Colloquia & Seminars

Many real-world problems require the creation of robust AI models that include both learning and planning for an agent (or a team of agents) in interaction with adversaries in a multi-agent environment. In such a complex setting, it is important to predict strategic behavior of the adversaries, as well as to anticipate potential adversarial manipulations that could deteriorate the learning outcomes and the decision quality of our agents. In this talk, I will discuss the challenges of modeling adversaries’ decision making and the security of machine learning in data-driven multi-agent competitive environments. I will present our algorithms to address these challenges that explore techniques in reinforcement learning, game theory, and optimization research. In addition, I will introduce some of the real-world applications of our algorithms in the domains of wildlife protection and public health.

Bio: Thanh Nguyen is an Assistant Professor in the Computer Science department at the University of Oregon (UO). Prior to UO, she was a postdoc at the University of Michigan and earned her PhD in Computer Science from the University of Southern California. Thanh’s work in the field of Artificial Intelligence is motivated by real-world societal problems, particularly in the areas of Public Safety and Security, Conservation, and Public Health. She brings together techniques from multi-agent systems, reinforcement learning, and game theory to solve problems in those areas, with the focus on studying adversary behavioral learning and deception in competitive multi-agent environments. Thanh’s work has been recognized by multiple awards, including the IAAI-16 Deployed Application Award, and the AAMAS-16 Runner-up of the Best Innovative Application Paper Award. Her works in wildlife protection and public health were evaluated and/or deployed in multiple countries around the world.

With the end of Moore's Law and the rise of compute- and data-intensive deep-learning (DL) applications, the focus on arduous new processor design has shifted towards a more effective and agile approach -- Intelligent Software to maximize the performance gains of DL hardware like GPUs.

In this talk, I will first highlight the importance of software innovation to bridge the gap between the increasingly diverse DL applications and the existing powerful DL hardware platforms. The second part of my talk will recap my research work on DL system software innovation, focusing on bridging the 1) Precision Mismatch between DL applications and high-performance GPU units like Tensor Cores (PPoPP '21 and SC '21), and 2) Computing Pattern Mismatch between the sparse and irregular DL applications such as Graph Neural Networks and the dense and regular tailored GPU computing paradigm (OSDI '21 and OSDI '23). Finally, I will conclude this talk with my vision and future work for building efficient, scalable, and secure DL systems.

Bio: Yuke Wang is a final-year Doctor of Philosophy (Ph.D.) candidate in the Department of computer science at the University of California, Santa Barbara (UCSB). He got his Bachelor of Engineering (B.E.) in software engineering from the University of Electronic Science and Technology of China (UESTC) in 2018. At UCSB, Yuke is working with Prof.Yufei Ding (Now at UC at San Diego, CSE). Yuke's research interests include Systems & Compiler for Deep Learning and GPU-based High-performance Computing. His projects cover graph neural network (GNN) optimization and its acceleration on GPUs. Yuke’s research has resulted in 20+ publications (with 10 first-authored papers) in top-tier conferences, including OSDI, ASPLOS, ISCA, USENIX ATC, PPoPP, and SC. Yuke’s research outcome has been adopted for further research in industries (e.g., NVIDIA, OctoML, and Alibaba) and academia (e.g., University of Washington and Pacific Northwest National Laboratory). Yuke is also the recipient of the NVIDIA Graduate Fellowship 2022 (Top-10 out of global applicants) and has industry experience at Microsoft Research, NVIDIA Research, and Alibaba. The ultimate goal of Yuke’s research is to facilitate efficient, scalable, and secure deep learning in the future.  https://www.wang-yuke.com/

Quantum algorithms claim to outperform classical algorithms by harnessing computational and communication properties unique to quantum systems, such as superposition and entanglement, e.g., Shor's factoring algorithm and Grover's algorithm. Do quantum algorithms live up to the claims?

Google's quantum supremacy announcement in 2019 has received broad questions from academia and industry due to the debatable estimate of 10,000 years' running time for the classical simulation task on the Summit supercomputer. Has “quantum supremacy" already come? Or will it come in one decade later?   We take a reinforcement learning approach for the classical simulation of quantum circuits and demonstrate its great potential by reporting an estimated simulation time of less than 4 days, a speedup of 5.40x over the state-of-the-art method.  Specifically, we utilize a tensor network approach and employ a deep reinforcement learning algorithm.   

Bio: Xiao-Yang (Yanglet) Liu joined RPI's CS department as a lecturer in September 2023. He holds Ph.D. and M.S. degrees in the Department of Electrical Engineering at Columbia University in 2023 and 2018, respectively. His research interests include quantum computing and tensor networks, deep reinforcement learning, and model-openness framework (licenses) in AI. Xiao-Yang has authored chapters to two graduate textbooks:  tensors for data processing, and reinforcement learning for cyber-physical systems. His papers got over 4400 citations, three of which are ESI-highly cited papers. He received NeurIPS scholar award 2022/2023 and ICAIF-JPM award 2022/2023. He is an academic member of Linux Foundation, LF AI & Data, FinOS, and CRAFT, collaborating on open-source projects FinGPT and FinRL.  As a (senior) PC member, he serves leading AI conferences such as NeurIPS, ICML, ICLR, AAAI, and ACM ICAIF. Xiao-Yang has chaired sessions at IJCAI 2019 and has been a leader organizer of multiple workshops and academic competitions, including NeurIPS 2020/2021 First/Second Workshop on Quantum Tensor Networks in Machine Learning (QTNML), ACM ICAIF FinRL competition 2023, and IJCAI 2020 Workshop on Tensor Networks Representations in Machine Learning.

