrew McGregor (from Fall 2020-Summer 2021). In Fall 2020, she was a visiting participant in the Probability, Geometry, and Computation in High Dimensions Program at the Simons Institute at Berkeley. Maryam received her Ph.D. in September 2020 from MIT, where she was advised by Prof. Ronitt Rubinfeld. Maryam was selected for the Rising Stars in EECS in 2018 and won the Neekeyfar Award from the Office of Graduate Education, MIT.
Learning-based software and systems are deeply embedded in our lives. However, despite the excellent performance of machine learning models on benchmarks, state-of-the-art methods like neural networks often fail once they encounter realistic settings. Since neural networks often learn correlations without reasoning with the right signals and knowledge, they fail when facing shifting distributions, unforeseen corruptions, and worst-case scenarios. In this talk, I will show how to build reliable and robust machine learning by tightly integrating context into the models. The context has two aspects: the intrinsic structure of natural data, and the extrinsic structure of domain knowledge. Both are crucial: By capitalizing on the intrinsic structure in natural images, I show that we can create robust computer vision systems, even in the worst case, an analytical result that also enjoys strong empirical gains. Through the integration of external knowledge, such as causal structure, my framework can instruct models to use the right signals for visual recognition, enabling new opportunities for controllable and interpretable models. I will also talk about future work in making machine learning robust, which I hope to transform us into an intelligent society.
Bio: Chengzhi Mao is a final-year Ph.D. student from the Department of Computer Science at Columbia University. He is advised by Prof. Junfeng Yang and Prof. Carl Vondrick. He received his B.S in E.E. from Tsinghua University. His research focuses on reliable and robust machine learning. His work has led to over ten publications and Orals at top conferences, which established a new generalization of robust models beyond feedforward inference. His work also connects causality to the vision domain. He serves as reviewers for several top conferences, including CVPR, ICCV, ECCV, ICLR, NeurIPS, IJCAI, and AAAI.
A framework for studying the percolation theory of interdependent networks will be presented. In interdependent networks, such as infrastructures, when nodes in one network fail, they cause dependent nodes in other networks to also fail. This may happen recursively and can lead to a cascade of failures and to a sudden abrupt fragmentation of the system of interdependent systems. This is in contrast to a single network where the fragmentation percolation transition due to failures is continuous. I will present analytical solutions based on percolation theory for the critical thresholds, cascading failures, and the giant functional component of a network of n interdependent networks. I will show that the general theory shows many novel processes and features that are not present in the percolation theory of single networks. I will also show that interdependent networks embedded in space are significantly more vulnerable, and the phase transition is much richer compared to non-embedded networks. In particular, small localized attacks of zero fraction but above a critical size may lead to cascading failures that dynamically propagate and yield an abrupt phase transition. I will finally discuss the consequences of the behavior of percolation of interdependent networks on phase transitions in real physical interdependent systems. I will discuss the recent theory and experiments on interdependent superconducting networks where we identified a novel abrupt transition, although each isolated system shows a continuous transition.
References:
[1] S. Buldyrev, G. Paul, H.E. Stanley, S. Havlin, Nature, 464, 08932 (2010).
[2] Jianxi Gao, S. Buldyrev, H. E. Stanley, S. Havlin, Nature Physics, 8, 40 (2012).
[3] A. Bashan et al, Nature Physics, 9, 667 (2013) [4] A Majdandzic et al, Nature Physics 10 (1), 34 (2014); Nature Comm. 7, 10850 (2016) [5] M. Danziger et al, Nature Physics 15(2), 178 (2019) [6] I Bonamassa et al, Interdependent superconducting networks, preprint arXiv:2207.01669 (2022) [7] B. Gross et al, arXiv:2208.00440, PRL, in press (2022)
Bio:
Professor Shlomo Havlin has made fundamental contributions to the physics of complex systems and statistical physics. These discoveries have impacted many other fields, such as medicine, biology, geophysics, and more. He has over 60,000 citations on ISI Web of Science and over 100,000 in Google Scholar. His h-index is 112 (142) in Web of Science (Google Scholar). Professor Havlin has been a Highly Cited Scientist in the last 3 years.
