April 2025 - Ai - Engineering Mailing Lists

No AI Seminar This Week - Check Out AI Week Events!
by Abou Mrad, Christian 28 Apr '25

28 Apr '25

Dear all, There will be no AI seminar this week. Instead, join AI Week 2025 (April 28 - May 2) for a week of AI exploration with hands-on sessions, community-led talks, and industry insights from partners like NVIDIA and Microsoft. Learn and connect across OSU! Find event details and registration here: https://arcs.oregonstate.edu/ai-week<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Farcs.oreg…> Future AI seminar info: https://engineering.oregonstate.edu/EECS/research/AI-seminars<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri…> Best, Christian Abou-Mrad Graduate Student Oregon State University

1 0

AI Seminar: Accelerating HPC Applications Using Machine Learning-based Surrogates
by Abou Mrad, Christian 24 Apr '25

24 Apr '25

Dear all, Our next AI seminar is scheduled for Friday, April 25th. Talk details: AI Seminar: Accelerating HPC Applications Using Machine Learning-based Surrogates Speaker: Wenqian Dong, Assistant Professor, Oregon State University Time: 2:00 PM Location: LINC 302 and Zoom Zoom link: link<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Foregonsta…> Talk Abstract: Large-scale scientific simulations drive scientific and engineering discovery across many domains, but face performance problems. These simulations typically involve complex physics computations but are difficult to scale efficiently on high-performance hardware. Machine Learning (ML)-based surrogates are revolutionizing the natural sciences. By employing reverse-engineering and automatic learning methodologies, ML surrogates can solve complex, unstructured problems with less computing power and execution time than needed by traditional direct and first-principle methods, with the advantages of lower implementation costs, stronger generalization capability, and lower computation overheads. In this talk, I will introduce the application of ML-based surrogates in two user-inspired High-Performance Computing (HPC) applications, and a generalized framework that automates the application process of ML-based surrogates in the HPC domain. Auto-HPCnet democratizes the usage of ML-based surrogates and is the first end-to-end framework that makes past proposals for the ML-based surrogate model practical and disciplined. Auto-HPCnet introduces a workflow to address unique challenges when applying the surrogates, such as feature acquisition and meeting the application-specific constraint on the quality of the final computation outcome. Also, the two real-world HPC applications include the Eulerian fluid simulation and the AC-OPF power grid simulation. In the Eulerian fluid simulation, we generate multiple surrogate candidates before the simulation and introduce a runtime scheduler that dynamically switches the ML surrogates to make the best efforts to reach the user’s requirement on simulation quality. We show that our method achieves 1.46x and 590x speedup, compared with a state-of-the-art ML model and the numerical fluid simulation respectively, while providing better simulation quality than the state-of-the-art model. In the AC-OPF power grid simulation, we generate multitask-learning (MTL) surrogates to predict the initial values of variables critical to the convergence of the power grid problem. The MTL models allow information sharing when predicting multiple dependent variables, while including customized layers to predict individual variables. The MTL model incorporates physics-informed learning to improve model accuracy and interpretability. These techniques bring 2.60× speedup on average (up to 3.28×) computed over 10,000 large-scale power grid problems, without losing solution optimality. Speaker Bio: Wenqian Dong is an assistant professor in the School of Electrical Engineering and Computer Science at Oregon State University. She earned her Ph.D. in EECS at the University of California, Merced, in Spring 2022. Recently, she is selected for the IEEE-CS Technical Community on High Performance Computing (TCHPC) Early Career Researchers Award for Excellence in High Performance Computing. Her research focuses on three main areas. She has contributed significantly to scientific machine learning, particularly in using machine learning to speed up HPC applications. Her work is showcased in conferences like SC’19 and SC’20. Wenqian has excelled in automatic machine learning, concentrating on creating machine learning models for HPC applications. Her papers in VLDB’21, HPDC’23, and ASPLOS’22 highlight her notable contributions. She’s skilled in optimizing system performance, aiming to enhance HPC applications’ quality and efficiency through system optimization. Her work presented at conferences like ICS’21, Eurosys’21, ICPP’18, and Parallel Computing’23 illustrates her dedication to this field. Her work has generated real impacts in the HPC community. For example, her work on power grid simulation using ML leads to 3.28 times performance improvement and highlighted Newswise as a DOE science innovation. Furthermore, Wenqian is committed to enrich the HPC community. Her commitment is apparent in her various roles as an organizer for ICPP’24, the MLBench’23 workshop, and as a member of the Technical Program Committee (TPC) for the SC’24, HPDC’24, CCGrid’24, IEEE Cloud 2023, IEEE Cluster 2023, AI4Science 2022 workshop, and the GPGPU 2023 workshop. For future AI seminars, please visit: https://engineering.oregonstate.edu/EECS/research/AI-seminars<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri…>. Best, Christian Abou-Mrad Graduate Student Oregon State University