Optimization problems with random objective functions are central in computer science, probability, and modern data science. Despite their ubiquity, finding efficient algorithms for solving these problems remains a major challenge. Interestingly, many random optimization problems share a common feature, dubbed as a statistical-computational gap: while the optimal value can be pinpointed non-constructively (through, e.g., probabilistic/information-theoretic tools), all known polynomial-time algorithms find strictly sub-optimal solutions. That is, an optimal solution can only be found through brute force search which is computationally expensive. 

In this talk, I will discuss an emerging theoretical framework for understanding the fundamental computational limits of random optimization problems, based on the Overlap Gap Property (OGP). This is an intricate geometrical property that achieves sharp algorithmic lower bounds against the best known polynomial-time algorithms for a wide range of random optimization problems. I will focus on two models to demonstrate the power of the OGP framework: (a) the symmetric binary perceptron, a random constraint satisfaction problem and a simple neural network classifying/storing random patterns, widely studied in computer science, probability, and statistics communities, and (b) the random number partitioning problem as well as its planted counterpart, a classical worst-case NP-hard problem whose average-case variant is closely related to the design of randomized controlled trials. In addition to yielding sharp algorithmic lower bounds, our techniques also give rise to new toolkits for the study of statistical-computational tradeoffs in other models, including the online setting.

Bio. Eren C. Kizildag is a Distinguished Postdoctoral Fellow at Columbia University, Department of Statistics. He received his PhD in Electrical Engineering and Computer Science from MIT in 2022, supervised by David Gamarnik. His research interests are at the intersection of computer science with probability, statistics, and data science. He is particularly interested in understanding statistical-computational tradeoffs in random computational problems and large-scale random models, as well as in mathematical foundations of machine learning and data science. 

In this talk, I will provide an overview of my contributions to quantum computing, specifically focusing on hardware software co-design for quantum computing by diving into the pulse level. Transitioning to a pulse-level workflow paradigm can reduce the circuit duration of quantum programs, enabling the execution of deeper quantum circuits on quantum machines with the same decoherence time. This shift in focus has been substantiated by my research to enhance solutions in practical areas such as quantum machine learning, quantum finance, and quantum chemistry. In the first part of the talk, I will introduce QPulse, a work that introduces a set of designs for parameterized pulses and evaluates them based on specific metrics, including their expressive capacity, entanglement capabilities, and effective parameter dimensions. Then, I will present NAPA, a cutting-edge native-pulse ansatz generator framework specifically tailored for variational quantum algorithms. By progressively searching the pulse-level circuit architecture, we build a pulse ansatz that demonstrates a significant advantage over the gate-level quantum circuit on benchmarking tasks. Finally, I will discuss my ongoing work and future research towards building efficient design automation tools and scalable hybrid classical-quantum algorithms to implement quantum technologies for practical real-world applications.

Bio: Zhiding Liang is currently a postgraduate student studying for his Ph.D. degree in the Department of Computer Science and Engineering at the University of Notre Dame under the supervision of Prof. Yiyu Shi. His current research interests include hardware-software co-design for quantum computing and quantum machine learning. The results of his research have been published in prestigious conferences and journals, including DAC, ICCAD, QCE, TCAD, and TVCG. He has been selected as a DAC Young Fellow in both 2021 and 2022. He has also been nominated as the recipient of the Edison Innovation Fellowship by the IDEA Center at the University of Notre Dame. He is devoted to quantum education and outreach; he is the co-founder of the Quantum Computer System (QuCS) Lecture Series, an impactful public online lecture series in the quantum computing community. He also led the organization of the first ACM/IEEE Quantum Computing for Drug Discovery Challenge at ICCAD, a top-tier computer science conference. He is one of the major contributors to the TorchQuantum library, which has been adopted by IBM Qiskit Ecosystem and PyTorch Ecosystem with 1.1K+ stars on GitHub. He received the B.S. in Electrical Engineering from the University of Wisconsin-Madison.

In this presentation, I'll talk about Cryptis, a tool we have been developing for verifying systems that feature cryptographic components, such as a key-value store server that connects to clients using some encryption protocol.  Cryptis is a separation logic embedded in the Coq proof assistant, which can be used to describe the implementation of protocols and systems that use these protocols.

Cryptis can check proofs of correctness involving arguments in the symbolic model of cryptography, thus guaranteeing that protocols can deliver strong security guarantees.

Bio:  Arthur Azevedo de Amorim joined the Department of Computer Science at RIT in 2023 as an assistant professor.  His research interests revolve around the use of programming-language and software-verification techniques to improve the security and reliability of software.

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.

 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.

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

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/.

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.

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.

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

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.

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).