He is a professor in the Physics Department at Bar-Ilan University. He received his PhD in 1972 from Bar Ilan University, and he has been a professor at BIU since 1984. Also, between the years of 1999 to 2001, he was the Dean of the Faculty of Exact Sciences, and from 1996 to 1999, he was the President of the Israel Physical Society. Professor Havlin is an IOP Honorary Fellow (England, 2021). He won the Senior Scientific Award of the International Complex Systems Society, the Israel prize in Physics (2018), Order of the Star of Italy, President of Italy (2017), the Rothschild Prize for Physical and Chemical Sciences, Israel (2014), the Lilienfeld Prize for "a most outstanding contribution to physics," APS, USA (2010), the Humboldt Senior Award, Germany (2006), the Distinguished Scientist Award, Chinese Academy of Sciences (2017), the Weizmann Prize for Exact Sciences, Israel (2009), the Nicholson Medal, American Physical Society, USA (2006), and many others.
Professor Havlin has been a leading pioneer in the development of network science, with over 800 papers and books in the fields of statistical physics, network science, and interdisciplinary physical sciences. His main research interests in the last 12 years have focused on interdependent networks, cascading failures, networks of networks, and their implications to real-world problems. The real-world systems applications include physiology, climate, infrastructures, finance, traffic, earthquakes, and others.
Abstract: The head of the Institute of Education Sciences has asked how we can use platform<https://www.the74million.org/article/schneider-garg-medical-researchers-find-cures-by-conducting-many-studies-and-failing-fast-we-need-to-do-the-same-for-education/>s to increase education sciences. We at ASSISTments have been addressing this<https://www.the74million.org/article/heffernan-how-can-we-know-if-ed-tech-works-by-encouraging-companies-researchers-to-share-data-government-funding-can-help/> very question. How do platforms like EdX, Khan Academy, or Canvas improve science? There is a crisis in American science referred to as the Reproducibility Crisis where many experimental results cannot be reproduced. We are trying to address that crisis by helping “good science” be done. People who control platforms have a responsibility to try to make them useful tools for learning what works. In Silicon Valley, every company is doing AB Testing to refine their individual products. That, in and of itself, is a good thing and we should use these platforms to figure out how to make them more effective. One of the ways we should do that is by experimenting with different ways of helping students succeed. ASSISTments<http://www.assistments.org/>, a platform I have created with 450,000 middle-school math students, is used to help scientists run studies. I will explain how we have over 100 experiments running inside the ASSISTments platform and how the ASSISTmentsTestBed.org allows external researchers to propose studies. I will also explain how proper oversight is done by our Institutional Review Board. Further, I will explain how users of this platform agree ahead of time to OpenScience procedures such as open-data, open-materials and pre-registration. I’ll illustrate some examples with the twenty-four randomized controlled trials<https://www.etrialstestbed.org/> that I have published as well as the three studies that have more recently come out from the platform by others. Finally, I will point to how we are anonymizing our data and how over 34 different external researchers have used our datasets to publish scientific studies<https://sites.google.com/site/assistmentsstudies/useourdata>. I would like to thank the U.S. Department of Education and the National Science Foundation for their support of over $32 million from 40+ grants.
Bio: Neil Heffernan is William Smith Dean's Professor of Computer Science and Director of the Learning Sciences & Technology program at Worcester Polytechnic Institute. He co-founded ASSISTments, a web-based learning platform, which he developed not only to help teachers be more effective in the classroom, but also so that he could use the platform to conduct studies to improve the quality of education. Professor Heffernan is passionate about educational data mining and enjoys supervising WPI students, helping them create ASSISTments content and features. Several student projects have resulted in peer-reviewed publications that compare different ways to optimize student learning. Professor Heffernan's goal is to give ASSISTments to millions across the U.S. and internationally as a free service.