1 1

AI Seminar: Efficiently Adapting Large Language Models for Safety, Multilinguality, and Long Context
by Abou Mrad, Christian 17 Apr '25

17 Apr '25

Dear all, Our next AI seminar is scheduled for Friday, April 18th. Talk details: AI Seminar: Efficiently Adapting Large Language Models for Safety, Multilinguality, and Long Context Speaker: Thien Huu Nguyen, Associate Professor, University of Oregon Time: 2:00 PM Location: LINC 302 and Zoom Zoom link: link<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Foregonsta…> Talk Abstract: Large Language Models (LLMs) have become essential to artificial intelligence, showcasing impressive fluency, reasoning, and adaptability. However, as these models scale to hundreds of billions of parameters, adapting them to new learning settings or desired behaviors can be costly, often requiring substantial data and compute for fine-tuning. In this talk, I will present our three recent works focused on developing efficient methods for adapting LLMs to new learning scenarios and expected behaviors. I will highlight three key challenges: ensuring safe responses, enhancing multilingual capabilities for low-resource languages, and improving long-context efficiency. For each, I will introduce our approach to efficient LLM adaptation, featuring activation editing, encoder-decoder combination, and linear models. I will conclude with our vision for future work. Speaker Bio: Thien Huu Nguyen is an Associate Professor in the Department of Computer Science at the University of Oregon. He obtained his Ph.D. in Natural Language Processing (NLP) from New York University, advised by Professors Ralph Grishman and Kyunghyun Cho, and completed a postdoctoral fellowship with Professor Yoshua Bengio at the University of Montreal. Thien's research focuses on Information Extraction, Text Analysis, Large Language Models (LLMs), Multilingual Learning, Representation Learning, and Deep Learning. He pioneered some of the initial deep learning models for Entity Recognition, Relation Extraction, Event Extraction, and Event Coreference Resolution in NLP. His recent work explores efficient LLMs and multilingual/multi-domain NLP models that achieve robust performance across diverse languages and domains. At the University of Oregon, Thien directs the NSF IUCRC Center for Big Learning (CBL). He has received the NSF CAREER Award for his works on Multilingual Learning for Information Extraction and was recognized as an AI 2000 Most Influential Scholar (Honorable Mention) in NLP. His toolkits for multilingual NLP and acronym identification/disambiguation won Outstanding and Best Demo Paper Awards at EACL. Thien was also an IBM Ph.D. Fellow, and his research has been supported by NSF, IARPA, Army Research Office, Adobe Research, and IBM Research. For future AI seminars, please visit: https://engineering.oregonstate.edu/EECS/research/AI-seminars<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri…>. Best, Christian Abou-Mrad Graduate Student Oregon State University

1 1

AI Seminar: From Sparse To Dense, and Back To Sparse Again?
by Abou Mrad, Christian 10 Apr '25

