Bio: Dr. Huan Liu is a Regents Professor and Ira A. Fulton Professor of Computer Science and Engineering at Arizona State University. He is the recipient of the ACM SIGKDD 2022 Innovation Award. His research interests are in data mining, machine learning, feature selection, social computing, social media mining, and artificial intelligence. He is a co-author of a text, Social Media Mining: An Introduction,Cambridge University Press. He is Editor in Chief of ACM TIST, and Field Chief Editor of Frontiers in Big Data and its Specialty Chief Editor of Data Mining and Management. He is a Fellow of ACM, AAAI, AAAS, and IEEE.

Speaking Topics

• Preliminary Discussion on LLM-based Feature Selection

Speaker's Notes

Large Language Models (LLMs) are powerful in some data-intensive applications. Can feature selection leverage LLMs to identify informative features within a dataset, complementing its dependence on traditional search strategies or statistical heuristics? This talk offers our preliminary study of LLM-based feature selection approaches and their robustness and efficiency, including direct prompting-based and search-based methods, and introduces an alternative, a tool-use-based approach, followed by empirical results and discussion.

Bio:Dr. Suzanne M. Shontz is the Associate Dean for Graduate and Online Education in the School of Engineering (SoE) and a Full Professor in the Department of Electrical Engineering and Computer Science at the University of Kansas (KU). She also holds courtesy appointments in the Department of Mechanical Engineering and the Bioengineering Program. In addition, she is the Director of the Mathematical Methods and Interdisciplinary Computing Center (MMICC) within the Information Sciences Institute at KU. Her research interests are in parallel scientific computing, with an emphasis on the development of unstructured mesh generation, numerical optimization, and machine learning methods, and their applications to image processing, computational medicine, materials, and other areas. Prof. Shontz is a Fellow of the International Meshing Roundtable and is the recipient of an NSF CAREER Award, NSF Presidential Early Career Award for Scientists and Engineers, and the James Corones Award in Leadership, Community Building, and Communication.

Speaking Topics

• Unsupervised Deep Learning-Based Deformable Image Registration for Cardiac Motion Estimation from Cine Cardiac MRI Images

Speaker's Notes

Estimated cardiac motion is used to assess the biomechanics of the cardiac chambers and is important for disease diagnosis and treatment planning. Cardiac motion can be estimated from a 4D cine cardiac MRI dataset by registering pairs of consecutive 3D images of the deforming heart. In this talk, I will present our hybrid convolutional neural network and Vision Transformer architecture for deformable image registration of 3D cine cardiac MRI images for cardiac motion estimation followed by experimental results and discussion.

Bio: Alix Schmidt is a Senior Research Scientist in Dow Chemical's Core R&D Information Research team in Midland, Michigan. During her 16 years in industry, Alix has held technical roles in polymer process research and development, high throughput research and combinatorial design, and chemical manufacturing analytics. She is currently a lead materials informatics researcher at Dow, with three granted patents and several pending patents on the subject of materials representation. Additionally, she leads the R&D Machine Learning Operations (MLOps) Strategy and the LLM Strategy for the successful deployment of traditional and generative AI into Dow's research, product design, and technical customer service organizations. Alix focuses on building a community around industrial materials informatics through guest lectures at several universities, chairing symposia at the AIChE and ACS conferences, and is co-editor of the 2024 book The Digital Transformation of Product Formulation: Concepts, Challenges, and Applications for Accelerated Innovation.

Speaking Topics

• The Myth of Clean Data: Real‑World Data Challenges in Materials Informatics

Speaker's Notes

In contrast to industries with many digital-native companies, the chemicals and materials industry is still dominated by large, established players. Far from digitally native, many older companies suffer from highly variable and siloed data infrastructure resulting from decades of mergers and acquisitions and changes in technology. There is a prevailing assumption that if enough money is spent to integrate systems and achieve a unified data fabric the data foundation for materials AI will be solved. In this talk, I will challenge this assumption and highlight some characteristics inherent to R&D and materials design data that pose difficulties to machine learning even in the case of "perfectly clean" data.