Muhammad Saad Atique
Taming Uncertainty in Social Networks
Advisor: Prof. Bolek Szymanski
Sahith Bhamidipati
Generating Synthetic Election Data
Advisor: Prof. Lirong Xia
Matthew Crotty
Determining Image Hardness using Ensemble Classifier Method
Advisor: Prof. Radoslav Ivanov
Olivia Lundelius
An Analysis of Improved Daltonization Algorithms
Advisor: Prof. Barb Cutler
Nicholas Lutrzykowski
Driver for Seizure Prediction Using EEG Data
Advisor: Prof. Bulent Yener
Richard Pawelkiewicz
Action at a Distance
Advisor: Prof. Bulent Yener
Noah Prisament
A Variation on the Hotelling-Downs Model with Facility Synergy: the Mall Effect
Advisor: Prof. Elliot Anshelevich
Jeff Putlock
Overcoming Missing Data and Labels During VFL
Advisor: Prof. Stacy Patterson
Vijay Sadashivaiah
Towards Explainable Transfer Learning
Advisor: Prof. Jim Hendler
Bishwajit Saha
Clustering with Associative Memories
Advisor: Prof. Mohammed Zaki
Mara Schwartz
Topical Analysis of Twitter User Clusters
Advisor: Prof. Tomek Strzalkowski
Xiao Shou
Event Former: A Self-Supervised Learning Paradigm for Temporal Point Process
Advisor: Prof. Kristin Bennett
The presentation considers computer vision, especially a point of view of applications. Digital image processing and analysis with machine learning methods enable efficient solutions for various areas of useful data-centric engineering applications. Challenges with domain adaptation, active learning, open set classification, and metric learning of similarities are considered. Computer Vision and Pattern Recognition Laboratory at LUT focuses on industrial computer vision, biomedical engineering, social signal processing, and data analytics. Different applications are given as examples based on specific data: fundus images in diagnosis of diabetic retinopathy, planktons in the Baltic Sea, Saimaa ringed seals in the Lake Saimaa, and logs in the sawmill industry.
Bio: Heikki Kälviäinen has been a Professor of Computer Science and Engineering since 1999. He is the head of the Computer Vision and Pattern Recognition Laboratory (CVPRL) at the Department of Computational Engineering of Lappeenranta-Lahti University of Technology LUT, Finland. He received his D.Sc. (Tech.) degree in Computer Science and Engineering in 1994 from the Department of Information Technology of LUT. Prof. Kälviäinen's research interests include computer vision, machine vision, pattern recognition, machine learning, and digital image processing and analysis.
Determining an ideal ordering for a sparse matrix is difficult. In the case of finding an ordering to minimize fill-in for factorization of a general sparse matrix, the problem is NP-Hard. Meanwhile, the selection of an ordering for an iterative method, such as Conjugate Gradients, that reduces iteration counts and efficiently uses the cache hierarchy is based on observations and rules of thumb. The selection of an ordering is exacerbated in applications that solve a series of sparse matrix problems that change over time, e.g., those found in some circuit simulations. Modern computing systems tend to be heterogeneous containing accelerators, such as a GPU or neural device, that can be co-scheduled with the main CPU. These accelerators are ideal for the computation of complex artificial neural networks. This work uses these accelerators to construct a neural network model that approximates an ordering for a given sparse matrix. This approximation model acts at accelerating the selection of an ordering during the application. Moreover, a trained model can be used iteratively in the case of applications where sparse matrices evolve during execution to determine if a new ordering should be implemented at some point in the computation to speed up the application.
Speaker Bio:
Joshua Booth is an Assistant Professor of Computer Science at the University of Alabama in Huntsville. He received his Ph.D. in Computer Science and Engineering at The Pennsylvania State University and was a Post-Doctoral Researcher in the Scalable Algorithm Division at Sandia National Laboratories where he constructed new sparse linear solver methods for their exascale computing initiative. Dr. Booth is deeply committed to teaching emerging computing systems and technologies to future generations and spent several years teaching at the top liberal art colleges of Bucknell University and Franklin & Marshall College. His research focuses on high-performance computing related to scalable sparse linear algebra, scheduling, resiliency, and system performance.
His contributions and potential have been recognized via being awarded a bronze-level ACM Graduate Student Research Award for his work on multilevel preconditioning and an NSF CAREER award related to the neural acceleration of irregular applications. Additionally, Dr. Booth is an active member of the community leading the Huntsville Computer Research Seminar and reviewing for numerous ACM and IEEE journals
The human mind is a remarkable thing. On the one hand, even "mundane" aspects of cognition, such as our ability to fold clean laundry, far outstrip the capabilities of the most advanced robotic systems. On the other hand, we know from decades of research in psychology that the human mind is also astonishingly limited. Human minds face strict limits on attention, working memory, perception, and other basic cognitive faculties. In this talk, I will discuss a framework for resolving this apparent tension that spans the fields of cognitive science and artificial intelligence, known as computational rationality. According to this framework, the mind utilizes computations and algorithms that maximize the expected utility of behavior, while subject to constraints on the ability to store, manipulate, and process information. In my research, I have focused primarily on the use of rate-distortion theory (a sub-field of information theory) to characterize these limits on human information processing. The approach will be demonstrated through computational models of perception, memory, decision making, and reinforcement learning.