10 Apr '25

Dear all, Our next AI seminar is scheduled for Friday, April 11th. Talk details: AI Seminar: From Sparse To Dense, and Back To Sparse Again? Speaker: Fuxin Li, Associate Professor, Oregon State University Time: 2:00 PM Location: LINC 302 and Zoom Zoom link: link<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Foregonsta…> Talk Abstract: Computer vision architectures used to be built on a sparse sample of points in the 80s and 90s. In the 2000s, dense models started to become popular for visual recognition as heuristically defined sparse models do not cover all the important parts of an image. However, with deep learning and end-to-end training approaches, this does not have to continue and sparse models may still have significant advantages in saving unnecessary computation as well as being more flexible. In this talk, I will talk about the deep point cloud convolutional backbones that we have developed in the past few years, including results on point cloud segmentation tasks, as well as recent applications on interaction modeling among objects, point cloud completion and world models for robot manipulation tasks. Point cloud approaches can also work well as 2D image recognition backbones. I will introduce our work AutoFocusFormer that uses point cloud backbones and decoders to work on 2D image recognition, with a novel adaptive downsampling module that enables the end-to-end learning of adaptive downsampling for dense prediction tasks such as segmentation. This is very helpful for detecting tiny objects faraway in the scene which would have been decimated by conventional grid downsampling approaches. Speaker Bio: Fuxin Li<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri…> is currently an associate professor in the School of Electrical Engineering and Computer Science at Oregon State University. He has held research positions at Apple Inc., University of Bonn and Georgia Institute of Technology. He had obtained a Ph.D. degree in the Institute of Automation, Chinese Academy of Sciences in 2009. He has won an NSF CAREER award and an Amazon Research Award, among other accolades. He is a program chair of CVPR 2025. He has published more than 90 papers in computer vision, machine learning, as well as applications of machine learning and computer vision. His main research interests are point cloud deep networks, human understanding of deep learning, video object segmentation, multi-target tracking and uncertainty estimation in deep learning. For future AI seminars, please visit: https://engineering.oregonstate.edu/EECS/research/AI-seminars<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri…>. Best, Christian Abou-Mrad Graduate Student Oregon State University

1 1

AI Seminar: New Frontiers in Deep Learning for Large-Scale Spatiotemporal Data
by Abou Mrad, Christian 03 Apr '25

03 Apr '25

Dear all, Our next AI seminar is scheduled for Friday, April 4th. Talk details: AI Seminar: New Frontiers in Deep Learning for Large-Scale Spatiotemporal Data Speaker: Rose Yu, Associate Professor, Computer Science and Engineering, UC San Diego Time: 2:00 PM Location: BEXL 102 and Zoom Zoom link: link<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Foregonsta…> Talk Abstract: Scientific applications such as climate science and public health generate massive data with rich spatiotemporal information. Existing deep learning models are still limited in handling complex spatiotemporal information. We will explain how to design deep learning models to learn from large-scale spatiotemporal data, especially for dealing with physical constraints, high-dimensional uncertainty, and complex spatiotemporal dependencies. We will showcase the application of these models to problems such as turbulence prediction, climate emulation, and generating what-if scenarios for epidemic prevention and control. Speaker Bio: Dr. Rose Yu is an associate professor at the University of California San Diego, Department of Computer Science and Engineering. She earned her Ph.D. in Computer Sciences at USC in 2017. She was subsequently a Postdoctoral Fellow at Caltech. Her research focuses on advancing machine learning techniques for large-scale spatiotemporal data analysis, with applications to sustainability, health, and physical sciences. A particular emphasis of her research is on physics-guided AI which aims to integrate first principles with data-driven models. She is a recipient of the Presidential Early Career Award (PECASE) -- the highest honor given by the White House to early career scientists, Army ECASE Award, NSF CAREER Award, Hellman Fellow, Faculty Research Award from JP Morgan, Facebook, Google, Amazon, and Adobe, Several Best Paper Awards, Best Dissertation Award at USC’. She was named as MIT Technology Review Innovators Under 35 in AI. For future AI seminars, please visit: https://engineering.oregonstate.edu/EECS/research/AI-seminars<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri…>. Best, Christian Abou-Mrad Graduate Student Oregon State University

1 1