Bio: Chris Sims received a B.S. in computer science from Cornell University (2003), followed by a Ph.D. in Cognitive Science from Rensselaer Polytechnic Institute (2009). After completing his Ph.D., Dr. Sims held a postdoctoral research position at the University of Rochester, and a faculty position at Drexel University before joining the faculty at RPI in 2017. Dr. Sims's overarching research interest lies in developing computational models of human intelligence. Specific research areas include reinforcement learning, visual memory and perceptual expertise, sensory-motor control and motor learning, and learning and decision-making under uncertainty.
The proliferation of mobile devices with sensing and tracking capabilities, cloud computing, smart home, and the internet of things (IoT) has been fundamentally digitizing people’s daily life, and unprecedentedly extending the scale of personal data collection. While vast amounts of individual data have been turned into fuels with big data technologies to create and drive business, they have brought forth vital privacy concerns and real risks. At the same time, it remains a significant challenge to effectively preserve data privacy with regulatory compliance while ensuring data utility effectively. In this talk, I will discuss privacy risks in different phases of a data life cycle, from data collection, and usage to publication. Along this line, I will go through our works on differential location privacy for location-based services, differentially private model publishing for deep learning, and differentially private location trace synthesis for location data publication. Finally, I will discuss my vision of future research opportunities for data privacy in the big data era.
Biography: Lei Yu is a Research Staff Member at IBM Thomas J. Watson Research. He received his Ph.D. from the school of computer science at Georgia Insitute of Technology. His research interests include data privacy, the security and privacy of machine learning, and mobile and cloud computing. His works were published in top-tier conferences, including IEEE S&P, CCS, NDSS, and INFOCOM. At IBM, his work focuses on log analytics, system anomaly detection, and data privacy for enterprise systems. He is a recipient of the 2021 IBM Research Accomplishment award.
Crowdsourcing studies how to aggregate a large collection of votes, which makes large-scale labeled datasets and modern AI possible. In this talk, I will mainly present our recent JAIR paper which focuses on labeling cost reduction problem of crowdsourcing. First, we reformulate the crowdsourcing problem as a statistical estimation problem where votes are approximated by worker models. Then doubly robust estimation is used to address this problem. We prove that the variance of estimation can be substantially reduced, hence the labeling cost can be reduced, even if the worker model is a poor approximation. Moreover, with adaptive worker/item selection rules, labeling cost can be further reduced. I conclude by summarizing future directions of doubly robust crowdsourcing and introducing more applications of voting, for example, private learning.
Bio:
Chong Liu is a PhD candidate in Computer Science at University of California, Santa Barbara (UCSB), advised by Prof. Yu-Xiang Wang. His research interests include ensemble learning, active learning, and global optimization. He has served on program committees of several conferences, e.g., AISTATS, AAAI, ICML, and NeurIPS. He is also serving on the editorial board of JMLR.
Presenters will be:
Aidan Lane
Fuzzy Search on Encrypted Data
Advisor: Konstanin Kuzmin
Leon Montealegre
OpenCircuits EE
Advisor: Wes Turner
Daniel Stevens
Article Similarity Detection
Advisor: Sibel Adali
Caitlin Crowley
Human Assisted Fuzzing
Advisor: Bulent Yener
Omar Malik
Resource-mediated Consensus Formation on Random Networks
Advisor: Bolek Szymanski
Mack Qian
Material Point Method for Continuum Material Simulation
Advisor: Barb Cutler
Brian Hotopp
Efficiently Computing and Visualizing Semantic Shift
Advisor: Sibel Adali
Alex Sidgwick
Memory Optimization for ECS Architecture
Advisor: Jasmine Plum
Inwon Kang
Detecting Preference in Text using Dependency and Co-Reference
Advisor: Lirong Xia
Aaron Cockley
TBD
Advisor: Bulent Yener
Siwen Zhang
Study on lung lesions detection and characterization using open source toolkit
Advisor: Wes Turner
Tobias Park
Serverless Federated Learning
Advisor: Stacy Patterson
Daniel Janikowski
Simulation of Light Transport through Opal Type Materials
Advisor: Barb Cutler
Jianan Lin
PNE and POA in Hotelling’s Game with Weighted Cost Functions
Advisor: Elliot Anshelevich
Stephen Trempel
TBD
Advisor: Bulent Yener and Brian Callahan
Networks (or graphs) are a powerful tool to model complex systems such as social networks, transportation networks, and the Web. The accurate modeling of such systems enables us to improve infrastructure, reduce conflicts in social media, and make better decisions in high-stakes settings. However, as graphs are highly combinatorial structures, these optimization and learning tasks require the design of efficient algorithms.
In this talk, I will describe three research directions in the context of network data. First, I will overview several combinatorial problems for graph optimization that I have addressed using classical approaches such as approximate and randomized algorithms. The second part will focus on a different and a more recent approach to solving combinatorial problems by leveraging the power of machine learning. More specifically, I will show how combining neural architectures on graphs with reinforcement learning solves popular data ming problems such as the influence maximization problem. In the last one, I will demonstrate how to deploy these methods on problems in computational social science with applications in decision-making for patent review systems and the stock market.
Bio: Sourav Medya is a research assistant professor in the Kellogg School of Management at Northwestern University. He is also affiliated with the Northwestern Institute of Complex Systems. He has received his Ph.D. in Computer Science at the University of California, Santa Barbara. His research has been published at several venues including VLDB, NeurIPS, WebConf (WWW), AAMAS, IJCAI, WSDM, SDM, ICDM, SIGKDD Explorations, and TKDE. He has also been a PC member for WSDM, WebConf, AAAI, SDM, AAMAS, ICLR, and IJCAI.
Sourav's research is focused on the problems at the intersection of graphs and machine learning. More specifically he designs data science tools that optimize graph-based processes and improve the quality as well as scalability of traditional graph combinatorial and mining problems. He also deploys these tools to solve problems in the interdisciplinary area of computational social science especially to improve innovation.
For over 50 years now, the choice between typed and untyped code has been the source of lively debates. Languages have therefore begun to support both styles, but doing so presents new challenges at the boundaries that link typed and untyped pieces. The challenges stem from three conflicting dimensions: the expressiveness of typed-untyped mixes, the guarantees that types provide, and the cost of enforcing the guarantees. Even though dozens of languages explore points in this complex design space, they tend to focus on one dimension and neglect the others, leading to a disorganized research landscape.
In this talk, I introduce principled methods to guide the design of languages that mix typed and untyped code. The methods characterize both the behaviors of static types and the run-time cost of enforcing them, and do so in a way that centers on their implications for developers. I have applied these methods to improve existing languages and to implement a new language that bridges major gaps in the design space. My ongoing work is using insights from programmers to drive further advances.
BIO: Ben Greenman is a postdoc at Brown University. He received his Ph.D. from Northeastern University in 2020, and B.S. and M.Eng. degrees from Cornell. His work is supported by the CRA CIFellows program and has led to collaborations with Instagram and RelationalAI.
Visual perception is indispensable in numerous applications such as autonomous vehicles. Today's visual perception algorithms are often developed under a closed-world paradigm (e.g., training machine-learned models over curated datasets), which assumes the data distribution and categorical labels are fixed a priori. This assumption is unrealistic in the real open world, which contains situations that are dynamic and unpredictable. As a result, closed-world visual perception systems appear to be brittle in the open-world. For example, autonomous vehicles with such systems could fail to recognize a never-before-seen overturned truck and crash into it. We are motivated to ask how to (1) detect all the object instances in the image, and (2) recognize the unknowns. In this talk, I will present my solutions and their applications in autonomous driving and natural science. I will also introduce more research topics in the direction of Open-World Visual Perception.
Bio
Shu Kong is a Postdoctoral Fellow in the Robotics Institute at Carnegie-Mellon University, supervised by Prof. Deva Ramanan. He earned a Ph.D. in Computer Science at the University of California-Irvine, advised by Prof. Charless Fowlkes. His research interests span computer vision and machine learning, and their applications to autonomous vehicles and natural science. His current research focuses on Open-World Visual Perception. His recent paper on this topic received Best Paper / Marr Prize Honorable Mention at ICCV 2021. He regularly serves on the program committee in major conferences of computer vision and machine learning. He also serves as the lead organizer of workshops on Open-World Visual Perception at CVPR 2021 and 2022. His latest interdisciplinary research includes building a high-throughput pollen analysis system, which was featured by the National Science Foundation as that "opens a new era of fossil pollen research".
Large amounts of text are written and published daily. As a result, applications such as reading through the documents to automatically extract useful and structured information from the text have become increasingly needed for people’s efficient absorption of information. They are essential for applications such as answering user questions, information retrieval, and knowledge base population.
In this talk, I will focus on the challenges of finding and organizing information about events and introduce my research on leveraging knowledge and reasoning for document-level information extraction. In the first part, I’ll introduce methods for better modeling the knowledge from context: (1) generative learning of output structures that better model the dependency between extracted events to enable more coherent extraction of information (i.e., event A happening in the earlier part of the document is usually correlated with event B in the later part). (2) How to utilize information retrieval to enable memory-based learning with even longer context.
In the second part, to better access relevant external knowledge encoded in large models for reducing the cost of human annotations, we propose a new question-answering formulation for the extraction problem. I will conclude by outlining a research agenda for building the next generation of efficient and intelligent machine reading systems with close to human-level reasoning capabilities.
Bio:
Xinya Du is a Postdoctoral Research Associate at the University of Illinois at Urbana-Champaign working with Prof. Heng Ji. He earned a Ph.D. degree in Computer Science from Cornell University, advised by Prof. Claire Cardie. Before Cornell, he received a bachelor's degree in Computer Science from Shanghai Jiao Tong University. His research is on natural language processing, especially methods that leverage knowledge & reasoning skills for document-level information extraction. His work has been published in leading NLP conferences such as ACL, EMNLP, NAACL and has been covered by major media like New Scientist. He has received awards including the CDAC Spotlight Rising Star award and SJTU National Scholarship.
AI is now being applied widely in society, including to support decision-making in important, resource-constrained efforts in conservation and public health. Such real-world use cases introduce new challenges, like noisy, limited data and human-in-the-loop decision-making. I show that ignoring these challenges can lead to suboptimal results in AI for social impact systems. For example, previous research has modeled illegal wildlife poaching using a defender-adversary security game with signaling to better allocate scarce conservation resources. However, this work has not considered detection uncertainty arising from noisy, limited data. In contrast, my work addresses uncertainty beginning in the data analysis stage, through to the higher-level reasoning stage of defender-adversary security games with signaling. I introduce novel techniques, such as additional randomized signaling in the security game, to handle uncertainty appropriately, thereby reducing losses to the defender. I show similar reasoning is important in public health, where we would like to predict disease prevalence with few ground truth samples in order to better inform policy, such as optimizing resource allocation. In addition to modeling such real-world efforts holistically, we must also work with all stakeholders in this research, including by making our field more inclusive through efforts like my nonprofit, Try AI.
Bio: Elizabeth Bondi is a PhD candidate in Computer Science at Harvard University advised by Prof. Milind Tambe. Her research interests include multi-agent systems, remote sensing, computer vision, and deep learning, especially applied to conservation and public health. Among her awards are Best Paper Runner up at AAAI 2021, Best Application Demo Award at AAMAS 2019, Best Paper Award at SPIE DCS 2016, and an Honorable Mention for the NSF Graduate Research Fellowship Program in 2017.
Quantum can solve complex problems that classical computers will never be able to.
In recent years, significant efforts have been devoted to building quantum computers to solve
real-world problems. To ensure the correctness of quantum systems, we develop the verification
techniques and testing algorithms for quantum systems. In the first part of this talk, I will overview
my work of efficient reasoning about quantum programs by developing verification techniques and
tools that leverage the power of Birkhoff & von Neumann quantum logic. In the second part, I will
review my work on quantum state tomography, i.e., learning the classical description of quantum
hardware, which closes a 40 years long-standing gap between the upper and lower bounds
for quantum state engineering.
Bio: Nengkun Yu is an associate professor in the Centre for Quantum Software and Information,
the University of Technology Sydney. He received his B.S. and PhD degrees from the Department
of Computer Science and Technology, Tsinghua University, Beijing, China, in July of 2008 and 2013.
He won the ACM SIGPLAN distinguished paper award at OOPSLA 2020 and the ACM SIGPLAN
distinguished paper award at PLDI 2021. His research interest focuses on quantum computing.
Message passing is the essential building block in many machine learning problems such as decentralized learning and graph neural networks. In this talk, I will introduce several innovative designs of message passing schemes that address the efficiency and security issues in machine learning. Specifically, first I will present a novel decentralized algorithm with compressed message passing that enables large-scale, efficient, and scalable distributed machine learning on big data. Then I will show how to significantly improve the security and robustness of graph neural networks by exploiting the structural information in data with a novel message passing design.
Biography: Xiaorui Liu is a Ph.D. candidate in the Department of Computer Science and Engineering at Michigan State University. His advisor is Prof. Jiliang Tang. His research interests include distributed and trustworthy machine learning, with a focus on big data and graph data. He was awarded the Best Paper Honorable Mention Award at ICHI 2019, MSU Engineering Distinguished Fellowship, and Cloud Computing Fellowship. He organized and co-presented four tutorials in KDD 2021, IJCAI 2021, and ICAPS 2021, and he has published innovative works in top-tier conferences such as NeurIPS, ICML, ICLR, KDD, and AISTATS. More information can be found on his homepage https://cse.msu.edu/~xiaorui/.
Today, cryptography has transcended beyond protecting data in transit. Many advanced cryptographic techniques are gaining traction due to their significance in emerging systems like blockchains, privacy-preserving computation, and cloud computing. However, designing cryptography for these systems is challenging as they require underlying cryptographic algorithms to provide strong security and privacy guarantees and admit very efficient implementations.
In the first part of this talk, I will present my work that explores fundamental connections between cryptography and blockchains. Blockchains rely on advanced cryptography like Zero-knowledge Proofs, and despite amazing advances in building efficient cryptographic tools, scalability is a major challenge plaguing blockchain-based applications like cryptocurrencies. I will discuss my work on improving prover’s time and memory overheads in Zero-knowledge Proofs, which is currently a primary bottleneck towards building more efficient zero-knowledge proofs. Then, I will discuss my work where I use blockchains to build new and useful cryptography.
Finally, I will discuss my work on designing cryptographic commitments, digital analogs of sealed envelopes, secure against man-in-the-middle attacks. While heuristic constructions are known, my work introduces new fundamental techniques to circumvent strong barriers established for achieving provably secure protocols with minimal interaction. Specifically, I build provably secure protocols that require minimal (to no) interaction between the sender and the receiver.
Bio:
Pratik Soni is a Postdoctoral Research Fellow in the School of Computer Science at Carnegie Mellon University. He received his Ph.D. from UC Santa Barbara in 2015. His research interests span a wide range of topics across cryptography, including zero-knowledge proofs, non-malleable cryptography, secure multi-party computation, and its connections with blockchain technology. His work at FOCS 2017 was invited to SIAM Journal of Computing's special issue, and he is currently serving as a Program Committee Member at ACM CCS 2022 and ASIACRYPT 2022.
Nowadays, security researchers are increasingly using AI to automate and facilitate security analysis. Although making some meaningful progress, AI has not maximized its capability in security yet due to two challenges. First, existing ML techniques have not reached security professionals' requirements in critical properties, such as interpretability and adversary-resistancy. Second, Security data imposes many new technical challenges, which break the assumptions of existing ML Models and thus jeopardize their efficacy.
In this talk, I will describe my research efforts to address the above challenges, with a primary focus on strengthening the interpretability of deep neural networks and deep reinforcement learning policies. Regarding deep neural networks, I will describe an explanation method for deep learning-based security applications and demonstrate how security analysts could benefit from this method to establish trust in blackbox models and conduct efficient finetuning. As for DRL policies, I will introduce a novel approach to draw critical states/actions of a DRL agent and show how to utilize the above explanations to scrutinize policy weaknesses, remediate policy errors, and even defend against adversarial attacks. Finally, I will conclude by highlighting my future plan towards strengthening the critical properties of advanced ML techniques and maximizing their capability in cyber defenses.
Bio:
Wenbo Guo is a Ph.D. Candidate at Penn State, advised by Professor Xinyu Xing. His research interests are machine learning and cybersecurity. His work includes strengthening the fundamental properties of machine learning models and designing customized machine learning models to handle security-unique challenges. He is a recipient of the IBM Ph.D. Fellowship (2020-2022), Facebook/Baidu Ph.D. Fellowship Finalist (2020), and ACM CCS Outstanding Paper Award (2018). His research has been featured by multiple mainstream media and has appeared in a diverse set of top-tier venues in security, machine learning, and data mining. Going beyond academic research, he also actively participates in many world-class cybersecurity competitions and has won the 2018 DEFCON/GeekPwn AI challenge finalist award.
Our visual world is naturally open, containing visual elements that are dynamic, vast, and unpredictable. However, existing computer vision models are often developed inside a closed-world paradigm, for example, recognizing objects or human actions from a fixed set of categories. If these models are exposed to the realistic complexity of the open world, they will be brittle and fail to generalize. For example, an autonomous driving vehicle may not be able to avoid an accident if it sees a turnover truck, because it has never seen this novelty in its training data. In order to enable visual understanding in open world, vision models need to be aware of the unknown novelties, interpret their decision processes to human users, and adapt themselves to the novelties. In this talk, I will describe our recent work on open-set recognition and interpretable visual precognition. Together, these approaches make strides towards assurance autonomous AI in open world.
Bio: Dr. Yu Kong is now an Assistant Professor directing the ACTION lab in the Golisano College of Computing and Information Sciences at Rochester Institute of Technology. He received B.Eng. degree in automation from Anhui University in 2006, and PhD degree in computer science from Beijing Institute of Technology, China, in 2012. He was a postdoctoral research associate in the Department of Computer Science and Engineering, State University of New York, Buffalo in 2012, and then in the Department of Electrical and Computer Engineering, Northeastern University, Boston, MA. Dr. Kong's research in Computer Vision and Machine Learning is supported by National Science Foundation, Army Research Office, and Office of Naval Research, etc. His work has been publishing on top-tier conferences and transactions in the AI community such as CVPR, ECCV, ICCV, T-PAMI, IJCV, etc. He is an Associate Editor for Springer Journal of Multimedia Systems, and also serves as reviewers and PC members for prestige journals and conferences, including T-PAMI, T-IP, T-NNLS, T-CSVT, CVPR, ICLR, AAAI, and IJCAI, etc. More information can be found on his webpage at https://people.rit.edu/yukics/.
Abstract:
In this talk I will discuss recent progress towards using human input to enable safe and robust autonomous systems. Whereas much work on robust machine learning and control seeks to be resilient to or remove the need for human input, my research seeks to directly and efficiently incorporate human input into the study of robust AI systems. One problem that arises when robots and other AI systems learn from human input is that there is often a large amount of uncertainty over the human’s true intent and the corresponding desired robot behavior. To address this problem, I will discuss prior and ongoing research along three main topics: (1) how to enable AI systems to efficiently and accurately maintain uncertainty over human intent, (2) how to generate risk-averse behaviors that are robust to this uncertainty, and (3) how robots and other AI systems can efficiently query for additional human input to actively reduce uncertainty and improve their performance. My talk will conclude with a discussion of my long-term vision for safe and robust autonomy, including learning from multi-modal human input, interpretable and verifiable robustness, and developing techniques for human-in-the-loop robust machine learning that generalize beyond reward function uncertainty.
Bio:
Daniel Brown is a postdoctoral scholar at UC Berkeley, advised by Anca Dragan and Ken Goldberg. His research focuses on safe and robust autonomous systems, with an emphasis on robot learning under uncertainty, human-AI interaction, and value alignment of AI systems. He evaluates his research across a range of applications, including autonomous driving, service robotics, and dexterous manipulation. Daniel received his Ph.D. in computer science from the University of Texas at Austin, where he worked with Scott Niekum on safe and efficient inverse reinforcement learning. Prior to starting his PhD, Daniel was a research scientist at the Air Force Research Lab's Information Directorate where he studied bio-inspired swarms and multi-agent systems. Daniel’s research has been nominated for two best-paper awards and he was selected in 2021 as a Robotics: Science and Systems Pioneer.